page_alloc.c 208 KB

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385138613871388138913901391139213931394139513961397139813991400140114021403140414051406140714081409141014111412141314141415141614171418141914201421142214231424142514261427142814291430143114321433143414351436143714381439144014411442144314441445144614471448144914501451145214531454145514561457145814591460146114621463146414651466146714681469147014711472147314741475147614771478147914801481148214831484148514861487148814891490149114921493149414951496149714981499150015011502150315041505150615071508150915101511151215131514151515161517151815191520152115221523152415251526152715281529153015311532153315341535153615371538153915401541154215431544154515461547154815491550155115521553155415551556155715581559156015611562156315641565156615671568156915701571157215731574157515761577157815791580158115821583158415851586158715881589159015911592159315941595159615971598159916001601160216031604160516061607160816091610161116121613161416151616161716181619162016211622162316241625162616271628162916301631163216331634163516361637163816391640164116421643164416451646164716481649165016511652165316541655165616571658165916601661166216631664166516661667166816691670167116721673167416751676167716781679168016811682168316841685168616871688168916901691169216931694169516961697169816991700170117021703170417051706170717081709171017111712171317141715171617171718171917201721172217231724172517261727172817291730173117321733173417351736173717381739174017411742174317441745174617471748174917501751175217531754175517561757175817591760176117621763176417651766176717681769177017711772177317741775177617771778177917801781178217831784178517861787178817891790179117921793179417951796179717981799180018011802180318041805180618071808180918101811181218131814181518161817181818191820182118221823182418251826182718281829183018311832183318341835183618371838183918401841184218431844184518461847184818491850185118521853185418551856185718581859186018611862186318641865186618671868186918701871187218731874187518761877187818791880188118821883188418851886188718881889189018911892189318941895189618971898189919001901190219031904190519061907190819091910191119121913191419151916191719181919192019211922192319241925192619271928192919301931193219331934193519361937193819391940194119421943194419451946194719481949195019511952195319541955195619571958195919601961196219631964196519661967196819691970197119721973197419751976197719781979198019811982198319841985198619871988198919901991199219931994199519961997199819992000200120022003200420052006200720082009201020112012201320142015201620172018201920202021202220232024202520262027202820292030203120322033203420352036203720382039204020412042204320442045204620472048204920502051205220532054205520562057205820592060206120622063206420652066206720682069207020712072207320742075207620772078207920802081208220832084208520862087208820892090209120922093209420952096209720982099210021012102210321042105210621072108210921102111211221132114211521162117211821192120212121222123212421252126212721282129213021312132213321342135213621372138213921402141214221432144214521462147214821492150215121522153215421552156215721582159216021612162216321642165216621672168216921702171217221732174217521762177217821792180218121822183218421852186218721882189219021912192219321942195219621972198219922002201220222032204220522062207220822092210221122122213221422152216221722182219222022212222222322242225222622272228222922302231223222332234223522362237223822392240224122422243224422452246224722482249225022512252225322542255225622572258225922602261226222632264226522662267226822692270227122722273227422752276227722782279228022812282228322842285228622872288228922902291229222932294229522962297229822992300230123022303230423052306230723082309231023112312231323142315231623172318231923202321232223232324232523262327232823292330233123322333233423352336233723382339234023412342234323442345234623472348234923502351235223532354235523562357235823592360236123622363236423652366236723682369237023712372237323742375237623772378237923802381238223832384238523862387238823892390239123922393239423952396239723982399240024012402240324042405240624072408240924102411241224132414241524162417241824192420242124222423242424252426242724282429243024312432243324342435243624372438243924402441244224432444244524462447244824492450245124522453245424552456245724582459246024612462246324642465246624672468246924702471247224732474247524762477247824792480248124822483248424852486248724882489249024912492249324942495249624972498249925002501250225032504250525062507250825092510251125122513251425152516251725182519252025212522252325242525252625272528252925302531253225332534253525362537253825392540254125422543254425452546254725482549255025512552255325542555255625572558255925602561256225632564256525662567256825692570257125722573257425752576257725782579258025812582258325842585258625872588258925902591259225932594259525962597259825992600260126022603260426052606260726082609261026112612261326142615261626172618261926202621262226232624262526262627262826292630263126322633263426352636263726382639264026412642264326442645264626472648264926502651265226532654265526562657265826592660266126622663266426652666266726682669267026712672267326742675267626772678267926802681268226832684268526862687268826892690269126922693269426952696269726982699270027012702270327042705270627072708270927102711271227132714271527162717271827192720272127222723272427252726272727282729273027312732273327342735273627372738273927402741274227432744274527462747274827492750275127522753275427552756275727582759276027612762276327642765276627672768276927702771277227732774277527762777277827792780278127822783278427852786278727882789279027912792279327942795279627972798279928002801280228032804280528062807280828092810281128122813281428152816281728182819282028212822282328242825282628272828282928302831283228332834283528362837283828392840284128422843284428452846284728482849285028512852285328542855285628572858285928602861286228632864286528662867286828692870287128722873287428752876287728782879288028812882288328842885288628872888288928902891289228932894289528962897289828992900290129022903290429052906290729082909291029112912291329142915291629172918291929202921292229232924292529262927292829292930293129322933293429352936293729382939294029412942294329442945294629472948294929502951295229532954295529562957295829592960296129622963296429652966296729682969297029712972297329742975297629772978297929802981298229832984298529862987298829892990299129922993299429952996299729982999300030013002300330043005300630073008300930103011301230133014301530163017301830193020302130223023302430253026302730283029303030313032303330343035303630373038303930403041304230433044304530463047304830493050305130523053305430553056305730583059306030613062306330643065306630673068306930703071307230733074307530763077307830793080308130823083308430853086308730883089309030913092309330943095309630973098309931003101310231033104310531063107310831093110311131123113311431153116311731183119312031213122312331243125312631273128312931303131313231333134313531363137313831393140314131423143314431453146314731483149315031513152315331543155315631573158315931603161316231633164316531663167316831693170317131723173317431753176317731783179318031813182318331843185318631873188318931903191319231933194319531963197319831993200320132023203320432053206320732083209321032113212321332143215321632173218321932203221322232233224322532263227322832293230323132323233323432353236323732383239324032413242324332443245324632473248324932503251325232533254325532563257325832593260326132623263326432653266326732683269327032713272327332743275327632773278327932803281328232833284328532863287328832893290329132923293329432953296329732983299330033013302330333043305330633073308330933103311331233133314331533163317331833193320332133223323332433253326332733283329333033313332333333343335333633373338333933403341334233433344334533463347334833493350335133523353335433553356335733583359336033613362336333643365336633673368336933703371337233733374337533763377337833793380338133823383338433853386338733883389339033913392339333943395339633973398339934003401340234033404340534063407340834093410341134123413341434153416341734183419342034213422342334243425342634273428342934303431343234333434343534363437343834393440344134423443344434453446344734483449345034513452345334543455345634573458345934603461346234633464346534663467346834693470347134723473347434753476347734783479348034813482348334843485348634873488348934903491349234933494349534963497349834993500350135023503350435053506350735083509351035113512351335143515351635173518351935203521352235233524352535263527352835293530353135323533353435353536353735383539354035413542354335443545354635473548354935503551355235533554355535563557355835593560356135623563356435653566356735683569357035713572357335743575357635773578357935803581358235833584358535863587358835893590359135923593359435953596359735983599360036013602360336043605360636073608360936103611361236133614361536163617361836193620362136223623362436253626362736283629363036313632363336343635363636373638363936403641364236433644364536463647364836493650365136523653365436553656365736583659366036613662366336643665366636673668366936703671367236733674367536763677367836793680368136823683368436853686368736883689369036913692369336943695369636973698369937003701370237033704370537063707370837093710371137123713371437153716371737183719372037213722372337243725372637273728372937303731373237333734373537363737373837393740374137423743374437453746374737483749375037513752375337543755375637573758375937603761376237633764376537663767376837693770377137723773377437753776377737783779378037813782378337843785378637873788378937903791379237933794379537963797379837993800380138023803380438053806380738083809381038113812381338143815381638173818381938203821382238233824382538263827382838293830383138323833383438353836383738383839384038413842384338443845384638473848384938503851385238533854385538563857385838593860386138623863386438653866386738683869387038713872387338743875387638773878387938803881388238833884388538863887388838893890389138923893389438953896389738983899390039013902390339043905390639073908390939103911391239133914391539163917391839193920392139223923392439253926392739283929393039313932393339343935393639373938393939403941394239433944394539463947394839493950395139523953395439553956395739583959396039613962396339643965396639673968396939703971397239733974397539763977397839793980398139823983398439853986398739883989399039913992399339943995399639973998399940004001400240034004400540064007400840094010401140124013401440154016401740184019402040214022402340244025402640274028402940304031403240334034403540364037403840394040404140424043404440454046404740484049405040514052405340544055405640574058405940604061406240634064406540664067406840694070407140724073407440754076407740784079408040814082408340844085408640874088408940904091409240934094409540964097409840994100410141024103410441054106410741084109411041114112411341144115411641174118411941204121412241234124412541264127412841294130413141324133413441354136413741384139414041414142414341444145414641474148414941504151415241534154415541564157415841594160416141624163416441654166416741684169417041714172417341744175417641774178417941804181418241834184418541864187418841894190419141924193419441954196419741984199420042014202420342044205420642074208420942104211421242134214421542164217421842194220422142224223422442254226422742284229423042314232423342344235423642374238423942404241424242434244424542464247424842494250425142524253425442554256425742584259426042614262426342644265426642674268426942704271427242734274427542764277427842794280428142824283428442854286428742884289429042914292429342944295429642974298429943004301430243034304430543064307430843094310431143124313431443154316431743184319432043214322432343244325432643274328432943304331433243334334433543364337433843394340434143424343434443454346434743484349435043514352435343544355435643574358435943604361436243634364436543664367436843694370437143724373437443754376437743784379438043814382438343844385438643874388438943904391439243934394439543964397439843994400440144024403440444054406440744084409441044114412441344144415441644174418441944204421442244234424442544264427442844294430443144324433443444354436443744384439444044414442444344444445444644474448444944504451445244534454445544564457445844594460446144624463446444654466446744684469447044714472447344744475447644774478447944804481448244834484448544864487448844894490449144924493449444954496449744984499450045014502450345044505450645074508450945104511451245134514451545164517451845194520452145224523452445254526452745284529453045314532453345344535453645374538453945404541454245434544454545464547454845494550455145524553455445554556455745584559456045614562456345644565456645674568456945704571457245734574457545764577457845794580458145824583458445854586458745884589459045914592459345944595459645974598459946004601460246034604460546064607460846094610461146124613461446154616461746184619462046214622462346244625462646274628462946304631463246334634463546364637463846394640464146424643464446454646464746484649465046514652465346544655465646574658465946604661466246634664466546664667466846694670467146724673467446754676467746784679468046814682468346844685468646874688468946904691469246934694469546964697469846994700470147024703470447054706470747084709471047114712471347144715471647174718471947204721472247234724472547264727472847294730473147324733473447354736473747384739474047414742474347444745474647474748474947504751475247534754475547564757475847594760476147624763476447654766476747684769477047714772477347744775477647774778477947804781478247834784478547864787478847894790479147924793479447954796479747984799480048014802480348044805480648074808480948104811481248134814481548164817481848194820482148224823482448254826482748284829483048314832483348344835483648374838483948404841484248434844484548464847484848494850485148524853485448554856485748584859486048614862486348644865486648674868486948704871487248734874487548764877487848794880488148824883488448854886488748884889489048914892489348944895489648974898489949004901490249034904490549064907490849094910491149124913491449154916491749184919492049214922492349244925492649274928492949304931493249334934493549364937493849394940494149424943494449454946494749484949495049514952495349544955495649574958495949604961496249634964496549664967496849694970497149724973497449754976497749784979498049814982498349844985498649874988498949904991499249934994499549964997499849995000500150025003500450055006500750085009501050115012501350145015501650175018501950205021502250235024502550265027502850295030503150325033503450355036503750385039504050415042504350445045504650475048504950505051505250535054505550565057505850595060506150625063506450655066506750685069507050715072507350745075507650775078507950805081508250835084508550865087508850895090509150925093509450955096509750985099510051015102510351045105510651075108510951105111511251135114511551165117511851195120512151225123512451255126512751285129513051315132513351345135513651375138513951405141514251435144514551465147514851495150515151525153515451555156515751585159516051615162516351645165516651675168516951705171517251735174517551765177517851795180518151825183518451855186518751885189519051915192519351945195519651975198519952005201520252035204520552065207520852095210521152125213521452155216521752185219522052215222522352245225522652275228522952305231523252335234523552365237523852395240524152425243524452455246524752485249525052515252525352545255525652575258525952605261526252635264526552665267526852695270527152725273527452755276527752785279528052815282528352845285528652875288528952905291529252935294529552965297529852995300530153025303530453055306530753085309531053115312531353145315531653175318531953205321532253235324532553265327532853295330533153325333533453355336533753385339534053415342534353445345534653475348534953505351535253535354535553565357535853595360536153625363536453655366536753685369537053715372537353745375537653775378537953805381538253835384538553865387538853895390539153925393539453955396539753985399540054015402540354045405540654075408540954105411541254135414541554165417541854195420542154225423542454255426542754285429543054315432543354345435543654375438543954405441544254435444544554465447544854495450545154525453545454555456545754585459546054615462546354645465546654675468546954705471547254735474547554765477547854795480548154825483548454855486548754885489549054915492549354945495549654975498549955005501550255035504550555065507550855095510551155125513551455155516551755185519552055215522552355245525552655275528552955305531553255335534553555365537553855395540554155425543554455455546554755485549555055515552555355545555555655575558555955605561556255635564556555665567556855695570557155725573557455755576557755785579558055815582558355845585558655875588558955905591559255935594559555965597559855995600560156025603560456055606560756085609561056115612561356145615561656175618561956205621562256235624562556265627562856295630563156325633563456355636563756385639564056415642564356445645564656475648564956505651565256535654565556565657565856595660566156625663566456655666566756685669567056715672567356745675567656775678567956805681568256835684568556865687568856895690569156925693569456955696569756985699570057015702570357045705570657075708570957105711571257135714571557165717571857195720572157225723572457255726572757285729573057315732573357345735573657375738573957405741574257435744574557465747574857495750575157525753575457555756575757585759576057615762576357645765576657675768576957705771577257735774577557765777577857795780578157825783578457855786578757885789579057915792579357945795579657975798579958005801580258035804580558065807580858095810581158125813581458155816581758185819582058215822582358245825582658275828582958305831583258335834583558365837583858395840584158425843584458455846584758485849585058515852585358545855585658575858585958605861586258635864586558665867586858695870587158725873587458755876587758785879588058815882588358845885588658875888588958905891589258935894589558965897589858995900590159025903590459055906590759085909591059115912591359145915591659175918591959205921592259235924592559265927592859295930593159325933593459355936593759385939594059415942594359445945594659475948594959505951595259535954595559565957595859595960596159625963596459655966596759685969597059715972597359745975597659775978597959805981598259835984598559865987598859895990599159925993599459955996599759985999600060016002600360046005600660076008600960106011601260136014601560166017601860196020602160226023602460256026602760286029603060316032603360346035603660376038603960406041604260436044604560466047604860496050605160526053605460556056605760586059606060616062606360646065606660676068606960706071607260736074607560766077607860796080608160826083608460856086608760886089609060916092609360946095609660976098609961006101610261036104610561066107610861096110611161126113611461156116611761186119612061216122612361246125612661276128612961306131613261336134613561366137613861396140614161426143614461456146614761486149615061516152615361546155615661576158615961606161616261636164616561666167616861696170617161726173617461756176617761786179618061816182618361846185618661876188618961906191619261936194619561966197619861996200620162026203620462056206620762086209621062116212621362146215621662176218621962206221622262236224622562266227622862296230623162326233623462356236623762386239624062416242624362446245624662476248624962506251625262536254625562566257625862596260626162626263626462656266626762686269627062716272627362746275627662776278627962806281628262836284628562866287628862896290629162926293629462956296629762986299630063016302630363046305630663076308630963106311631263136314631563166317631863196320632163226323632463256326632763286329633063316332633363346335633663376338633963406341634263436344634563466347634863496350635163526353635463556356635763586359636063616362636363646365636663676368636963706371637263736374637563766377637863796380638163826383638463856386638763886389639063916392639363946395639663976398639964006401640264036404640564066407640864096410641164126413641464156416641764186419642064216422642364246425642664276428642964306431643264336434643564366437643864396440644164426443644464456446644764486449645064516452645364546455645664576458645964606461646264636464646564666467646864696470647164726473647464756476647764786479648064816482648364846485648664876488648964906491649264936494649564966497649864996500650165026503650465056506650765086509651065116512651365146515651665176518651965206521652265236524652565266527652865296530653165326533653465356536653765386539654065416542654365446545654665476548654965506551655265536554655565566557655865596560656165626563656465656566656765686569657065716572657365746575657665776578657965806581658265836584658565866587658865896590659165926593659465956596659765986599660066016602660366046605660666076608660966106611661266136614661566166617661866196620662166226623662466256626662766286629663066316632663366346635663666376638663966406641664266436644664566466647664866496650665166526653665466556656665766586659666066616662666366646665666666676668666966706671667266736674667566766677667866796680668166826683668466856686668766886689669066916692669366946695669666976698669967006701670267036704670567066707670867096710671167126713671467156716671767186719672067216722672367246725672667276728672967306731673267336734673567366737673867396740674167426743674467456746674767486749675067516752675367546755675667576758675967606761676267636764676567666767676867696770677167726773677467756776677767786779678067816782678367846785678667876788678967906791679267936794679567966797679867996800680168026803680468056806680768086809681068116812681368146815681668176818681968206821682268236824682568266827682868296830683168326833683468356836683768386839684068416842684368446845684668476848684968506851685268536854685568566857685868596860686168626863686468656866686768686869687068716872687368746875687668776878687968806881688268836884688568866887688868896890689168926893689468956896689768986899690069016902690369046905690669076908690969106911691269136914691569166917691869196920692169226923692469256926692769286929693069316932693369346935693669376938693969406941694269436944694569466947694869496950695169526953695469556956695769586959696069616962696369646965696669676968696969706971697269736974697569766977697869796980698169826983698469856986698769886989699069916992699369946995699669976998699970007001700270037004700570067007700870097010701170127013701470157016701770187019702070217022702370247025702670277028702970307031703270337034703570367037703870397040704170427043704470457046704770487049705070517052705370547055705670577058705970607061706270637064706570667067706870697070707170727073707470757076707770787079708070817082708370847085708670877088708970907091709270937094709570967097709870997100710171027103710471057106710771087109711071117112711371147115711671177118711971207121712271237124712571267127712871297130713171327133713471357136713771387139714071417142714371447145714671477148714971507151715271537154715571567157715871597160716171627163716471657166716771687169717071717172717371747175717671777178717971807181718271837184718571867187718871897190719171927193719471957196719771987199720072017202720372047205720672077208720972107211721272137214721572167217721872197220722172227223722472257226722772287229723072317232723372347235723672377238723972407241724272437244724572467247724872497250725172527253725472557256725772587259726072617262726372647265726672677268726972707271727272737274727572767277727872797280728172827283728472857286728772887289729072917292729372947295729672977298729973007301730273037304730573067307730873097310731173127313731473157316731773187319732073217322732373247325732673277328732973307331733273337334733573367337733873397340734173427343734473457346734773487349735073517352735373547355735673577358735973607361736273637364736573667367736873697370737173727373737473757376737773787379738073817382738373847385738673877388738973907391739273937394739573967397739873997400740174027403740474057406740774087409741074117412741374147415741674177418741974207421742274237424742574267427742874297430743174327433743474357436743774387439744074417442744374447445744674477448744974507451745274537454745574567457745874597460746174627463746474657466746774687469747074717472747374747475747674777478747974807481748274837484748574867487748874897490749174927493749474957496749774987499750075017502750375047505750675077508750975107511751275137514751575167517751875197520752175227523752475257526752775287529753075317532753375347535753675377538753975407541754275437544754575467547754875497550755175527553755475557556755775587559756075617562756375647565756675677568756975707571757275737574757575767577757875797580758175827583758475857586758775887589759075917592759375947595759675977598759976007601760276037604760576067607760876097610761176127613761476157616761776187619762076217622762376247625762676277628762976307631763276337634763576367637763876397640764176427643764476457646764776487649765076517652765376547655765676577658765976607661
  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/kasan.h>
  28. #include <linux/module.h>
  29. #include <linux/suspend.h>
  30. #include <linux/pagevec.h>
  31. #include <linux/blkdev.h>
  32. #include <linux/slab.h>
  33. #include <linux/ratelimit.h>
  34. #include <linux/oom.h>
  35. #include <linux/notifier.h>
  36. #include <linux/topology.h>
  37. #include <linux/sysctl.h>
  38. #include <linux/cpu.h>
  39. #include <linux/cpuset.h>
  40. #include <linux/memory_hotplug.h>
  41. #include <linux/nodemask.h>
  42. #include <linux/vmalloc.h>
  43. #include <linux/vmstat.h>
  44. #include <linux/mempolicy.h>
  45. #include <linux/memremap.h>
  46. #include <linux/stop_machine.h>
  47. #include <linux/sort.h>
  48. #include <linux/pfn.h>
  49. #include <linux/backing-dev.h>
  50. #include <linux/fault-inject.h>
  51. #include <linux/page-isolation.h>
  52. #include <linux/page_ext.h>
  53. #include <linux/debugobjects.h>
  54. #include <linux/kmemleak.h>
  55. #include <linux/compaction.h>
  56. #include <trace/events/kmem.h>
  57. #include <trace/events/oom.h>
  58. #include <linux/prefetch.h>
  59. #include <linux/mm_inline.h>
  60. #include <linux/migrate.h>
  61. #include <linux/hugetlb.h>
  62. #include <linux/sched/rt.h>
  63. #include <linux/sched/mm.h>
  64. #include <linux/page_owner.h>
  65. #include <linux/kthread.h>
  66. #include <linux/memcontrol.h>
  67. #include <asm/sections.h>
  68. #include <asm/tlbflush.h>
  69. #include <asm/div64.h>
  70. #include "internal.h"
  71. /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
  72. static DEFINE_MUTEX(pcp_batch_high_lock);
  73. #define MIN_PERCPU_PAGELIST_FRACTION (8)
  74. #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
  75. DEFINE_PER_CPU(int, numa_node);
  76. EXPORT_PER_CPU_SYMBOL(numa_node);
  77. #endif
  78. #ifdef CONFIG_HAVE_MEMORYLESS_NODES
  79. /*
  80. * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
  81. * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
  82. * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
  83. * defined in <linux/topology.h>.
  84. */
  85. DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */
  86. EXPORT_PER_CPU_SYMBOL(_numa_mem_);
  87. int _node_numa_mem_[MAX_NUMNODES];
  88. #endif
  89. /* work_structs for global per-cpu drains */
  90. DEFINE_MUTEX(pcpu_drain_mutex);
  91. DEFINE_PER_CPU(struct work_struct, pcpu_drain);
  92. #ifdef CONFIG_GCC_PLUGIN_LATENT_ENTROPY
  93. volatile unsigned long latent_entropy __latent_entropy;
  94. EXPORT_SYMBOL(latent_entropy);
  95. #endif
  96. /*
  97. * Array of node states.
  98. */
  99. nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
  100. [N_POSSIBLE] = NODE_MASK_ALL,
  101. [N_ONLINE] = { { [0] = 1UL } },
  102. #ifndef CONFIG_NUMA
  103. [N_NORMAL_MEMORY] = { { [0] = 1UL } },
  104. #ifdef CONFIG_HIGHMEM
  105. [N_HIGH_MEMORY] = { { [0] = 1UL } },
  106. #endif
  107. #ifdef CONFIG_MOVABLE_NODE
  108. [N_MEMORY] = { { [0] = 1UL } },
  109. #endif
  110. [N_CPU] = { { [0] = 1UL } },
  111. #endif /* NUMA */
  112. };
  113. EXPORT_SYMBOL(node_states);
  114. /* Protect totalram_pages and zone->managed_pages */
  115. static DEFINE_SPINLOCK(managed_page_count_lock);
  116. unsigned long totalram_pages __read_mostly;
  117. unsigned long totalreserve_pages __read_mostly;
  118. unsigned long totalcma_pages __read_mostly;
  119. int percpu_pagelist_fraction;
  120. gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
  121. /*
  122. * A cached value of the page's pageblock's migratetype, used when the page is
  123. * put on a pcplist. Used to avoid the pageblock migratetype lookup when
  124. * freeing from pcplists in most cases, at the cost of possibly becoming stale.
  125. * Also the migratetype set in the page does not necessarily match the pcplist
  126. * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any
  127. * other index - this ensures that it will be put on the correct CMA freelist.
  128. */
  129. static inline int get_pcppage_migratetype(struct page *page)
  130. {
  131. return page->index;
  132. }
  133. static inline void set_pcppage_migratetype(struct page *page, int migratetype)
  134. {
  135. page->index = migratetype;
  136. }
  137. #ifdef CONFIG_PM_SLEEP
  138. /*
  139. * The following functions are used by the suspend/hibernate code to temporarily
  140. * change gfp_allowed_mask in order to avoid using I/O during memory allocations
  141. * while devices are suspended. To avoid races with the suspend/hibernate code,
  142. * they should always be called with pm_mutex held (gfp_allowed_mask also should
  143. * only be modified with pm_mutex held, unless the suspend/hibernate code is
  144. * guaranteed not to run in parallel with that modification).
  145. */
  146. static gfp_t saved_gfp_mask;
  147. void pm_restore_gfp_mask(void)
  148. {
  149. WARN_ON(!mutex_is_locked(&pm_mutex));
  150. if (saved_gfp_mask) {
  151. gfp_allowed_mask = saved_gfp_mask;
  152. saved_gfp_mask = 0;
  153. }
  154. }
  155. void pm_restrict_gfp_mask(void)
  156. {
  157. WARN_ON(!mutex_is_locked(&pm_mutex));
  158. WARN_ON(saved_gfp_mask);
  159. saved_gfp_mask = gfp_allowed_mask;
  160. gfp_allowed_mask &= ~(__GFP_IO | __GFP_FS);
  161. }
  162. bool pm_suspended_storage(void)
  163. {
  164. if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS))
  165. return false;
  166. return true;
  167. }
  168. #endif /* CONFIG_PM_SLEEP */
  169. #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
  170. unsigned int pageblock_order __read_mostly;
  171. #endif
  172. static void __free_pages_ok(struct page *page, unsigned int order);
  173. /*
  174. * results with 256, 32 in the lowmem_reserve sysctl:
  175. * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
  176. * 1G machine -> (16M dma, 784M normal, 224M high)
  177. * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
  178. * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
  179. * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA
  180. *
  181. * TBD: should special case ZONE_DMA32 machines here - in those we normally
  182. * don't need any ZONE_NORMAL reservation
  183. */
  184. int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
  185. #ifdef CONFIG_ZONE_DMA
  186. 256,
  187. #endif
  188. #ifdef CONFIG_ZONE_DMA32
  189. 256,
  190. #endif
  191. #ifdef CONFIG_HIGHMEM
  192. 32,
  193. #endif
  194. 32,
  195. };
  196. EXPORT_SYMBOL(totalram_pages);
  197. static char * const zone_names[MAX_NR_ZONES] = {
  198. #ifdef CONFIG_ZONE_DMA
  199. "DMA",
  200. #endif
  201. #ifdef CONFIG_ZONE_DMA32
  202. "DMA32",
  203. #endif
  204. "Normal",
  205. #ifdef CONFIG_HIGHMEM
  206. "HighMem",
  207. #endif
  208. "Movable",
  209. #ifdef CONFIG_ZONE_DEVICE
  210. "Device",
  211. #endif
  212. };
  213. char * const migratetype_names[MIGRATE_TYPES] = {
  214. "Unmovable",
  215. "Movable",
  216. "Reclaimable",
  217. "HighAtomic",
  218. #ifdef CONFIG_CMA
  219. "CMA",
  220. #endif
  221. #ifdef CONFIG_MEMORY_ISOLATION
  222. "Isolate",
  223. #endif
  224. };
  225. compound_page_dtor * const compound_page_dtors[] = {
  226. NULL,
  227. free_compound_page,
  228. #ifdef CONFIG_HUGETLB_PAGE
  229. free_huge_page,
  230. #endif
  231. #ifdef CONFIG_TRANSPARENT_HUGEPAGE
  232. free_transhuge_page,
  233. #endif
  234. };
  235. int min_free_kbytes = 1024;
  236. int user_min_free_kbytes = -1;
  237. int watermark_scale_factor = 10;
  238. static unsigned long __meminitdata nr_kernel_pages;
  239. static unsigned long __meminitdata nr_all_pages;
  240. static unsigned long __meminitdata dma_reserve;
  241. #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
  242. static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
  243. static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
  244. static unsigned long __initdata required_kernelcore;
  245. static unsigned long __initdata required_movablecore;
  246. static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
  247. static bool mirrored_kernelcore;
  248. /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
  249. int movable_zone;
  250. EXPORT_SYMBOL(movable_zone);
  251. #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
  252. #if MAX_NUMNODES > 1
  253. int nr_node_ids __read_mostly = MAX_NUMNODES;
  254. int nr_online_nodes __read_mostly = 1;
  255. EXPORT_SYMBOL(nr_node_ids);
  256. EXPORT_SYMBOL(nr_online_nodes);
  257. #endif
  258. int page_group_by_mobility_disabled __read_mostly;
  259. #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
  260. static inline void reset_deferred_meminit(pg_data_t *pgdat)
  261. {
  262. pgdat->first_deferred_pfn = ULONG_MAX;
  263. }
  264. /* Returns true if the struct page for the pfn is uninitialised */
  265. static inline bool __meminit early_page_uninitialised(unsigned long pfn)
  266. {
  267. int nid = early_pfn_to_nid(pfn);
  268. if (node_online(nid) && pfn >= NODE_DATA(nid)->first_deferred_pfn)
  269. return true;
  270. return false;
  271. }
  272. /*
  273. * Returns false when the remaining initialisation should be deferred until
  274. * later in the boot cycle when it can be parallelised.
  275. */
  276. static inline bool update_defer_init(pg_data_t *pgdat,
  277. unsigned long pfn, unsigned long zone_end,
  278. unsigned long *nr_initialised)
  279. {
  280. unsigned long max_initialise;
  281. /* Always populate low zones for address-contrained allocations */
  282. if (zone_end < pgdat_end_pfn(pgdat))
  283. return true;
  284. /*
  285. * Initialise at least 2G of a node but also take into account that
  286. * two large system hashes that can take up 1GB for 0.25TB/node.
  287. */
  288. max_initialise = max(2UL << (30 - PAGE_SHIFT),
  289. (pgdat->node_spanned_pages >> 8));
  290. (*nr_initialised)++;
  291. if ((*nr_initialised > max_initialise) &&
  292. (pfn & (PAGES_PER_SECTION - 1)) == 0) {
  293. pgdat->first_deferred_pfn = pfn;
  294. return false;
  295. }
  296. return true;
  297. }
  298. #else
  299. static inline void reset_deferred_meminit(pg_data_t *pgdat)
  300. {
  301. }
  302. static inline bool early_page_uninitialised(unsigned long pfn)
  303. {
  304. return false;
  305. }
  306. static inline bool update_defer_init(pg_data_t *pgdat,
  307. unsigned long pfn, unsigned long zone_end,
  308. unsigned long *nr_initialised)
  309. {
  310. return true;
  311. }
  312. #endif
  313. /* Return a pointer to the bitmap storing bits affecting a block of pages */
  314. static inline unsigned long *get_pageblock_bitmap(struct page *page,
  315. unsigned long pfn)
  316. {
  317. #ifdef CONFIG_SPARSEMEM
  318. return __pfn_to_section(pfn)->pageblock_flags;
  319. #else
  320. return page_zone(page)->pageblock_flags;
  321. #endif /* CONFIG_SPARSEMEM */
  322. }
  323. static inline int pfn_to_bitidx(struct page *page, unsigned long pfn)
  324. {
  325. #ifdef CONFIG_SPARSEMEM
  326. pfn &= (PAGES_PER_SECTION-1);
  327. return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
  328. #else
  329. pfn = pfn - round_down(page_zone(page)->zone_start_pfn, pageblock_nr_pages);
  330. return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
  331. #endif /* CONFIG_SPARSEMEM */
  332. }
  333. /**
  334. * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
  335. * @page: The page within the block of interest
  336. * @pfn: The target page frame number
  337. * @end_bitidx: The last bit of interest to retrieve
  338. * @mask: mask of bits that the caller is interested in
  339. *
  340. * Return: pageblock_bits flags
  341. */
  342. static __always_inline unsigned long __get_pfnblock_flags_mask(struct page *page,
  343. unsigned long pfn,
  344. unsigned long end_bitidx,
  345. unsigned long mask)
  346. {
  347. unsigned long *bitmap;
  348. unsigned long bitidx, word_bitidx;
  349. unsigned long word;
  350. bitmap = get_pageblock_bitmap(page, pfn);
  351. bitidx = pfn_to_bitidx(page, pfn);
  352. word_bitidx = bitidx / BITS_PER_LONG;
  353. bitidx &= (BITS_PER_LONG-1);
  354. word = bitmap[word_bitidx];
  355. bitidx += end_bitidx;
  356. return (word >> (BITS_PER_LONG - bitidx - 1)) & mask;
  357. }
  358. unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn,
  359. unsigned long end_bitidx,
  360. unsigned long mask)
  361. {
  362. return __get_pfnblock_flags_mask(page, pfn, end_bitidx, mask);
  363. }
  364. static __always_inline int get_pfnblock_migratetype(struct page *page, unsigned long pfn)
  365. {
  366. return __get_pfnblock_flags_mask(page, pfn, PB_migrate_end, MIGRATETYPE_MASK);
  367. }
  368. /**
  369. * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
  370. * @page: The page within the block of interest
  371. * @flags: The flags to set
  372. * @pfn: The target page frame number
  373. * @end_bitidx: The last bit of interest
  374. * @mask: mask of bits that the caller is interested in
  375. */
  376. void set_pfnblock_flags_mask(struct page *page, unsigned long flags,
  377. unsigned long pfn,
  378. unsigned long end_bitidx,
  379. unsigned long mask)
  380. {
  381. unsigned long *bitmap;
  382. unsigned long bitidx, word_bitidx;
  383. unsigned long old_word, word;
  384. BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4);
  385. bitmap = get_pageblock_bitmap(page, pfn);
  386. bitidx = pfn_to_bitidx(page, pfn);
  387. word_bitidx = bitidx / BITS_PER_LONG;
  388. bitidx &= (BITS_PER_LONG-1);
  389. VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page);
  390. bitidx += end_bitidx;
  391. mask <<= (BITS_PER_LONG - bitidx - 1);
  392. flags <<= (BITS_PER_LONG - bitidx - 1);
  393. word = READ_ONCE(bitmap[word_bitidx]);
  394. for (;;) {
  395. old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags);
  396. if (word == old_word)
  397. break;
  398. word = old_word;
  399. }
  400. }
  401. void set_pageblock_migratetype(struct page *page, int migratetype)
  402. {
  403. if (unlikely(page_group_by_mobility_disabled &&
  404. migratetype < MIGRATE_PCPTYPES))
  405. migratetype = MIGRATE_UNMOVABLE;
  406. set_pageblock_flags_group(page, (unsigned long)migratetype,
  407. PB_migrate, PB_migrate_end);
  408. }
  409. #ifdef CONFIG_DEBUG_VM
  410. static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
  411. {
  412. int ret = 0;
  413. unsigned seq;
  414. unsigned long pfn = page_to_pfn(page);
  415. unsigned long sp, start_pfn;
  416. do {
  417. seq = zone_span_seqbegin(zone);
  418. start_pfn = zone->zone_start_pfn;
  419. sp = zone->spanned_pages;
  420. if (!zone_spans_pfn(zone, pfn))
  421. ret = 1;
  422. } while (zone_span_seqretry(zone, seq));
  423. if (ret)
  424. pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
  425. pfn, zone_to_nid(zone), zone->name,
  426. start_pfn, start_pfn + sp);
  427. return ret;
  428. }
  429. static int page_is_consistent(struct zone *zone, struct page *page)
  430. {
  431. if (!pfn_valid_within(page_to_pfn(page)))
  432. return 0;
  433. if (zone != page_zone(page))
  434. return 0;
  435. return 1;
  436. }
  437. /*
  438. * Temporary debugging check for pages not lying within a given zone.
  439. */
  440. static int bad_range(struct zone *zone, struct page *page)
  441. {
  442. if (page_outside_zone_boundaries(zone, page))
  443. return 1;
  444. if (!page_is_consistent(zone, page))
  445. return 1;
  446. return 0;
  447. }
  448. #else
  449. static inline int bad_range(struct zone *zone, struct page *page)
  450. {
  451. return 0;
  452. }
  453. #endif
  454. static void bad_page(struct page *page, const char *reason,
  455. unsigned long bad_flags)
  456. {
  457. static unsigned long resume;
  458. static unsigned long nr_shown;
  459. static unsigned long nr_unshown;
  460. /*
  461. * Allow a burst of 60 reports, then keep quiet for that minute;
  462. * or allow a steady drip of one report per second.
  463. */
  464. if (nr_shown == 60) {
  465. if (time_before(jiffies, resume)) {
  466. nr_unshown++;
  467. goto out;
  468. }
  469. if (nr_unshown) {
  470. pr_alert(
  471. "BUG: Bad page state: %lu messages suppressed\n",
  472. nr_unshown);
  473. nr_unshown = 0;
  474. }
  475. nr_shown = 0;
  476. }
  477. if (nr_shown++ == 0)
  478. resume = jiffies + 60 * HZ;
  479. pr_alert("BUG: Bad page state in process %s pfn:%05lx\n",
  480. current->comm, page_to_pfn(page));
  481. __dump_page(page, reason);
  482. bad_flags &= page->flags;
  483. if (bad_flags)
  484. pr_alert("bad because of flags: %#lx(%pGp)\n",
  485. bad_flags, &bad_flags);
  486. dump_page_owner(page);
  487. print_modules();
  488. dump_stack();
  489. out:
  490. /* Leave bad fields for debug, except PageBuddy could make trouble */
  491. page_mapcount_reset(page); /* remove PageBuddy */
  492. add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
  493. }
  494. /*
  495. * Higher-order pages are called "compound pages". They are structured thusly:
  496. *
  497. * The first PAGE_SIZE page is called the "head page" and have PG_head set.
  498. *
  499. * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded
  500. * in bit 0 of page->compound_head. The rest of bits is pointer to head page.
  501. *
  502. * The first tail page's ->compound_dtor holds the offset in array of compound
  503. * page destructors. See compound_page_dtors.
  504. *
  505. * The first tail page's ->compound_order holds the order of allocation.
  506. * This usage means that zero-order pages may not be compound.
  507. */
  508. void free_compound_page(struct page *page)
  509. {
  510. __free_pages_ok(page, compound_order(page));
  511. }
  512. void prep_compound_page(struct page *page, unsigned int order)
  513. {
  514. int i;
  515. int nr_pages = 1 << order;
  516. set_compound_page_dtor(page, COMPOUND_PAGE_DTOR);
  517. set_compound_order(page, order);
  518. __SetPageHead(page);
  519. for (i = 1; i < nr_pages; i++) {
  520. struct page *p = page + i;
  521. set_page_count(p, 0);
  522. p->mapping = TAIL_MAPPING;
  523. set_compound_head(p, page);
  524. }
  525. atomic_set(compound_mapcount_ptr(page), -1);
  526. }
  527. #ifdef CONFIG_DEBUG_PAGEALLOC
  528. unsigned int _debug_guardpage_minorder;
  529. bool _debug_pagealloc_enabled __read_mostly
  530. = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT);
  531. EXPORT_SYMBOL(_debug_pagealloc_enabled);
  532. bool _debug_guardpage_enabled __read_mostly;
  533. static int __init early_debug_pagealloc(char *buf)
  534. {
  535. if (!buf)
  536. return -EINVAL;
  537. return kstrtobool(buf, &_debug_pagealloc_enabled);
  538. }
  539. early_param("debug_pagealloc", early_debug_pagealloc);
  540. static bool need_debug_guardpage(void)
  541. {
  542. /* If we don't use debug_pagealloc, we don't need guard page */
  543. if (!debug_pagealloc_enabled())
  544. return false;
  545. if (!debug_guardpage_minorder())
  546. return false;
  547. return true;
  548. }
  549. static void init_debug_guardpage(void)
  550. {
  551. if (!debug_pagealloc_enabled())
  552. return;
  553. if (!debug_guardpage_minorder())
  554. return;
  555. _debug_guardpage_enabled = true;
  556. }
  557. struct page_ext_operations debug_guardpage_ops = {
  558. .need = need_debug_guardpage,
  559. .init = init_debug_guardpage,
  560. };
  561. static int __init debug_guardpage_minorder_setup(char *buf)
  562. {
  563. unsigned long res;
  564. if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) {
  565. pr_err("Bad debug_guardpage_minorder value\n");
  566. return 0;
  567. }
  568. _debug_guardpage_minorder = res;
  569. pr_info("Setting debug_guardpage_minorder to %lu\n", res);
  570. return 0;
  571. }
  572. early_param("debug_guardpage_minorder", debug_guardpage_minorder_setup);
  573. static inline bool set_page_guard(struct zone *zone, struct page *page,
  574. unsigned int order, int migratetype)
  575. {
  576. struct page_ext *page_ext;
  577. if (!debug_guardpage_enabled())
  578. return false;
  579. if (order >= debug_guardpage_minorder())
  580. return false;
  581. page_ext = lookup_page_ext(page);
  582. if (unlikely(!page_ext))
  583. return false;
  584. __set_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
  585. INIT_LIST_HEAD(&page->lru);
  586. set_page_private(page, order);
  587. /* Guard pages are not available for any usage */
  588. __mod_zone_freepage_state(zone, -(1 << order), migratetype);
  589. return true;
  590. }
  591. static inline void clear_page_guard(struct zone *zone, struct page *page,
  592. unsigned int order, int migratetype)
  593. {
  594. struct page_ext *page_ext;
  595. if (!debug_guardpage_enabled())
  596. return;
  597. page_ext = lookup_page_ext(page);
  598. if (unlikely(!page_ext))
  599. return;
  600. __clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
  601. set_page_private(page, 0);
  602. if (!is_migrate_isolate(migratetype))
  603. __mod_zone_freepage_state(zone, (1 << order), migratetype);
  604. }
  605. #else
  606. struct page_ext_operations debug_guardpage_ops;
  607. static inline bool set_page_guard(struct zone *zone, struct page *page,
  608. unsigned int order, int migratetype) { return false; }
  609. static inline void clear_page_guard(struct zone *zone, struct page *page,
  610. unsigned int order, int migratetype) {}
  611. #endif
  612. static inline void set_page_order(struct page *page, unsigned int order)
  613. {
  614. set_page_private(page, order);
  615. __SetPageBuddy(page);
  616. }
  617. static inline void rmv_page_order(struct page *page)
  618. {
  619. __ClearPageBuddy(page);
  620. set_page_private(page, 0);
  621. }
  622. /*
  623. * This function checks whether a page is free && is the buddy
  624. * we can do coalesce a page and its buddy if
  625. * (a) the buddy is not in a hole (check before calling!) &&
  626. * (b) the buddy is in the buddy system &&
  627. * (c) a page and its buddy have the same order &&
  628. * (d) a page and its buddy are in the same zone.
  629. *
  630. * For recording whether a page is in the buddy system, we set ->_mapcount
  631. * PAGE_BUDDY_MAPCOUNT_VALUE.
  632. * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
  633. * serialized by zone->lock.
  634. *
  635. * For recording page's order, we use page_private(page).
  636. */
  637. static inline int page_is_buddy(struct page *page, struct page *buddy,
  638. unsigned int order)
  639. {
  640. if (page_is_guard(buddy) && page_order(buddy) == order) {
  641. if (page_zone_id(page) != page_zone_id(buddy))
  642. return 0;
  643. VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
  644. return 1;
  645. }
  646. if (PageBuddy(buddy) && page_order(buddy) == order) {
  647. /*
  648. * zone check is done late to avoid uselessly
  649. * calculating zone/node ids for pages that could
  650. * never merge.
  651. */
  652. if (page_zone_id(page) != page_zone_id(buddy))
  653. return 0;
  654. VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
  655. return 1;
  656. }
  657. return 0;
  658. }
  659. /*
  660. * Freeing function for a buddy system allocator.
  661. *
  662. * The concept of a buddy system is to maintain direct-mapped table
  663. * (containing bit values) for memory blocks of various "orders".
  664. * The bottom level table contains the map for the smallest allocatable
  665. * units of memory (here, pages), and each level above it describes
  666. * pairs of units from the levels below, hence, "buddies".
  667. * At a high level, all that happens here is marking the table entry
  668. * at the bottom level available, and propagating the changes upward
  669. * as necessary, plus some accounting needed to play nicely with other
  670. * parts of the VM system.
  671. * At each level, we keep a list of pages, which are heads of continuous
  672. * free pages of length of (1 << order) and marked with _mapcount
  673. * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
  674. * field.
  675. * So when we are allocating or freeing one, we can derive the state of the
  676. * other. That is, if we allocate a small block, and both were
  677. * free, the remainder of the region must be split into blocks.
  678. * If a block is freed, and its buddy is also free, then this
  679. * triggers coalescing into a block of larger size.
  680. *
  681. * -- nyc
  682. */
  683. static inline void __free_one_page(struct page *page,
  684. unsigned long pfn,
  685. struct zone *zone, unsigned int order,
  686. int migratetype)
  687. {
  688. unsigned long combined_pfn;
  689. unsigned long uninitialized_var(buddy_pfn);
  690. struct page *buddy;
  691. unsigned int max_order;
  692. max_order = min_t(unsigned int, MAX_ORDER, pageblock_order + 1);
  693. VM_BUG_ON(!zone_is_initialized(zone));
  694. VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page);
  695. VM_BUG_ON(migratetype == -1);
  696. if (likely(!is_migrate_isolate(migratetype)))
  697. __mod_zone_freepage_state(zone, 1 << order, migratetype);
  698. VM_BUG_ON_PAGE(pfn & ((1 << order) - 1), page);
  699. VM_BUG_ON_PAGE(bad_range(zone, page), page);
  700. continue_merging:
  701. while (order < max_order - 1) {
  702. buddy_pfn = __find_buddy_pfn(pfn, order);
  703. buddy = page + (buddy_pfn - pfn);
  704. if (!pfn_valid_within(buddy_pfn))
  705. goto done_merging;
  706. if (!page_is_buddy(page, buddy, order))
  707. goto done_merging;
  708. /*
  709. * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
  710. * merge with it and move up one order.
  711. */
  712. if (page_is_guard(buddy)) {
  713. clear_page_guard(zone, buddy, order, migratetype);
  714. } else {
  715. list_del(&buddy->lru);
  716. zone->free_area[order].nr_free--;
  717. rmv_page_order(buddy);
  718. }
  719. combined_pfn = buddy_pfn & pfn;
  720. page = page + (combined_pfn - pfn);
  721. pfn = combined_pfn;
  722. order++;
  723. }
  724. if (max_order < MAX_ORDER) {
  725. /* If we are here, it means order is >= pageblock_order.
  726. * We want to prevent merge between freepages on isolate
  727. * pageblock and normal pageblock. Without this, pageblock
  728. * isolation could cause incorrect freepage or CMA accounting.
  729. *
  730. * We don't want to hit this code for the more frequent
  731. * low-order merging.
  732. */
  733. if (unlikely(has_isolate_pageblock(zone))) {
  734. int buddy_mt;
  735. buddy_pfn = __find_buddy_pfn(pfn, order);
  736. buddy = page + (buddy_pfn - pfn);
  737. buddy_mt = get_pageblock_migratetype(buddy);
  738. if (migratetype != buddy_mt
  739. && (is_migrate_isolate(migratetype) ||
  740. is_migrate_isolate(buddy_mt)))
  741. goto done_merging;
  742. }
  743. max_order++;
  744. goto continue_merging;
  745. }
  746. done_merging:
  747. set_page_order(page, order);
  748. /*
  749. * If this is not the largest possible page, check if the buddy
  750. * of the next-highest order is free. If it is, it's possible
  751. * that pages are being freed that will coalesce soon. In case,
  752. * that is happening, add the free page to the tail of the list
  753. * so it's less likely to be used soon and more likely to be merged
  754. * as a higher order page
  755. */
  756. if ((order < MAX_ORDER-2) && pfn_valid_within(buddy_pfn)) {
  757. struct page *higher_page, *higher_buddy;
  758. combined_pfn = buddy_pfn & pfn;
  759. higher_page = page + (combined_pfn - pfn);
  760. buddy_pfn = __find_buddy_pfn(combined_pfn, order + 1);
  761. higher_buddy = higher_page + (buddy_pfn - combined_pfn);
  762. if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
  763. list_add_tail(&page->lru,
  764. &zone->free_area[order].free_list[migratetype]);
  765. goto out;
  766. }
  767. }
  768. list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
  769. out:
  770. zone->free_area[order].nr_free++;
  771. }
  772. /*
  773. * A bad page could be due to a number of fields. Instead of multiple branches,
  774. * try and check multiple fields with one check. The caller must do a detailed
  775. * check if necessary.
  776. */
  777. static inline bool page_expected_state(struct page *page,
  778. unsigned long check_flags)
  779. {
  780. if (unlikely(atomic_read(&page->_mapcount) != -1))
  781. return false;
  782. if (unlikely((unsigned long)page->mapping |
  783. page_ref_count(page) |
  784. #ifdef CONFIG_MEMCG
  785. (unsigned long)page->mem_cgroup |
  786. #endif
  787. (page->flags & check_flags)))
  788. return false;
  789. return true;
  790. }
  791. static void free_pages_check_bad(struct page *page)
  792. {
  793. const char *bad_reason;
  794. unsigned long bad_flags;
  795. bad_reason = NULL;
  796. bad_flags = 0;
  797. if (unlikely(atomic_read(&page->_mapcount) != -1))
  798. bad_reason = "nonzero mapcount";
  799. if (unlikely(page->mapping != NULL))
  800. bad_reason = "non-NULL mapping";
  801. if (unlikely(page_ref_count(page) != 0))
  802. bad_reason = "nonzero _refcount";
  803. if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) {
  804. bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
  805. bad_flags = PAGE_FLAGS_CHECK_AT_FREE;
  806. }
  807. #ifdef CONFIG_MEMCG
  808. if (unlikely(page->mem_cgroup))
  809. bad_reason = "page still charged to cgroup";
  810. #endif
  811. bad_page(page, bad_reason, bad_flags);
  812. }
  813. static inline int free_pages_check(struct page *page)
  814. {
  815. if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE)))
  816. return 0;
  817. /* Something has gone sideways, find it */
  818. free_pages_check_bad(page);
  819. return 1;
  820. }
  821. static int free_tail_pages_check(struct page *head_page, struct page *page)
  822. {
  823. int ret = 1;
  824. /*
  825. * We rely page->lru.next never has bit 0 set, unless the page
  826. * is PageTail(). Let's make sure that's true even for poisoned ->lru.
  827. */
  828. BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1);
  829. if (!IS_ENABLED(CONFIG_DEBUG_VM)) {
  830. ret = 0;
  831. goto out;
  832. }
  833. switch (page - head_page) {
  834. case 1:
  835. /* the first tail page: ->mapping is compound_mapcount() */
  836. if (unlikely(compound_mapcount(page))) {
  837. bad_page(page, "nonzero compound_mapcount", 0);
  838. goto out;
  839. }
  840. break;
  841. case 2:
  842. /*
  843. * the second tail page: ->mapping is
  844. * page_deferred_list().next -- ignore value.
  845. */
  846. break;
  847. default:
  848. if (page->mapping != TAIL_MAPPING) {
  849. bad_page(page, "corrupted mapping in tail page", 0);
  850. goto out;
  851. }
  852. break;
  853. }
  854. if (unlikely(!PageTail(page))) {
  855. bad_page(page, "PageTail not set", 0);
  856. goto out;
  857. }
  858. if (unlikely(compound_head(page) != head_page)) {
  859. bad_page(page, "compound_head not consistent", 0);
  860. goto out;
  861. }
  862. ret = 0;
  863. out:
  864. page->mapping = NULL;
  865. clear_compound_head(page);
  866. return ret;
  867. }
  868. static __always_inline bool free_pages_prepare(struct page *page,
  869. unsigned int order, bool check_free)
  870. {
  871. int bad = 0;
  872. VM_BUG_ON_PAGE(PageTail(page), page);
  873. trace_mm_page_free(page, order);
  874. kmemcheck_free_shadow(page, order);
  875. /*
  876. * Check tail pages before head page information is cleared to
  877. * avoid checking PageCompound for order-0 pages.
  878. */
  879. if (unlikely(order)) {
  880. bool compound = PageCompound(page);
  881. int i;
  882. VM_BUG_ON_PAGE(compound && compound_order(page) != order, page);
  883. if (compound)
  884. ClearPageDoubleMap(page);
  885. for (i = 1; i < (1 << order); i++) {
  886. if (compound)
  887. bad += free_tail_pages_check(page, page + i);
  888. if (unlikely(free_pages_check(page + i))) {
  889. bad++;
  890. continue;
  891. }
  892. (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
  893. }
  894. }
  895. if (PageMappingFlags(page))
  896. page->mapping = NULL;
  897. if (memcg_kmem_enabled() && PageKmemcg(page))
  898. memcg_kmem_uncharge(page, order);
  899. if (check_free)
  900. bad += free_pages_check(page);
  901. if (bad)
  902. return false;
  903. page_cpupid_reset_last(page);
  904. page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
  905. reset_page_owner(page, order);
  906. if (!PageHighMem(page)) {
  907. debug_check_no_locks_freed(page_address(page),
  908. PAGE_SIZE << order);
  909. debug_check_no_obj_freed(page_address(page),
  910. PAGE_SIZE << order);
  911. }
  912. arch_free_page(page, order);
  913. kernel_poison_pages(page, 1 << order, 0);
  914. kernel_map_pages(page, 1 << order, 0);
  915. kasan_free_pages(page, order);
  916. return true;
  917. }
  918. #ifdef CONFIG_DEBUG_VM
  919. static inline bool free_pcp_prepare(struct page *page)
  920. {
  921. return free_pages_prepare(page, 0, true);
  922. }
  923. static inline bool bulkfree_pcp_prepare(struct page *page)
  924. {
  925. return false;
  926. }
  927. #else
  928. static bool free_pcp_prepare(struct page *page)
  929. {
  930. return free_pages_prepare(page, 0, false);
  931. }
  932. static bool bulkfree_pcp_prepare(struct page *page)
  933. {
  934. return free_pages_check(page);
  935. }
  936. #endif /* CONFIG_DEBUG_VM */
  937. /*
  938. * Frees a number of pages from the PCP lists
  939. * Assumes all pages on list are in same zone, and of same order.
  940. * count is the number of pages to free.
  941. *
  942. * If the zone was previously in an "all pages pinned" state then look to
  943. * see if this freeing clears that state.
  944. *
  945. * And clear the zone's pages_scanned counter, to hold off the "all pages are
  946. * pinned" detection logic.
  947. */
  948. static void free_pcppages_bulk(struct zone *zone, int count,
  949. struct per_cpu_pages *pcp)
  950. {
  951. int migratetype = 0;
  952. int batch_free = 0;
  953. unsigned long nr_scanned, flags;
  954. bool isolated_pageblocks;
  955. spin_lock_irqsave(&zone->lock, flags);
  956. isolated_pageblocks = has_isolate_pageblock(zone);
  957. nr_scanned = node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED);
  958. if (nr_scanned)
  959. __mod_node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED, -nr_scanned);
  960. while (count) {
  961. struct page *page;
  962. struct list_head *list;
  963. /*
  964. * Remove pages from lists in a round-robin fashion. A
  965. * batch_free count is maintained that is incremented when an
  966. * empty list is encountered. This is so more pages are freed
  967. * off fuller lists instead of spinning excessively around empty
  968. * lists
  969. */
  970. do {
  971. batch_free++;
  972. if (++migratetype == MIGRATE_PCPTYPES)
  973. migratetype = 0;
  974. list = &pcp->lists[migratetype];
  975. } while (list_empty(list));
  976. /* This is the only non-empty list. Free them all. */
  977. if (batch_free == MIGRATE_PCPTYPES)
  978. batch_free = count;
  979. do {
  980. int mt; /* migratetype of the to-be-freed page */
  981. page = list_last_entry(list, struct page, lru);
  982. /* must delete as __free_one_page list manipulates */
  983. list_del(&page->lru);
  984. mt = get_pcppage_migratetype(page);
  985. /* MIGRATE_ISOLATE page should not go to pcplists */
  986. VM_BUG_ON_PAGE(is_migrate_isolate(mt), page);
  987. /* Pageblock could have been isolated meanwhile */
  988. if (unlikely(isolated_pageblocks))
  989. mt = get_pageblock_migratetype(page);
  990. if (bulkfree_pcp_prepare(page))
  991. continue;
  992. __free_one_page(page, page_to_pfn(page), zone, 0, mt);
  993. trace_mm_page_pcpu_drain(page, 0, mt);
  994. } while (--count && --batch_free && !list_empty(list));
  995. }
  996. spin_unlock_irqrestore(&zone->lock, flags);
  997. }
  998. static void free_one_page(struct zone *zone,
  999. struct page *page, unsigned long pfn,
  1000. unsigned int order,
  1001. int migratetype)
  1002. {
  1003. unsigned long nr_scanned, flags;
  1004. spin_lock_irqsave(&zone->lock, flags);
  1005. __count_vm_events(PGFREE, 1 << order);
  1006. nr_scanned = node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED);
  1007. if (nr_scanned)
  1008. __mod_node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED, -nr_scanned);
  1009. if (unlikely(has_isolate_pageblock(zone) ||
  1010. is_migrate_isolate(migratetype))) {
  1011. migratetype = get_pfnblock_migratetype(page, pfn);
  1012. }
  1013. __free_one_page(page, pfn, zone, order, migratetype);
  1014. spin_unlock_irqrestore(&zone->lock, flags);
  1015. }
  1016. static void __meminit __init_single_page(struct page *page, unsigned long pfn,
  1017. unsigned long zone, int nid)
  1018. {
  1019. set_page_links(page, zone, nid, pfn);
  1020. init_page_count(page);
  1021. page_mapcount_reset(page);
  1022. page_cpupid_reset_last(page);
  1023. INIT_LIST_HEAD(&page->lru);
  1024. #ifdef WANT_PAGE_VIRTUAL
  1025. /* The shift won't overflow because ZONE_NORMAL is below 4G. */
  1026. if (!is_highmem_idx(zone))
  1027. set_page_address(page, __va(pfn << PAGE_SHIFT));
  1028. #endif
  1029. }
  1030. static void __meminit __init_single_pfn(unsigned long pfn, unsigned long zone,
  1031. int nid)
  1032. {
  1033. return __init_single_page(pfn_to_page(pfn), pfn, zone, nid);
  1034. }
  1035. #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
  1036. static void init_reserved_page(unsigned long pfn)
  1037. {
  1038. pg_data_t *pgdat;
  1039. int nid, zid;
  1040. if (!early_page_uninitialised(pfn))
  1041. return;
  1042. nid = early_pfn_to_nid(pfn);
  1043. pgdat = NODE_DATA(nid);
  1044. for (zid = 0; zid < MAX_NR_ZONES; zid++) {
  1045. struct zone *zone = &pgdat->node_zones[zid];
  1046. if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone))
  1047. break;
  1048. }
  1049. __init_single_pfn(pfn, zid, nid);
  1050. }
  1051. #else
  1052. static inline void init_reserved_page(unsigned long pfn)
  1053. {
  1054. }
  1055. #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
  1056. /*
  1057. * Initialised pages do not have PageReserved set. This function is
  1058. * called for each range allocated by the bootmem allocator and
  1059. * marks the pages PageReserved. The remaining valid pages are later
  1060. * sent to the buddy page allocator.
  1061. */
  1062. void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end)
  1063. {
  1064. unsigned long start_pfn = PFN_DOWN(start);
  1065. unsigned long end_pfn = PFN_UP(end);
  1066. for (; start_pfn < end_pfn; start_pfn++) {
  1067. if (pfn_valid(start_pfn)) {
  1068. struct page *page = pfn_to_page(start_pfn);
  1069. init_reserved_page(start_pfn);
  1070. /* Avoid false-positive PageTail() */
  1071. INIT_LIST_HEAD(&page->lru);
  1072. SetPageReserved(page);
  1073. }
  1074. }
  1075. }
  1076. static void __free_pages_ok(struct page *page, unsigned int order)
  1077. {
  1078. int migratetype;
  1079. unsigned long pfn = page_to_pfn(page);
  1080. if (!free_pages_prepare(page, order, true))
  1081. return;
  1082. migratetype = get_pfnblock_migratetype(page, pfn);
  1083. free_one_page(page_zone(page), page, pfn, order, migratetype);
  1084. }
  1085. static void __init __free_pages_boot_core(struct page *page, unsigned int order)
  1086. {
  1087. unsigned int nr_pages = 1 << order;
  1088. struct page *p = page;
  1089. unsigned int loop;
  1090. prefetchw(p);
  1091. for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
  1092. prefetchw(p + 1);
  1093. __ClearPageReserved(p);
  1094. set_page_count(p, 0);
  1095. }
  1096. __ClearPageReserved(p);
  1097. set_page_count(p, 0);
  1098. page_zone(page)->managed_pages += nr_pages;
  1099. set_page_refcounted(page);
  1100. __free_pages(page, order);
  1101. }
  1102. #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
  1103. defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
  1104. static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata;
  1105. int __meminit early_pfn_to_nid(unsigned long pfn)
  1106. {
  1107. static DEFINE_SPINLOCK(early_pfn_lock);
  1108. int nid;
  1109. spin_lock(&early_pfn_lock);
  1110. nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache);
  1111. if (nid < 0)
  1112. nid = first_online_node;
  1113. spin_unlock(&early_pfn_lock);
  1114. return nid;
  1115. }
  1116. #endif
  1117. #ifdef CONFIG_NODES_SPAN_OTHER_NODES
  1118. static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node,
  1119. struct mminit_pfnnid_cache *state)
  1120. {
  1121. int nid;
  1122. nid = __early_pfn_to_nid(pfn, state);
  1123. if (nid >= 0 && nid != node)
  1124. return false;
  1125. return true;
  1126. }
  1127. /* Only safe to use early in boot when initialisation is single-threaded */
  1128. static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
  1129. {
  1130. return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache);
  1131. }
  1132. #else
  1133. static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
  1134. {
  1135. return true;
  1136. }
  1137. static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node,
  1138. struct mminit_pfnnid_cache *state)
  1139. {
  1140. return true;
  1141. }
  1142. #endif
  1143. void __init __free_pages_bootmem(struct page *page, unsigned long pfn,
  1144. unsigned int order)
  1145. {
  1146. if (early_page_uninitialised(pfn))
  1147. return;
  1148. return __free_pages_boot_core(page, order);
  1149. }
  1150. /*
  1151. * Check that the whole (or subset of) a pageblock given by the interval of
  1152. * [start_pfn, end_pfn) is valid and within the same zone, before scanning it
  1153. * with the migration of free compaction scanner. The scanners then need to
  1154. * use only pfn_valid_within() check for arches that allow holes within
  1155. * pageblocks.
  1156. *
  1157. * Return struct page pointer of start_pfn, or NULL if checks were not passed.
  1158. *
  1159. * It's possible on some configurations to have a setup like node0 node1 node0
  1160. * i.e. it's possible that all pages within a zones range of pages do not
  1161. * belong to a single zone. We assume that a border between node0 and node1
  1162. * can occur within a single pageblock, but not a node0 node1 node0
  1163. * interleaving within a single pageblock. It is therefore sufficient to check
  1164. * the first and last page of a pageblock and avoid checking each individual
  1165. * page in a pageblock.
  1166. */
  1167. struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
  1168. unsigned long end_pfn, struct zone *zone)
  1169. {
  1170. struct page *start_page;
  1171. struct page *end_page;
  1172. /* end_pfn is one past the range we are checking */
  1173. end_pfn--;
  1174. if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn))
  1175. return NULL;
  1176. start_page = pfn_to_page(start_pfn);
  1177. if (page_zone(start_page) != zone)
  1178. return NULL;
  1179. end_page = pfn_to_page(end_pfn);
  1180. /* This gives a shorter code than deriving page_zone(end_page) */
  1181. if (page_zone_id(start_page) != page_zone_id(end_page))
  1182. return NULL;
  1183. return start_page;
  1184. }
  1185. void set_zone_contiguous(struct zone *zone)
  1186. {
  1187. unsigned long block_start_pfn = zone->zone_start_pfn;
  1188. unsigned long block_end_pfn;
  1189. block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages);
  1190. for (; block_start_pfn < zone_end_pfn(zone);
  1191. block_start_pfn = block_end_pfn,
  1192. block_end_pfn += pageblock_nr_pages) {
  1193. block_end_pfn = min(block_end_pfn, zone_end_pfn(zone));
  1194. if (!__pageblock_pfn_to_page(block_start_pfn,
  1195. block_end_pfn, zone))
  1196. return;
  1197. }
  1198. /* We confirm that there is no hole */
  1199. zone->contiguous = true;
  1200. }
  1201. void clear_zone_contiguous(struct zone *zone)
  1202. {
  1203. zone->contiguous = false;
  1204. }
  1205. #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
  1206. static void __init deferred_free_range(struct page *page,
  1207. unsigned long pfn, int nr_pages)
  1208. {
  1209. int i;
  1210. if (!page)
  1211. return;
  1212. /* Free a large naturally-aligned chunk if possible */
  1213. if (nr_pages == pageblock_nr_pages &&
  1214. (pfn & (pageblock_nr_pages - 1)) == 0) {
  1215. set_pageblock_migratetype(page, MIGRATE_MOVABLE);
  1216. __free_pages_boot_core(page, pageblock_order);
  1217. return;
  1218. }
  1219. for (i = 0; i < nr_pages; i++, page++, pfn++) {
  1220. if ((pfn & (pageblock_nr_pages - 1)) == 0)
  1221. set_pageblock_migratetype(page, MIGRATE_MOVABLE);
  1222. __free_pages_boot_core(page, 0);
  1223. }
  1224. }
  1225. /* Completion tracking for deferred_init_memmap() threads */
  1226. static atomic_t pgdat_init_n_undone __initdata;
  1227. static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp);
  1228. static inline void __init pgdat_init_report_one_done(void)
  1229. {
  1230. if (atomic_dec_and_test(&pgdat_init_n_undone))
  1231. complete(&pgdat_init_all_done_comp);
  1232. }
  1233. /* Initialise remaining memory on a node */
  1234. static int __init deferred_init_memmap(void *data)
  1235. {
  1236. pg_data_t *pgdat = data;
  1237. int nid = pgdat->node_id;
  1238. struct mminit_pfnnid_cache nid_init_state = { };
  1239. unsigned long start = jiffies;
  1240. unsigned long nr_pages = 0;
  1241. unsigned long walk_start, walk_end;
  1242. int i, zid;
  1243. struct zone *zone;
  1244. unsigned long first_init_pfn = pgdat->first_deferred_pfn;
  1245. const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
  1246. if (first_init_pfn == ULONG_MAX) {
  1247. pgdat_init_report_one_done();
  1248. return 0;
  1249. }
  1250. /* Bind memory initialisation thread to a local node if possible */
  1251. if (!cpumask_empty(cpumask))
  1252. set_cpus_allowed_ptr(current, cpumask);
  1253. /* Sanity check boundaries */
  1254. BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn);
  1255. BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat));
  1256. pgdat->first_deferred_pfn = ULONG_MAX;
  1257. /* Only the highest zone is deferred so find it */
  1258. for (zid = 0; zid < MAX_NR_ZONES; zid++) {
  1259. zone = pgdat->node_zones + zid;
  1260. if (first_init_pfn < zone_end_pfn(zone))
  1261. break;
  1262. }
  1263. for_each_mem_pfn_range(i, nid, &walk_start, &walk_end, NULL) {
  1264. unsigned long pfn, end_pfn;
  1265. struct page *page = NULL;
  1266. struct page *free_base_page = NULL;
  1267. unsigned long free_base_pfn = 0;
  1268. int nr_to_free = 0;
  1269. end_pfn = min(walk_end, zone_end_pfn(zone));
  1270. pfn = first_init_pfn;
  1271. if (pfn < walk_start)
  1272. pfn = walk_start;
  1273. if (pfn < zone->zone_start_pfn)
  1274. pfn = zone->zone_start_pfn;
  1275. for (; pfn < end_pfn; pfn++) {
  1276. if (!pfn_valid_within(pfn))
  1277. goto free_range;
  1278. /*
  1279. * Ensure pfn_valid is checked every
  1280. * pageblock_nr_pages for memory holes
  1281. */
  1282. if ((pfn & (pageblock_nr_pages - 1)) == 0) {
  1283. if (!pfn_valid(pfn)) {
  1284. page = NULL;
  1285. goto free_range;
  1286. }
  1287. }
  1288. if (!meminit_pfn_in_nid(pfn, nid, &nid_init_state)) {
  1289. page = NULL;
  1290. goto free_range;
  1291. }
  1292. /* Minimise pfn page lookups and scheduler checks */
  1293. if (page && (pfn & (pageblock_nr_pages - 1)) != 0) {
  1294. page++;
  1295. } else {
  1296. nr_pages += nr_to_free;
  1297. deferred_free_range(free_base_page,
  1298. free_base_pfn, nr_to_free);
  1299. free_base_page = NULL;
  1300. free_base_pfn = nr_to_free = 0;
  1301. page = pfn_to_page(pfn);
  1302. cond_resched();
  1303. }
  1304. if (page->flags) {
  1305. VM_BUG_ON(page_zone(page) != zone);
  1306. goto free_range;
  1307. }
  1308. __init_single_page(page, pfn, zid, nid);
  1309. if (!free_base_page) {
  1310. free_base_page = page;
  1311. free_base_pfn = pfn;
  1312. nr_to_free = 0;
  1313. }
  1314. nr_to_free++;
  1315. /* Where possible, batch up pages for a single free */
  1316. continue;
  1317. free_range:
  1318. /* Free the current block of pages to allocator */
  1319. nr_pages += nr_to_free;
  1320. deferred_free_range(free_base_page, free_base_pfn,
  1321. nr_to_free);
  1322. free_base_page = NULL;
  1323. free_base_pfn = nr_to_free = 0;
  1324. }
  1325. /* Free the last block of pages to allocator */
  1326. nr_pages += nr_to_free;
  1327. deferred_free_range(free_base_page, free_base_pfn, nr_to_free);
  1328. first_init_pfn = max(end_pfn, first_init_pfn);
  1329. }
  1330. /* Sanity check that the next zone really is unpopulated */
  1331. WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone));
  1332. pr_info("node %d initialised, %lu pages in %ums\n", nid, nr_pages,
  1333. jiffies_to_msecs(jiffies - start));
  1334. pgdat_init_report_one_done();
  1335. return 0;
  1336. }
  1337. #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
  1338. void __init page_alloc_init_late(void)
  1339. {
  1340. struct zone *zone;
  1341. #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
  1342. int nid;
  1343. /* There will be num_node_state(N_MEMORY) threads */
  1344. atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY));
  1345. for_each_node_state(nid, N_MEMORY) {
  1346. kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid);
  1347. }
  1348. /* Block until all are initialised */
  1349. wait_for_completion(&pgdat_init_all_done_comp);
  1350. /* Reinit limits that are based on free pages after the kernel is up */
  1351. files_maxfiles_init();
  1352. #endif
  1353. for_each_populated_zone(zone)
  1354. set_zone_contiguous(zone);
  1355. }
  1356. #ifdef CONFIG_CMA
  1357. /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
  1358. void __init init_cma_reserved_pageblock(struct page *page)
  1359. {
  1360. unsigned i = pageblock_nr_pages;
  1361. struct page *p = page;
  1362. do {
  1363. __ClearPageReserved(p);
  1364. set_page_count(p, 0);
  1365. } while (++p, --i);
  1366. set_pageblock_migratetype(page, MIGRATE_CMA);
  1367. if (pageblock_order >= MAX_ORDER) {
  1368. i = pageblock_nr_pages;
  1369. p = page;
  1370. do {
  1371. set_page_refcounted(p);
  1372. __free_pages(p, MAX_ORDER - 1);
  1373. p += MAX_ORDER_NR_PAGES;
  1374. } while (i -= MAX_ORDER_NR_PAGES);
  1375. } else {
  1376. set_page_refcounted(page);
  1377. __free_pages(page, pageblock_order);
  1378. }
  1379. adjust_managed_page_count(page, pageblock_nr_pages);
  1380. }
  1381. #endif
  1382. /*
  1383. * The order of subdivision here is critical for the IO subsystem.
  1384. * Please do not alter this order without good reasons and regression
  1385. * testing. Specifically, as large blocks of memory are subdivided,
  1386. * the order in which smaller blocks are delivered depends on the order
  1387. * they're subdivided in this function. This is the primary factor
  1388. * influencing the order in which pages are delivered to the IO
  1389. * subsystem according to empirical testing, and this is also justified
  1390. * by considering the behavior of a buddy system containing a single
  1391. * large block of memory acted on by a series of small allocations.
  1392. * This behavior is a critical factor in sglist merging's success.
  1393. *
  1394. * -- nyc
  1395. */
  1396. static inline void expand(struct zone *zone, struct page *page,
  1397. int low, int high, struct free_area *area,
  1398. int migratetype)
  1399. {
  1400. unsigned long size = 1 << high;
  1401. while (high > low) {
  1402. area--;
  1403. high--;
  1404. size >>= 1;
  1405. VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
  1406. /*
  1407. * Mark as guard pages (or page), that will allow to
  1408. * merge back to allocator when buddy will be freed.
  1409. * Corresponding page table entries will not be touched,
  1410. * pages will stay not present in virtual address space
  1411. */
  1412. if (set_page_guard(zone, &page[size], high, migratetype))
  1413. continue;
  1414. list_add(&page[size].lru, &area->free_list[migratetype]);
  1415. area->nr_free++;
  1416. set_page_order(&page[size], high);
  1417. }
  1418. }
  1419. static void check_new_page_bad(struct page *page)
  1420. {
  1421. const char *bad_reason = NULL;
  1422. unsigned long bad_flags = 0;
  1423. if (unlikely(atomic_read(&page->_mapcount) != -1))
  1424. bad_reason = "nonzero mapcount";
  1425. if (unlikely(page->mapping != NULL))
  1426. bad_reason = "non-NULL mapping";
  1427. if (unlikely(page_ref_count(page) != 0))
  1428. bad_reason = "nonzero _count";
  1429. if (unlikely(page->flags & __PG_HWPOISON)) {
  1430. bad_reason = "HWPoisoned (hardware-corrupted)";
  1431. bad_flags = __PG_HWPOISON;
  1432. /* Don't complain about hwpoisoned pages */
  1433. page_mapcount_reset(page); /* remove PageBuddy */
  1434. return;
  1435. }
  1436. if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) {
  1437. bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set";
  1438. bad_flags = PAGE_FLAGS_CHECK_AT_PREP;
  1439. }
  1440. #ifdef CONFIG_MEMCG
  1441. if (unlikely(page->mem_cgroup))
  1442. bad_reason = "page still charged to cgroup";
  1443. #endif
  1444. bad_page(page, bad_reason, bad_flags);
  1445. }
  1446. /*
  1447. * This page is about to be returned from the page allocator
  1448. */
  1449. static inline int check_new_page(struct page *page)
  1450. {
  1451. if (likely(page_expected_state(page,
  1452. PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON)))
  1453. return 0;
  1454. check_new_page_bad(page);
  1455. return 1;
  1456. }
  1457. static inline bool free_pages_prezeroed(bool poisoned)
  1458. {
  1459. return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) &&
  1460. page_poisoning_enabled() && poisoned;
  1461. }
  1462. #ifdef CONFIG_DEBUG_VM
  1463. static bool check_pcp_refill(struct page *page)
  1464. {
  1465. return false;
  1466. }
  1467. static bool check_new_pcp(struct page *page)
  1468. {
  1469. return check_new_page(page);
  1470. }
  1471. #else
  1472. static bool check_pcp_refill(struct page *page)
  1473. {
  1474. return check_new_page(page);
  1475. }
  1476. static bool check_new_pcp(struct page *page)
  1477. {
  1478. return false;
  1479. }
  1480. #endif /* CONFIG_DEBUG_VM */
  1481. static bool check_new_pages(struct page *page, unsigned int order)
  1482. {
  1483. int i;
  1484. for (i = 0; i < (1 << order); i++) {
  1485. struct page *p = page + i;
  1486. if (unlikely(check_new_page(p)))
  1487. return true;
  1488. }
  1489. return false;
  1490. }
  1491. inline void post_alloc_hook(struct page *page, unsigned int order,
  1492. gfp_t gfp_flags)
  1493. {
  1494. set_page_private(page, 0);
  1495. set_page_refcounted(page);
  1496. arch_alloc_page(page, order);
  1497. kernel_map_pages(page, 1 << order, 1);
  1498. kernel_poison_pages(page, 1 << order, 1);
  1499. kasan_alloc_pages(page, order);
  1500. set_page_owner(page, order, gfp_flags);
  1501. }
  1502. static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags,
  1503. unsigned int alloc_flags)
  1504. {
  1505. int i;
  1506. bool poisoned = true;
  1507. for (i = 0; i < (1 << order); i++) {
  1508. struct page *p = page + i;
  1509. if (poisoned)
  1510. poisoned &= page_is_poisoned(p);
  1511. }
  1512. post_alloc_hook(page, order, gfp_flags);
  1513. if (!free_pages_prezeroed(poisoned) && (gfp_flags & __GFP_ZERO))
  1514. for (i = 0; i < (1 << order); i++)
  1515. clear_highpage(page + i);
  1516. if (order && (gfp_flags & __GFP_COMP))
  1517. prep_compound_page(page, order);
  1518. /*
  1519. * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
  1520. * allocate the page. The expectation is that the caller is taking
  1521. * steps that will free more memory. The caller should avoid the page
  1522. * being used for !PFMEMALLOC purposes.
  1523. */
  1524. if (alloc_flags & ALLOC_NO_WATERMARKS)
  1525. set_page_pfmemalloc(page);
  1526. else
  1527. clear_page_pfmemalloc(page);
  1528. }
  1529. /*
  1530. * Go through the free lists for the given migratetype and remove
  1531. * the smallest available page from the freelists
  1532. */
  1533. static inline
  1534. struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
  1535. int migratetype)
  1536. {
  1537. unsigned int current_order;
  1538. struct free_area *area;
  1539. struct page *page;
  1540. /* Find a page of the appropriate size in the preferred list */
  1541. for (current_order = order; current_order < MAX_ORDER; ++current_order) {
  1542. area = &(zone->free_area[current_order]);
  1543. page = list_first_entry_or_null(&area->free_list[migratetype],
  1544. struct page, lru);
  1545. if (!page)
  1546. continue;
  1547. list_del(&page->lru);
  1548. rmv_page_order(page);
  1549. area->nr_free--;
  1550. expand(zone, page, order, current_order, area, migratetype);
  1551. set_pcppage_migratetype(page, migratetype);
  1552. return page;
  1553. }
  1554. return NULL;
  1555. }
  1556. /*
  1557. * This array describes the order lists are fallen back to when
  1558. * the free lists for the desirable migrate type are depleted
  1559. */
  1560. static int fallbacks[MIGRATE_TYPES][4] = {
  1561. [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES },
  1562. [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES },
  1563. [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES },
  1564. #ifdef CONFIG_CMA
  1565. [MIGRATE_CMA] = { MIGRATE_TYPES }, /* Never used */
  1566. #endif
  1567. #ifdef CONFIG_MEMORY_ISOLATION
  1568. [MIGRATE_ISOLATE] = { MIGRATE_TYPES }, /* Never used */
  1569. #endif
  1570. };
  1571. #ifdef CONFIG_CMA
  1572. static struct page *__rmqueue_cma_fallback(struct zone *zone,
  1573. unsigned int order)
  1574. {
  1575. return __rmqueue_smallest(zone, order, MIGRATE_CMA);
  1576. }
  1577. #else
  1578. static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
  1579. unsigned int order) { return NULL; }
  1580. #endif
  1581. /*
  1582. * Move the free pages in a range to the free lists of the requested type.
  1583. * Note that start_page and end_pages are not aligned on a pageblock
  1584. * boundary. If alignment is required, use move_freepages_block()
  1585. */
  1586. int move_freepages(struct zone *zone,
  1587. struct page *start_page, struct page *end_page,
  1588. int migratetype)
  1589. {
  1590. struct page *page;
  1591. unsigned int order;
  1592. int pages_moved = 0;
  1593. #ifndef CONFIG_HOLES_IN_ZONE
  1594. /*
  1595. * page_zone is not safe to call in this context when
  1596. * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
  1597. * anyway as we check zone boundaries in move_freepages_block().
  1598. * Remove at a later date when no bug reports exist related to
  1599. * grouping pages by mobility
  1600. */
  1601. VM_BUG_ON(page_zone(start_page) != page_zone(end_page));
  1602. #endif
  1603. for (page = start_page; page <= end_page;) {
  1604. if (!pfn_valid_within(page_to_pfn(page))) {
  1605. page++;
  1606. continue;
  1607. }
  1608. /* Make sure we are not inadvertently changing nodes */
  1609. VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
  1610. if (!PageBuddy(page)) {
  1611. page++;
  1612. continue;
  1613. }
  1614. order = page_order(page);
  1615. list_move(&page->lru,
  1616. &zone->free_area[order].free_list[migratetype]);
  1617. page += 1 << order;
  1618. pages_moved += 1 << order;
  1619. }
  1620. return pages_moved;
  1621. }
  1622. int move_freepages_block(struct zone *zone, struct page *page,
  1623. int migratetype)
  1624. {
  1625. unsigned long start_pfn, end_pfn;
  1626. struct page *start_page, *end_page;
  1627. start_pfn = page_to_pfn(page);
  1628. start_pfn = start_pfn & ~(pageblock_nr_pages-1);
  1629. start_page = pfn_to_page(start_pfn);
  1630. end_page = start_page + pageblock_nr_pages - 1;
  1631. end_pfn = start_pfn + pageblock_nr_pages - 1;
  1632. /* Do not cross zone boundaries */
  1633. if (!zone_spans_pfn(zone, start_pfn))
  1634. start_page = page;
  1635. if (!zone_spans_pfn(zone, end_pfn))
  1636. return 0;
  1637. return move_freepages(zone, start_page, end_page, migratetype);
  1638. }
  1639. static void change_pageblock_range(struct page *pageblock_page,
  1640. int start_order, int migratetype)
  1641. {
  1642. int nr_pageblocks = 1 << (start_order - pageblock_order);
  1643. while (nr_pageblocks--) {
  1644. set_pageblock_migratetype(pageblock_page, migratetype);
  1645. pageblock_page += pageblock_nr_pages;
  1646. }
  1647. }
  1648. /*
  1649. * When we are falling back to another migratetype during allocation, try to
  1650. * steal extra free pages from the same pageblocks to satisfy further
  1651. * allocations, instead of polluting multiple pageblocks.
  1652. *
  1653. * If we are stealing a relatively large buddy page, it is likely there will
  1654. * be more free pages in the pageblock, so try to steal them all. For
  1655. * reclaimable and unmovable allocations, we steal regardless of page size,
  1656. * as fragmentation caused by those allocations polluting movable pageblocks
  1657. * is worse than movable allocations stealing from unmovable and reclaimable
  1658. * pageblocks.
  1659. */
  1660. static bool can_steal_fallback(unsigned int order, int start_mt)
  1661. {
  1662. /*
  1663. * Leaving this order check is intended, although there is
  1664. * relaxed order check in next check. The reason is that
  1665. * we can actually steal whole pageblock if this condition met,
  1666. * but, below check doesn't guarantee it and that is just heuristic
  1667. * so could be changed anytime.
  1668. */
  1669. if (order >= pageblock_order)
  1670. return true;
  1671. if (order >= pageblock_order / 2 ||
  1672. start_mt == MIGRATE_RECLAIMABLE ||
  1673. start_mt == MIGRATE_UNMOVABLE ||
  1674. page_group_by_mobility_disabled)
  1675. return true;
  1676. return false;
  1677. }
  1678. /*
  1679. * This function implements actual steal behaviour. If order is large enough,
  1680. * we can steal whole pageblock. If not, we first move freepages in this
  1681. * pageblock and check whether half of pages are moved or not. If half of
  1682. * pages are moved, we can change migratetype of pageblock and permanently
  1683. * use it's pages as requested migratetype in the future.
  1684. */
  1685. static void steal_suitable_fallback(struct zone *zone, struct page *page,
  1686. int start_type)
  1687. {
  1688. unsigned int current_order = page_order(page);
  1689. int pages;
  1690. /* Take ownership for orders >= pageblock_order */
  1691. if (current_order >= pageblock_order) {
  1692. change_pageblock_range(page, current_order, start_type);
  1693. return;
  1694. }
  1695. pages = move_freepages_block(zone, page, start_type);
  1696. /* Claim the whole block if over half of it is free */
  1697. if (pages >= (1 << (pageblock_order-1)) ||
  1698. page_group_by_mobility_disabled)
  1699. set_pageblock_migratetype(page, start_type);
  1700. }
  1701. /*
  1702. * Check whether there is a suitable fallback freepage with requested order.
  1703. * If only_stealable is true, this function returns fallback_mt only if
  1704. * we can steal other freepages all together. This would help to reduce
  1705. * fragmentation due to mixed migratetype pages in one pageblock.
  1706. */
  1707. int find_suitable_fallback(struct free_area *area, unsigned int order,
  1708. int migratetype, bool only_stealable, bool *can_steal)
  1709. {
  1710. int i;
  1711. int fallback_mt;
  1712. if (area->nr_free == 0)
  1713. return -1;
  1714. *can_steal = false;
  1715. for (i = 0;; i++) {
  1716. fallback_mt = fallbacks[migratetype][i];
  1717. if (fallback_mt == MIGRATE_TYPES)
  1718. break;
  1719. if (list_empty(&area->free_list[fallback_mt]))
  1720. continue;
  1721. if (can_steal_fallback(order, migratetype))
  1722. *can_steal = true;
  1723. if (!only_stealable)
  1724. return fallback_mt;
  1725. if (*can_steal)
  1726. return fallback_mt;
  1727. }
  1728. return -1;
  1729. }
  1730. /*
  1731. * Reserve a pageblock for exclusive use of high-order atomic allocations if
  1732. * there are no empty page blocks that contain a page with a suitable order
  1733. */
  1734. static void reserve_highatomic_pageblock(struct page *page, struct zone *zone,
  1735. unsigned int alloc_order)
  1736. {
  1737. int mt;
  1738. unsigned long max_managed, flags;
  1739. /*
  1740. * Limit the number reserved to 1 pageblock or roughly 1% of a zone.
  1741. * Check is race-prone but harmless.
  1742. */
  1743. max_managed = (zone->managed_pages / 100) + pageblock_nr_pages;
  1744. if (zone->nr_reserved_highatomic >= max_managed)
  1745. return;
  1746. spin_lock_irqsave(&zone->lock, flags);
  1747. /* Recheck the nr_reserved_highatomic limit under the lock */
  1748. if (zone->nr_reserved_highatomic >= max_managed)
  1749. goto out_unlock;
  1750. /* Yoink! */
  1751. mt = get_pageblock_migratetype(page);
  1752. if (mt != MIGRATE_HIGHATOMIC &&
  1753. !is_migrate_isolate(mt) && !is_migrate_cma(mt)) {
  1754. zone->nr_reserved_highatomic += pageblock_nr_pages;
  1755. set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC);
  1756. move_freepages_block(zone, page, MIGRATE_HIGHATOMIC);
  1757. }
  1758. out_unlock:
  1759. spin_unlock_irqrestore(&zone->lock, flags);
  1760. }
  1761. /*
  1762. * Used when an allocation is about to fail under memory pressure. This
  1763. * potentially hurts the reliability of high-order allocations when under
  1764. * intense memory pressure but failed atomic allocations should be easier
  1765. * to recover from than an OOM.
  1766. *
  1767. * If @force is true, try to unreserve a pageblock even though highatomic
  1768. * pageblock is exhausted.
  1769. */
  1770. static bool unreserve_highatomic_pageblock(const struct alloc_context *ac,
  1771. bool force)
  1772. {
  1773. struct zonelist *zonelist = ac->zonelist;
  1774. unsigned long flags;
  1775. struct zoneref *z;
  1776. struct zone *zone;
  1777. struct page *page;
  1778. int order;
  1779. bool ret;
  1780. for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx,
  1781. ac->nodemask) {
  1782. /*
  1783. * Preserve at least one pageblock unless memory pressure
  1784. * is really high.
  1785. */
  1786. if (!force && zone->nr_reserved_highatomic <=
  1787. pageblock_nr_pages)
  1788. continue;
  1789. spin_lock_irqsave(&zone->lock, flags);
  1790. for (order = 0; order < MAX_ORDER; order++) {
  1791. struct free_area *area = &(zone->free_area[order]);
  1792. page = list_first_entry_or_null(
  1793. &area->free_list[MIGRATE_HIGHATOMIC],
  1794. struct page, lru);
  1795. if (!page)
  1796. continue;
  1797. /*
  1798. * In page freeing path, migratetype change is racy so
  1799. * we can counter several free pages in a pageblock
  1800. * in this loop althoug we changed the pageblock type
  1801. * from highatomic to ac->migratetype. So we should
  1802. * adjust the count once.
  1803. */
  1804. if (get_pageblock_migratetype(page) ==
  1805. MIGRATE_HIGHATOMIC) {
  1806. /*
  1807. * It should never happen but changes to
  1808. * locking could inadvertently allow a per-cpu
  1809. * drain to add pages to MIGRATE_HIGHATOMIC
  1810. * while unreserving so be safe and watch for
  1811. * underflows.
  1812. */
  1813. zone->nr_reserved_highatomic -= min(
  1814. pageblock_nr_pages,
  1815. zone->nr_reserved_highatomic);
  1816. }
  1817. /*
  1818. * Convert to ac->migratetype and avoid the normal
  1819. * pageblock stealing heuristics. Minimally, the caller
  1820. * is doing the work and needs the pages. More
  1821. * importantly, if the block was always converted to
  1822. * MIGRATE_UNMOVABLE or another type then the number
  1823. * of pageblocks that cannot be completely freed
  1824. * may increase.
  1825. */
  1826. set_pageblock_migratetype(page, ac->migratetype);
  1827. ret = move_freepages_block(zone, page, ac->migratetype);
  1828. if (ret) {
  1829. spin_unlock_irqrestore(&zone->lock, flags);
  1830. return ret;
  1831. }
  1832. }
  1833. spin_unlock_irqrestore(&zone->lock, flags);
  1834. }
  1835. return false;
  1836. }
  1837. /* Remove an element from the buddy allocator from the fallback list */
  1838. static inline struct page *
  1839. __rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype)
  1840. {
  1841. struct free_area *area;
  1842. unsigned int current_order;
  1843. struct page *page;
  1844. int fallback_mt;
  1845. bool can_steal;
  1846. /* Find the largest possible block of pages in the other list */
  1847. for (current_order = MAX_ORDER-1;
  1848. current_order >= order && current_order <= MAX_ORDER-1;
  1849. --current_order) {
  1850. area = &(zone->free_area[current_order]);
  1851. fallback_mt = find_suitable_fallback(area, current_order,
  1852. start_migratetype, false, &can_steal);
  1853. if (fallback_mt == -1)
  1854. continue;
  1855. page = list_first_entry(&area->free_list[fallback_mt],
  1856. struct page, lru);
  1857. if (can_steal &&
  1858. get_pageblock_migratetype(page) != MIGRATE_HIGHATOMIC)
  1859. steal_suitable_fallback(zone, page, start_migratetype);
  1860. /* Remove the page from the freelists */
  1861. area->nr_free--;
  1862. list_del(&page->lru);
  1863. rmv_page_order(page);
  1864. expand(zone, page, order, current_order, area,
  1865. start_migratetype);
  1866. /*
  1867. * The pcppage_migratetype may differ from pageblock's
  1868. * migratetype depending on the decisions in
  1869. * find_suitable_fallback(). This is OK as long as it does not
  1870. * differ for MIGRATE_CMA pageblocks. Those can be used as
  1871. * fallback only via special __rmqueue_cma_fallback() function
  1872. */
  1873. set_pcppage_migratetype(page, start_migratetype);
  1874. trace_mm_page_alloc_extfrag(page, order, current_order,
  1875. start_migratetype, fallback_mt);
  1876. return page;
  1877. }
  1878. return NULL;
  1879. }
  1880. /*
  1881. * Do the hard work of removing an element from the buddy allocator.
  1882. * Call me with the zone->lock already held.
  1883. */
  1884. static struct page *__rmqueue(struct zone *zone, unsigned int order,
  1885. int migratetype)
  1886. {
  1887. struct page *page;
  1888. page = __rmqueue_smallest(zone, order, migratetype);
  1889. if (unlikely(!page)) {
  1890. if (migratetype == MIGRATE_MOVABLE)
  1891. page = __rmqueue_cma_fallback(zone, order);
  1892. if (!page)
  1893. page = __rmqueue_fallback(zone, order, migratetype);
  1894. }
  1895. trace_mm_page_alloc_zone_locked(page, order, migratetype);
  1896. return page;
  1897. }
  1898. /*
  1899. * Obtain a specified number of elements from the buddy allocator, all under
  1900. * a single hold of the lock, for efficiency. Add them to the supplied list.
  1901. * Returns the number of new pages which were placed at *list.
  1902. */
  1903. static int rmqueue_bulk(struct zone *zone, unsigned int order,
  1904. unsigned long count, struct list_head *list,
  1905. int migratetype, bool cold)
  1906. {
  1907. int i, alloced = 0;
  1908. unsigned long flags;
  1909. spin_lock_irqsave(&zone->lock, flags);
  1910. for (i = 0; i < count; ++i) {
  1911. struct page *page = __rmqueue(zone, order, migratetype);
  1912. if (unlikely(page == NULL))
  1913. break;
  1914. if (unlikely(check_pcp_refill(page)))
  1915. continue;
  1916. /*
  1917. * Split buddy pages returned by expand() are received here
  1918. * in physical page order. The page is added to the callers and
  1919. * list and the list head then moves forward. From the callers
  1920. * perspective, the linked list is ordered by page number in
  1921. * some conditions. This is useful for IO devices that can
  1922. * merge IO requests if the physical pages are ordered
  1923. * properly.
  1924. */
  1925. if (likely(!cold))
  1926. list_add(&page->lru, list);
  1927. else
  1928. list_add_tail(&page->lru, list);
  1929. list = &page->lru;
  1930. alloced++;
  1931. if (is_migrate_cma(get_pcppage_migratetype(page)))
  1932. __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
  1933. -(1 << order));
  1934. }
  1935. /*
  1936. * i pages were removed from the buddy list even if some leak due
  1937. * to check_pcp_refill failing so adjust NR_FREE_PAGES based
  1938. * on i. Do not confuse with 'alloced' which is the number of
  1939. * pages added to the pcp list.
  1940. */
  1941. __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
  1942. spin_unlock_irqrestore(&zone->lock, flags);
  1943. return alloced;
  1944. }
  1945. #ifdef CONFIG_NUMA
  1946. /*
  1947. * Called from the vmstat counter updater to drain pagesets of this
  1948. * currently executing processor on remote nodes after they have
  1949. * expired.
  1950. *
  1951. * Note that this function must be called with the thread pinned to
  1952. * a single processor.
  1953. */
  1954. void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
  1955. {
  1956. unsigned long flags;
  1957. int to_drain, batch;
  1958. local_irq_save(flags);
  1959. batch = READ_ONCE(pcp->batch);
  1960. to_drain = min(pcp->count, batch);
  1961. if (to_drain > 0) {
  1962. free_pcppages_bulk(zone, to_drain, pcp);
  1963. pcp->count -= to_drain;
  1964. }
  1965. local_irq_restore(flags);
  1966. }
  1967. #endif
  1968. /*
  1969. * Drain pcplists of the indicated processor and zone.
  1970. *
  1971. * The processor must either be the current processor and the
  1972. * thread pinned to the current processor or a processor that
  1973. * is not online.
  1974. */
  1975. static void drain_pages_zone(unsigned int cpu, struct zone *zone)
  1976. {
  1977. unsigned long flags;
  1978. struct per_cpu_pageset *pset;
  1979. struct per_cpu_pages *pcp;
  1980. local_irq_save(flags);
  1981. pset = per_cpu_ptr(zone->pageset, cpu);
  1982. pcp = &pset->pcp;
  1983. if (pcp->count) {
  1984. free_pcppages_bulk(zone, pcp->count, pcp);
  1985. pcp->count = 0;
  1986. }
  1987. local_irq_restore(flags);
  1988. }
  1989. /*
  1990. * Drain pcplists of all zones on the indicated processor.
  1991. *
  1992. * The processor must either be the current processor and the
  1993. * thread pinned to the current processor or a processor that
  1994. * is not online.
  1995. */
  1996. static void drain_pages(unsigned int cpu)
  1997. {
  1998. struct zone *zone;
  1999. for_each_populated_zone(zone) {
  2000. drain_pages_zone(cpu, zone);
  2001. }
  2002. }
  2003. /*
  2004. * Spill all of this CPU's per-cpu pages back into the buddy allocator.
  2005. *
  2006. * The CPU has to be pinned. When zone parameter is non-NULL, spill just
  2007. * the single zone's pages.
  2008. */
  2009. void drain_local_pages(struct zone *zone)
  2010. {
  2011. int cpu = smp_processor_id();
  2012. if (zone)
  2013. drain_pages_zone(cpu, zone);
  2014. else
  2015. drain_pages(cpu);
  2016. }
  2017. static void drain_local_pages_wq(struct work_struct *work)
  2018. {
  2019. /*
  2020. * drain_all_pages doesn't use proper cpu hotplug protection so
  2021. * we can race with cpu offline when the WQ can move this from
  2022. * a cpu pinned worker to an unbound one. We can operate on a different
  2023. * cpu which is allright but we also have to make sure to not move to
  2024. * a different one.
  2025. */
  2026. preempt_disable();
  2027. drain_local_pages(NULL);
  2028. preempt_enable();
  2029. }
  2030. /*
  2031. * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
  2032. *
  2033. * When zone parameter is non-NULL, spill just the single zone's pages.
  2034. *
  2035. * Note that this can be extremely slow as the draining happens in a workqueue.
  2036. */
  2037. void drain_all_pages(struct zone *zone)
  2038. {
  2039. int cpu;
  2040. /*
  2041. * Allocate in the BSS so we wont require allocation in
  2042. * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
  2043. */
  2044. static cpumask_t cpus_with_pcps;
  2045. /* Workqueues cannot recurse */
  2046. if (current->flags & PF_WQ_WORKER)
  2047. return;
  2048. /*
  2049. * Do not drain if one is already in progress unless it's specific to
  2050. * a zone. Such callers are primarily CMA and memory hotplug and need
  2051. * the drain to be complete when the call returns.
  2052. */
  2053. if (unlikely(!mutex_trylock(&pcpu_drain_mutex))) {
  2054. if (!zone)
  2055. return;
  2056. mutex_lock(&pcpu_drain_mutex);
  2057. }
  2058. /*
  2059. * We don't care about racing with CPU hotplug event
  2060. * as offline notification will cause the notified
  2061. * cpu to drain that CPU pcps and on_each_cpu_mask
  2062. * disables preemption as part of its processing
  2063. */
  2064. for_each_online_cpu(cpu) {
  2065. struct per_cpu_pageset *pcp;
  2066. struct zone *z;
  2067. bool has_pcps = false;
  2068. if (zone) {
  2069. pcp = per_cpu_ptr(zone->pageset, cpu);
  2070. if (pcp->pcp.count)
  2071. has_pcps = true;
  2072. } else {
  2073. for_each_populated_zone(z) {
  2074. pcp = per_cpu_ptr(z->pageset, cpu);
  2075. if (pcp->pcp.count) {
  2076. has_pcps = true;
  2077. break;
  2078. }
  2079. }
  2080. }
  2081. if (has_pcps)
  2082. cpumask_set_cpu(cpu, &cpus_with_pcps);
  2083. else
  2084. cpumask_clear_cpu(cpu, &cpus_with_pcps);
  2085. }
  2086. for_each_cpu(cpu, &cpus_with_pcps) {
  2087. struct work_struct *work = per_cpu_ptr(&pcpu_drain, cpu);
  2088. INIT_WORK(work, drain_local_pages_wq);
  2089. schedule_work_on(cpu, work);
  2090. }
  2091. for_each_cpu(cpu, &cpus_with_pcps)
  2092. flush_work(per_cpu_ptr(&pcpu_drain, cpu));
  2093. mutex_unlock(&pcpu_drain_mutex);
  2094. }
  2095. #ifdef CONFIG_HIBERNATION
  2096. void mark_free_pages(struct zone *zone)
  2097. {
  2098. unsigned long pfn, max_zone_pfn;
  2099. unsigned long flags;
  2100. unsigned int order, t;
  2101. struct page *page;
  2102. if (zone_is_empty(zone))
  2103. return;
  2104. spin_lock_irqsave(&zone->lock, flags);
  2105. max_zone_pfn = zone_end_pfn(zone);
  2106. for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
  2107. if (pfn_valid(pfn)) {
  2108. page = pfn_to_page(pfn);
  2109. if (page_zone(page) != zone)
  2110. continue;
  2111. if (!swsusp_page_is_forbidden(page))
  2112. swsusp_unset_page_free(page);
  2113. }
  2114. for_each_migratetype_order(order, t) {
  2115. list_for_each_entry(page,
  2116. &zone->free_area[order].free_list[t], lru) {
  2117. unsigned long i;
  2118. pfn = page_to_pfn(page);
  2119. for (i = 0; i < (1UL << order); i++)
  2120. swsusp_set_page_free(pfn_to_page(pfn + i));
  2121. }
  2122. }
  2123. spin_unlock_irqrestore(&zone->lock, flags);
  2124. }
  2125. #endif /* CONFIG_PM */
  2126. /*
  2127. * Free a 0-order page
  2128. * cold == true ? free a cold page : free a hot page
  2129. */
  2130. void free_hot_cold_page(struct page *page, bool cold)
  2131. {
  2132. struct zone *zone = page_zone(page);
  2133. struct per_cpu_pages *pcp;
  2134. unsigned long pfn = page_to_pfn(page);
  2135. int migratetype;
  2136. if (in_interrupt()) {
  2137. __free_pages_ok(page, 0);
  2138. return;
  2139. }
  2140. if (!free_pcp_prepare(page))
  2141. return;
  2142. migratetype = get_pfnblock_migratetype(page, pfn);
  2143. set_pcppage_migratetype(page, migratetype);
  2144. preempt_disable();
  2145. /*
  2146. * We only track unmovable, reclaimable and movable on pcp lists.
  2147. * Free ISOLATE pages back to the allocator because they are being
  2148. * offlined but treat RESERVE as movable pages so we can get those
  2149. * areas back if necessary. Otherwise, we may have to free
  2150. * excessively into the page allocator
  2151. */
  2152. if (migratetype >= MIGRATE_PCPTYPES) {
  2153. if (unlikely(is_migrate_isolate(migratetype))) {
  2154. free_one_page(zone, page, pfn, 0, migratetype);
  2155. goto out;
  2156. }
  2157. migratetype = MIGRATE_MOVABLE;
  2158. }
  2159. __count_vm_event(PGFREE);
  2160. pcp = &this_cpu_ptr(zone->pageset)->pcp;
  2161. if (!cold)
  2162. list_add(&page->lru, &pcp->lists[migratetype]);
  2163. else
  2164. list_add_tail(&page->lru, &pcp->lists[migratetype]);
  2165. pcp->count++;
  2166. if (pcp->count >= pcp->high) {
  2167. unsigned long batch = READ_ONCE(pcp->batch);
  2168. free_pcppages_bulk(zone, batch, pcp);
  2169. pcp->count -= batch;
  2170. }
  2171. out:
  2172. preempt_enable();
  2173. }
  2174. /*
  2175. * Free a list of 0-order pages
  2176. */
  2177. void free_hot_cold_page_list(struct list_head *list, bool cold)
  2178. {
  2179. struct page *page, *next;
  2180. list_for_each_entry_safe(page, next, list, lru) {
  2181. trace_mm_page_free_batched(page, cold);
  2182. free_hot_cold_page(page, cold);
  2183. }
  2184. }
  2185. /*
  2186. * split_page takes a non-compound higher-order page, and splits it into
  2187. * n (1<<order) sub-pages: page[0..n]
  2188. * Each sub-page must be freed individually.
  2189. *
  2190. * Note: this is probably too low level an operation for use in drivers.
  2191. * Please consult with lkml before using this in your driver.
  2192. */
  2193. void split_page(struct page *page, unsigned int order)
  2194. {
  2195. int i;
  2196. VM_BUG_ON_PAGE(PageCompound(page), page);
  2197. VM_BUG_ON_PAGE(!page_count(page), page);
  2198. #ifdef CONFIG_KMEMCHECK
  2199. /*
  2200. * Split shadow pages too, because free(page[0]) would
  2201. * otherwise free the whole shadow.
  2202. */
  2203. if (kmemcheck_page_is_tracked(page))
  2204. split_page(virt_to_page(page[0].shadow), order);
  2205. #endif
  2206. for (i = 1; i < (1 << order); i++)
  2207. set_page_refcounted(page + i);
  2208. split_page_owner(page, order);
  2209. }
  2210. EXPORT_SYMBOL_GPL(split_page);
  2211. int __isolate_free_page(struct page *page, unsigned int order)
  2212. {
  2213. unsigned long watermark;
  2214. struct zone *zone;
  2215. int mt;
  2216. BUG_ON(!PageBuddy(page));
  2217. zone = page_zone(page);
  2218. mt = get_pageblock_migratetype(page);
  2219. if (!is_migrate_isolate(mt)) {
  2220. /*
  2221. * Obey watermarks as if the page was being allocated. We can
  2222. * emulate a high-order watermark check with a raised order-0
  2223. * watermark, because we already know our high-order page
  2224. * exists.
  2225. */
  2226. watermark = min_wmark_pages(zone) + (1UL << order);
  2227. if (!zone_watermark_ok(zone, 0, watermark, 0, ALLOC_CMA))
  2228. return 0;
  2229. __mod_zone_freepage_state(zone, -(1UL << order), mt);
  2230. }
  2231. /* Remove page from free list */
  2232. list_del(&page->lru);
  2233. zone->free_area[order].nr_free--;
  2234. rmv_page_order(page);
  2235. /*
  2236. * Set the pageblock if the isolated page is at least half of a
  2237. * pageblock
  2238. */
  2239. if (order >= pageblock_order - 1) {
  2240. struct page *endpage = page + (1 << order) - 1;
  2241. for (; page < endpage; page += pageblock_nr_pages) {
  2242. int mt = get_pageblock_migratetype(page);
  2243. if (!is_migrate_isolate(mt) && !is_migrate_cma(mt)
  2244. && mt != MIGRATE_HIGHATOMIC)
  2245. set_pageblock_migratetype(page,
  2246. MIGRATE_MOVABLE);
  2247. }
  2248. }
  2249. return 1UL << order;
  2250. }
  2251. /*
  2252. * Update NUMA hit/miss statistics
  2253. *
  2254. * Must be called with interrupts disabled.
  2255. */
  2256. static inline void zone_statistics(struct zone *preferred_zone, struct zone *z)
  2257. {
  2258. #ifdef CONFIG_NUMA
  2259. enum zone_stat_item local_stat = NUMA_LOCAL;
  2260. if (z->node != numa_node_id())
  2261. local_stat = NUMA_OTHER;
  2262. if (z->node == preferred_zone->node)
  2263. __inc_zone_state(z, NUMA_HIT);
  2264. else {
  2265. __inc_zone_state(z, NUMA_MISS);
  2266. __inc_zone_state(preferred_zone, NUMA_FOREIGN);
  2267. }
  2268. __inc_zone_state(z, local_stat);
  2269. #endif
  2270. }
  2271. /* Remove page from the per-cpu list, caller must protect the list */
  2272. static struct page *__rmqueue_pcplist(struct zone *zone, int migratetype,
  2273. bool cold, struct per_cpu_pages *pcp,
  2274. struct list_head *list)
  2275. {
  2276. struct page *page;
  2277. VM_BUG_ON(in_interrupt());
  2278. do {
  2279. if (list_empty(list)) {
  2280. pcp->count += rmqueue_bulk(zone, 0,
  2281. pcp->batch, list,
  2282. migratetype, cold);
  2283. if (unlikely(list_empty(list)))
  2284. return NULL;
  2285. }
  2286. if (cold)
  2287. page = list_last_entry(list, struct page, lru);
  2288. else
  2289. page = list_first_entry(list, struct page, lru);
  2290. list_del(&page->lru);
  2291. pcp->count--;
  2292. } while (check_new_pcp(page));
  2293. return page;
  2294. }
  2295. /* Lock and remove page from the per-cpu list */
  2296. static struct page *rmqueue_pcplist(struct zone *preferred_zone,
  2297. struct zone *zone, unsigned int order,
  2298. gfp_t gfp_flags, int migratetype)
  2299. {
  2300. struct per_cpu_pages *pcp;
  2301. struct list_head *list;
  2302. bool cold = ((gfp_flags & __GFP_COLD) != 0);
  2303. struct page *page;
  2304. preempt_disable();
  2305. pcp = &this_cpu_ptr(zone->pageset)->pcp;
  2306. list = &pcp->lists[migratetype];
  2307. page = __rmqueue_pcplist(zone, migratetype, cold, pcp, list);
  2308. if (page) {
  2309. __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order);
  2310. zone_statistics(preferred_zone, zone);
  2311. }
  2312. preempt_enable();
  2313. return page;
  2314. }
  2315. /*
  2316. * Allocate a page from the given zone. Use pcplists for order-0 allocations.
  2317. */
  2318. static inline
  2319. struct page *rmqueue(struct zone *preferred_zone,
  2320. struct zone *zone, unsigned int order,
  2321. gfp_t gfp_flags, unsigned int alloc_flags,
  2322. int migratetype)
  2323. {
  2324. unsigned long flags;
  2325. struct page *page;
  2326. if (likely(order == 0) && !in_interrupt()) {
  2327. page = rmqueue_pcplist(preferred_zone, zone, order,
  2328. gfp_flags, migratetype);
  2329. goto out;
  2330. }
  2331. /*
  2332. * We most definitely don't want callers attempting to
  2333. * allocate greater than order-1 page units with __GFP_NOFAIL.
  2334. */
  2335. WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1));
  2336. spin_lock_irqsave(&zone->lock, flags);
  2337. do {
  2338. page = NULL;
  2339. if (alloc_flags & ALLOC_HARDER) {
  2340. page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC);
  2341. if (page)
  2342. trace_mm_page_alloc_zone_locked(page, order, migratetype);
  2343. }
  2344. if (!page)
  2345. page = __rmqueue(zone, order, migratetype);
  2346. } while (page && check_new_pages(page, order));
  2347. spin_unlock(&zone->lock);
  2348. if (!page)
  2349. goto failed;
  2350. __mod_zone_freepage_state(zone, -(1 << order),
  2351. get_pcppage_migratetype(page));
  2352. __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order);
  2353. zone_statistics(preferred_zone, zone);
  2354. local_irq_restore(flags);
  2355. out:
  2356. VM_BUG_ON_PAGE(page && bad_range(zone, page), page);
  2357. return page;
  2358. failed:
  2359. local_irq_restore(flags);
  2360. return NULL;
  2361. }
  2362. #ifdef CONFIG_FAIL_PAGE_ALLOC
  2363. static struct {
  2364. struct fault_attr attr;
  2365. bool ignore_gfp_highmem;
  2366. bool ignore_gfp_reclaim;
  2367. u32 min_order;
  2368. } fail_page_alloc = {
  2369. .attr = FAULT_ATTR_INITIALIZER,
  2370. .ignore_gfp_reclaim = true,
  2371. .ignore_gfp_highmem = true,
  2372. .min_order = 1,
  2373. };
  2374. static int __init setup_fail_page_alloc(char *str)
  2375. {
  2376. return setup_fault_attr(&fail_page_alloc.attr, str);
  2377. }
  2378. __setup("fail_page_alloc=", setup_fail_page_alloc);
  2379. static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
  2380. {
  2381. if (order < fail_page_alloc.min_order)
  2382. return false;
  2383. if (gfp_mask & __GFP_NOFAIL)
  2384. return false;
  2385. if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
  2386. return false;
  2387. if (fail_page_alloc.ignore_gfp_reclaim &&
  2388. (gfp_mask & __GFP_DIRECT_RECLAIM))
  2389. return false;
  2390. return should_fail(&fail_page_alloc.attr, 1 << order);
  2391. }
  2392. #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
  2393. static int __init fail_page_alloc_debugfs(void)
  2394. {
  2395. umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
  2396. struct dentry *dir;
  2397. dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
  2398. &fail_page_alloc.attr);
  2399. if (IS_ERR(dir))
  2400. return PTR_ERR(dir);
  2401. if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
  2402. &fail_page_alloc.ignore_gfp_reclaim))
  2403. goto fail;
  2404. if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
  2405. &fail_page_alloc.ignore_gfp_highmem))
  2406. goto fail;
  2407. if (!debugfs_create_u32("min-order", mode, dir,
  2408. &fail_page_alloc.min_order))
  2409. goto fail;
  2410. return 0;
  2411. fail:
  2412. debugfs_remove_recursive(dir);
  2413. return -ENOMEM;
  2414. }
  2415. late_initcall(fail_page_alloc_debugfs);
  2416. #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
  2417. #else /* CONFIG_FAIL_PAGE_ALLOC */
  2418. static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
  2419. {
  2420. return false;
  2421. }
  2422. #endif /* CONFIG_FAIL_PAGE_ALLOC */
  2423. /*
  2424. * Return true if free base pages are above 'mark'. For high-order checks it
  2425. * will return true of the order-0 watermark is reached and there is at least
  2426. * one free page of a suitable size. Checking now avoids taking the zone lock
  2427. * to check in the allocation paths if no pages are free.
  2428. */
  2429. bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
  2430. int classzone_idx, unsigned int alloc_flags,
  2431. long free_pages)
  2432. {
  2433. long min = mark;
  2434. int o;
  2435. const bool alloc_harder = (alloc_flags & ALLOC_HARDER);
  2436. /* free_pages may go negative - that's OK */
  2437. free_pages -= (1 << order) - 1;
  2438. if (alloc_flags & ALLOC_HIGH)
  2439. min -= min / 2;
  2440. /*
  2441. * If the caller does not have rights to ALLOC_HARDER then subtract
  2442. * the high-atomic reserves. This will over-estimate the size of the
  2443. * atomic reserve but it avoids a search.
  2444. */
  2445. if (likely(!alloc_harder))
  2446. free_pages -= z->nr_reserved_highatomic;
  2447. else
  2448. min -= min / 4;
  2449. #ifdef CONFIG_CMA
  2450. /* If allocation can't use CMA areas don't use free CMA pages */
  2451. if (!(alloc_flags & ALLOC_CMA))
  2452. free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES);
  2453. #endif
  2454. /*
  2455. * Check watermarks for an order-0 allocation request. If these
  2456. * are not met, then a high-order request also cannot go ahead
  2457. * even if a suitable page happened to be free.
  2458. */
  2459. if (free_pages <= min + z->lowmem_reserve[classzone_idx])
  2460. return false;
  2461. /* If this is an order-0 request then the watermark is fine */
  2462. if (!order)
  2463. return true;
  2464. /* For a high-order request, check at least one suitable page is free */
  2465. for (o = order; o < MAX_ORDER; o++) {
  2466. struct free_area *area = &z->free_area[o];
  2467. int mt;
  2468. if (!area->nr_free)
  2469. continue;
  2470. if (alloc_harder)
  2471. return true;
  2472. for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) {
  2473. if (!list_empty(&area->free_list[mt]))
  2474. return true;
  2475. }
  2476. #ifdef CONFIG_CMA
  2477. if ((alloc_flags & ALLOC_CMA) &&
  2478. !list_empty(&area->free_list[MIGRATE_CMA])) {
  2479. return true;
  2480. }
  2481. #endif
  2482. }
  2483. return false;
  2484. }
  2485. bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
  2486. int classzone_idx, unsigned int alloc_flags)
  2487. {
  2488. return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
  2489. zone_page_state(z, NR_FREE_PAGES));
  2490. }
  2491. static inline bool zone_watermark_fast(struct zone *z, unsigned int order,
  2492. unsigned long mark, int classzone_idx, unsigned int alloc_flags)
  2493. {
  2494. long free_pages = zone_page_state(z, NR_FREE_PAGES);
  2495. long cma_pages = 0;
  2496. #ifdef CONFIG_CMA
  2497. /* If allocation can't use CMA areas don't use free CMA pages */
  2498. if (!(alloc_flags & ALLOC_CMA))
  2499. cma_pages = zone_page_state(z, NR_FREE_CMA_PAGES);
  2500. #endif
  2501. /*
  2502. * Fast check for order-0 only. If this fails then the reserves
  2503. * need to be calculated. There is a corner case where the check
  2504. * passes but only the high-order atomic reserve are free. If
  2505. * the caller is !atomic then it'll uselessly search the free
  2506. * list. That corner case is then slower but it is harmless.
  2507. */
  2508. if (!order && (free_pages - cma_pages) > mark + z->lowmem_reserve[classzone_idx])
  2509. return true;
  2510. return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
  2511. free_pages);
  2512. }
  2513. bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
  2514. unsigned long mark, int classzone_idx)
  2515. {
  2516. long free_pages = zone_page_state(z, NR_FREE_PAGES);
  2517. if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
  2518. free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
  2519. return __zone_watermark_ok(z, order, mark, classzone_idx, 0,
  2520. free_pages);
  2521. }
  2522. #ifdef CONFIG_NUMA
  2523. static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
  2524. {
  2525. return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <=
  2526. RECLAIM_DISTANCE;
  2527. }
  2528. #else /* CONFIG_NUMA */
  2529. static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
  2530. {
  2531. return true;
  2532. }
  2533. #endif /* CONFIG_NUMA */
  2534. /*
  2535. * get_page_from_freelist goes through the zonelist trying to allocate
  2536. * a page.
  2537. */
  2538. static struct page *
  2539. get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags,
  2540. const struct alloc_context *ac)
  2541. {
  2542. struct zoneref *z = ac->preferred_zoneref;
  2543. struct zone *zone;
  2544. struct pglist_data *last_pgdat_dirty_limit = NULL;
  2545. /*
  2546. * Scan zonelist, looking for a zone with enough free.
  2547. * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
  2548. */
  2549. for_next_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
  2550. ac->nodemask) {
  2551. struct page *page;
  2552. unsigned long mark;
  2553. if (cpusets_enabled() &&
  2554. (alloc_flags & ALLOC_CPUSET) &&
  2555. !__cpuset_zone_allowed(zone, gfp_mask))
  2556. continue;
  2557. /*
  2558. * When allocating a page cache page for writing, we
  2559. * want to get it from a node that is within its dirty
  2560. * limit, such that no single node holds more than its
  2561. * proportional share of globally allowed dirty pages.
  2562. * The dirty limits take into account the node's
  2563. * lowmem reserves and high watermark so that kswapd
  2564. * should be able to balance it without having to
  2565. * write pages from its LRU list.
  2566. *
  2567. * XXX: For now, allow allocations to potentially
  2568. * exceed the per-node dirty limit in the slowpath
  2569. * (spread_dirty_pages unset) before going into reclaim,
  2570. * which is important when on a NUMA setup the allowed
  2571. * nodes are together not big enough to reach the
  2572. * global limit. The proper fix for these situations
  2573. * will require awareness of nodes in the
  2574. * dirty-throttling and the flusher threads.
  2575. */
  2576. if (ac->spread_dirty_pages) {
  2577. if (last_pgdat_dirty_limit == zone->zone_pgdat)
  2578. continue;
  2579. if (!node_dirty_ok(zone->zone_pgdat)) {
  2580. last_pgdat_dirty_limit = zone->zone_pgdat;
  2581. continue;
  2582. }
  2583. }
  2584. mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
  2585. if (!zone_watermark_fast(zone, order, mark,
  2586. ac_classzone_idx(ac), alloc_flags)) {
  2587. int ret;
  2588. /* Checked here to keep the fast path fast */
  2589. BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
  2590. if (alloc_flags & ALLOC_NO_WATERMARKS)
  2591. goto try_this_zone;
  2592. if (node_reclaim_mode == 0 ||
  2593. !zone_allows_reclaim(ac->preferred_zoneref->zone, zone))
  2594. continue;
  2595. ret = node_reclaim(zone->zone_pgdat, gfp_mask, order);
  2596. switch (ret) {
  2597. case NODE_RECLAIM_NOSCAN:
  2598. /* did not scan */
  2599. continue;
  2600. case NODE_RECLAIM_FULL:
  2601. /* scanned but unreclaimable */
  2602. continue;
  2603. default:
  2604. /* did we reclaim enough */
  2605. if (zone_watermark_ok(zone, order, mark,
  2606. ac_classzone_idx(ac), alloc_flags))
  2607. goto try_this_zone;
  2608. continue;
  2609. }
  2610. }
  2611. try_this_zone:
  2612. page = rmqueue(ac->preferred_zoneref->zone, zone, order,
  2613. gfp_mask, alloc_flags, ac->migratetype);
  2614. if (page) {
  2615. prep_new_page(page, order, gfp_mask, alloc_flags);
  2616. /*
  2617. * If this is a high-order atomic allocation then check
  2618. * if the pageblock should be reserved for the future
  2619. */
  2620. if (unlikely(order && (alloc_flags & ALLOC_HARDER)))
  2621. reserve_highatomic_pageblock(page, zone, order);
  2622. return page;
  2623. }
  2624. }
  2625. return NULL;
  2626. }
  2627. /*
  2628. * Large machines with many possible nodes should not always dump per-node
  2629. * meminfo in irq context.
  2630. */
  2631. static inline bool should_suppress_show_mem(void)
  2632. {
  2633. bool ret = false;
  2634. #if NODES_SHIFT > 8
  2635. ret = in_interrupt();
  2636. #endif
  2637. return ret;
  2638. }
  2639. static void warn_alloc_show_mem(gfp_t gfp_mask, nodemask_t *nodemask)
  2640. {
  2641. unsigned int filter = SHOW_MEM_FILTER_NODES;
  2642. static DEFINE_RATELIMIT_STATE(show_mem_rs, HZ, 1);
  2643. if (should_suppress_show_mem() || !__ratelimit(&show_mem_rs))
  2644. return;
  2645. /*
  2646. * This documents exceptions given to allocations in certain
  2647. * contexts that are allowed to allocate outside current's set
  2648. * of allowed nodes.
  2649. */
  2650. if (!(gfp_mask & __GFP_NOMEMALLOC))
  2651. if (test_thread_flag(TIF_MEMDIE) ||
  2652. (current->flags & (PF_MEMALLOC | PF_EXITING)))
  2653. filter &= ~SHOW_MEM_FILTER_NODES;
  2654. if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM))
  2655. filter &= ~SHOW_MEM_FILTER_NODES;
  2656. show_mem(filter, nodemask);
  2657. }
  2658. void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...)
  2659. {
  2660. struct va_format vaf;
  2661. va_list args;
  2662. static DEFINE_RATELIMIT_STATE(nopage_rs, DEFAULT_RATELIMIT_INTERVAL,
  2663. DEFAULT_RATELIMIT_BURST);
  2664. if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) ||
  2665. debug_guardpage_minorder() > 0)
  2666. return;
  2667. pr_warn("%s: ", current->comm);
  2668. va_start(args, fmt);
  2669. vaf.fmt = fmt;
  2670. vaf.va = &args;
  2671. pr_cont("%pV", &vaf);
  2672. va_end(args);
  2673. pr_cont(", mode:%#x(%pGg), nodemask=", gfp_mask, &gfp_mask);
  2674. if (nodemask)
  2675. pr_cont("%*pbl\n", nodemask_pr_args(nodemask));
  2676. else
  2677. pr_cont("(null)\n");
  2678. cpuset_print_current_mems_allowed();
  2679. dump_stack();
  2680. warn_alloc_show_mem(gfp_mask, nodemask);
  2681. }
  2682. static inline struct page *
  2683. __alloc_pages_cpuset_fallback(gfp_t gfp_mask, unsigned int order,
  2684. unsigned int alloc_flags,
  2685. const struct alloc_context *ac)
  2686. {
  2687. struct page *page;
  2688. page = get_page_from_freelist(gfp_mask, order,
  2689. alloc_flags|ALLOC_CPUSET, ac);
  2690. /*
  2691. * fallback to ignore cpuset restriction if our nodes
  2692. * are depleted
  2693. */
  2694. if (!page)
  2695. page = get_page_from_freelist(gfp_mask, order,
  2696. alloc_flags, ac);
  2697. return page;
  2698. }
  2699. static inline struct page *
  2700. __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
  2701. const struct alloc_context *ac, unsigned long *did_some_progress)
  2702. {
  2703. struct oom_control oc = {
  2704. .zonelist = ac->zonelist,
  2705. .nodemask = ac->nodemask,
  2706. .memcg = NULL,
  2707. .gfp_mask = gfp_mask,
  2708. .order = order,
  2709. };
  2710. struct page *page;
  2711. *did_some_progress = 0;
  2712. /*
  2713. * Acquire the oom lock. If that fails, somebody else is
  2714. * making progress for us.
  2715. */
  2716. if (!mutex_trylock(&oom_lock)) {
  2717. *did_some_progress = 1;
  2718. schedule_timeout_uninterruptible(1);
  2719. return NULL;
  2720. }
  2721. /*
  2722. * Go through the zonelist yet one more time, keep very high watermark
  2723. * here, this is only to catch a parallel oom killing, we must fail if
  2724. * we're still under heavy pressure.
  2725. */
  2726. page = get_page_from_freelist(gfp_mask | __GFP_HARDWALL, order,
  2727. ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac);
  2728. if (page)
  2729. goto out;
  2730. /* Coredumps can quickly deplete all memory reserves */
  2731. if (current->flags & PF_DUMPCORE)
  2732. goto out;
  2733. /* The OOM killer will not help higher order allocs */
  2734. if (order > PAGE_ALLOC_COSTLY_ORDER)
  2735. goto out;
  2736. /* The OOM killer does not needlessly kill tasks for lowmem */
  2737. if (ac->high_zoneidx < ZONE_NORMAL)
  2738. goto out;
  2739. if (pm_suspended_storage())
  2740. goto out;
  2741. /*
  2742. * XXX: GFP_NOFS allocations should rather fail than rely on
  2743. * other request to make a forward progress.
  2744. * We are in an unfortunate situation where out_of_memory cannot
  2745. * do much for this context but let's try it to at least get
  2746. * access to memory reserved if the current task is killed (see
  2747. * out_of_memory). Once filesystems are ready to handle allocation
  2748. * failures more gracefully we should just bail out here.
  2749. */
  2750. /* The OOM killer may not free memory on a specific node */
  2751. if (gfp_mask & __GFP_THISNODE)
  2752. goto out;
  2753. /* Exhausted what can be done so it's blamo time */
  2754. if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) {
  2755. *did_some_progress = 1;
  2756. /*
  2757. * Help non-failing allocations by giving them access to memory
  2758. * reserves
  2759. */
  2760. if (gfp_mask & __GFP_NOFAIL)
  2761. page = __alloc_pages_cpuset_fallback(gfp_mask, order,
  2762. ALLOC_NO_WATERMARKS, ac);
  2763. }
  2764. out:
  2765. mutex_unlock(&oom_lock);
  2766. return page;
  2767. }
  2768. /*
  2769. * Maximum number of compaction retries wit a progress before OOM
  2770. * killer is consider as the only way to move forward.
  2771. */
  2772. #define MAX_COMPACT_RETRIES 16
  2773. #ifdef CONFIG_COMPACTION
  2774. /* Try memory compaction for high-order allocations before reclaim */
  2775. static struct page *
  2776. __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
  2777. unsigned int alloc_flags, const struct alloc_context *ac,
  2778. enum compact_priority prio, enum compact_result *compact_result)
  2779. {
  2780. struct page *page;
  2781. if (!order)
  2782. return NULL;
  2783. current->flags |= PF_MEMALLOC;
  2784. *compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac,
  2785. prio);
  2786. current->flags &= ~PF_MEMALLOC;
  2787. if (*compact_result <= COMPACT_INACTIVE)
  2788. return NULL;
  2789. /*
  2790. * At least in one zone compaction wasn't deferred or skipped, so let's
  2791. * count a compaction stall
  2792. */
  2793. count_vm_event(COMPACTSTALL);
  2794. page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
  2795. if (page) {
  2796. struct zone *zone = page_zone(page);
  2797. zone->compact_blockskip_flush = false;
  2798. compaction_defer_reset(zone, order, true);
  2799. count_vm_event(COMPACTSUCCESS);
  2800. return page;
  2801. }
  2802. /*
  2803. * It's bad if compaction run occurs and fails. The most likely reason
  2804. * is that pages exist, but not enough to satisfy watermarks.
  2805. */
  2806. count_vm_event(COMPACTFAIL);
  2807. cond_resched();
  2808. return NULL;
  2809. }
  2810. static inline bool
  2811. should_compact_retry(struct alloc_context *ac, int order, int alloc_flags,
  2812. enum compact_result compact_result,
  2813. enum compact_priority *compact_priority,
  2814. int *compaction_retries)
  2815. {
  2816. int max_retries = MAX_COMPACT_RETRIES;
  2817. int min_priority;
  2818. bool ret = false;
  2819. int retries = *compaction_retries;
  2820. enum compact_priority priority = *compact_priority;
  2821. if (!order)
  2822. return false;
  2823. if (compaction_made_progress(compact_result))
  2824. (*compaction_retries)++;
  2825. /*
  2826. * compaction considers all the zone as desperately out of memory
  2827. * so it doesn't really make much sense to retry except when the
  2828. * failure could be caused by insufficient priority
  2829. */
  2830. if (compaction_failed(compact_result))
  2831. goto check_priority;
  2832. /*
  2833. * make sure the compaction wasn't deferred or didn't bail out early
  2834. * due to locks contention before we declare that we should give up.
  2835. * But do not retry if the given zonelist is not suitable for
  2836. * compaction.
  2837. */
  2838. if (compaction_withdrawn(compact_result)) {
  2839. ret = compaction_zonelist_suitable(ac, order, alloc_flags);
  2840. goto out;
  2841. }
  2842. /*
  2843. * !costly requests are much more important than __GFP_REPEAT
  2844. * costly ones because they are de facto nofail and invoke OOM
  2845. * killer to move on while costly can fail and users are ready
  2846. * to cope with that. 1/4 retries is rather arbitrary but we
  2847. * would need much more detailed feedback from compaction to
  2848. * make a better decision.
  2849. */
  2850. if (order > PAGE_ALLOC_COSTLY_ORDER)
  2851. max_retries /= 4;
  2852. if (*compaction_retries <= max_retries) {
  2853. ret = true;
  2854. goto out;
  2855. }
  2856. /*
  2857. * Make sure there are attempts at the highest priority if we exhausted
  2858. * all retries or failed at the lower priorities.
  2859. */
  2860. check_priority:
  2861. min_priority = (order > PAGE_ALLOC_COSTLY_ORDER) ?
  2862. MIN_COMPACT_COSTLY_PRIORITY : MIN_COMPACT_PRIORITY;
  2863. if (*compact_priority > min_priority) {
  2864. (*compact_priority)--;
  2865. *compaction_retries = 0;
  2866. ret = true;
  2867. }
  2868. out:
  2869. trace_compact_retry(order, priority, compact_result, retries, max_retries, ret);
  2870. return ret;
  2871. }
  2872. #else
  2873. static inline struct page *
  2874. __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
  2875. unsigned int alloc_flags, const struct alloc_context *ac,
  2876. enum compact_priority prio, enum compact_result *compact_result)
  2877. {
  2878. *compact_result = COMPACT_SKIPPED;
  2879. return NULL;
  2880. }
  2881. static inline bool
  2882. should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags,
  2883. enum compact_result compact_result,
  2884. enum compact_priority *compact_priority,
  2885. int *compaction_retries)
  2886. {
  2887. struct zone *zone;
  2888. struct zoneref *z;
  2889. if (!order || order > PAGE_ALLOC_COSTLY_ORDER)
  2890. return false;
  2891. /*
  2892. * There are setups with compaction disabled which would prefer to loop
  2893. * inside the allocator rather than hit the oom killer prematurely.
  2894. * Let's give them a good hope and keep retrying while the order-0
  2895. * watermarks are OK.
  2896. */
  2897. for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
  2898. ac->nodemask) {
  2899. if (zone_watermark_ok(zone, 0, min_wmark_pages(zone),
  2900. ac_classzone_idx(ac), alloc_flags))
  2901. return true;
  2902. }
  2903. return false;
  2904. }
  2905. #endif /* CONFIG_COMPACTION */
  2906. /* Perform direct synchronous page reclaim */
  2907. static int
  2908. __perform_reclaim(gfp_t gfp_mask, unsigned int order,
  2909. const struct alloc_context *ac)
  2910. {
  2911. struct reclaim_state reclaim_state;
  2912. int progress;
  2913. cond_resched();
  2914. /* We now go into synchronous reclaim */
  2915. cpuset_memory_pressure_bump();
  2916. current->flags |= PF_MEMALLOC;
  2917. lockdep_set_current_reclaim_state(gfp_mask);
  2918. reclaim_state.reclaimed_slab = 0;
  2919. current->reclaim_state = &reclaim_state;
  2920. progress = try_to_free_pages(ac->zonelist, order, gfp_mask,
  2921. ac->nodemask);
  2922. current->reclaim_state = NULL;
  2923. lockdep_clear_current_reclaim_state();
  2924. current->flags &= ~PF_MEMALLOC;
  2925. cond_resched();
  2926. return progress;
  2927. }
  2928. /* The really slow allocator path where we enter direct reclaim */
  2929. static inline struct page *
  2930. __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
  2931. unsigned int alloc_flags, const struct alloc_context *ac,
  2932. unsigned long *did_some_progress)
  2933. {
  2934. struct page *page = NULL;
  2935. bool drained = false;
  2936. *did_some_progress = __perform_reclaim(gfp_mask, order, ac);
  2937. if (unlikely(!(*did_some_progress)))
  2938. return NULL;
  2939. retry:
  2940. page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
  2941. /*
  2942. * If an allocation failed after direct reclaim, it could be because
  2943. * pages are pinned on the per-cpu lists or in high alloc reserves.
  2944. * Shrink them them and try again
  2945. */
  2946. if (!page && !drained) {
  2947. unreserve_highatomic_pageblock(ac, false);
  2948. drain_all_pages(NULL);
  2949. drained = true;
  2950. goto retry;
  2951. }
  2952. return page;
  2953. }
  2954. static void wake_all_kswapds(unsigned int order, const struct alloc_context *ac)
  2955. {
  2956. struct zoneref *z;
  2957. struct zone *zone;
  2958. pg_data_t *last_pgdat = NULL;
  2959. for_each_zone_zonelist_nodemask(zone, z, ac->zonelist,
  2960. ac->high_zoneidx, ac->nodemask) {
  2961. if (last_pgdat != zone->zone_pgdat)
  2962. wakeup_kswapd(zone, order, ac->high_zoneidx);
  2963. last_pgdat = zone->zone_pgdat;
  2964. }
  2965. }
  2966. static inline unsigned int
  2967. gfp_to_alloc_flags(gfp_t gfp_mask)
  2968. {
  2969. unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
  2970. /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
  2971. BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
  2972. /*
  2973. * The caller may dip into page reserves a bit more if the caller
  2974. * cannot run direct reclaim, or if the caller has realtime scheduling
  2975. * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
  2976. * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH).
  2977. */
  2978. alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
  2979. if (gfp_mask & __GFP_ATOMIC) {
  2980. /*
  2981. * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
  2982. * if it can't schedule.
  2983. */
  2984. if (!(gfp_mask & __GFP_NOMEMALLOC))
  2985. alloc_flags |= ALLOC_HARDER;
  2986. /*
  2987. * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
  2988. * comment for __cpuset_node_allowed().
  2989. */
  2990. alloc_flags &= ~ALLOC_CPUSET;
  2991. } else if (unlikely(rt_task(current)) && !in_interrupt())
  2992. alloc_flags |= ALLOC_HARDER;
  2993. #ifdef CONFIG_CMA
  2994. if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
  2995. alloc_flags |= ALLOC_CMA;
  2996. #endif
  2997. return alloc_flags;
  2998. }
  2999. bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
  3000. {
  3001. if (unlikely(gfp_mask & __GFP_NOMEMALLOC))
  3002. return false;
  3003. if (gfp_mask & __GFP_MEMALLOC)
  3004. return true;
  3005. if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
  3006. return true;
  3007. if (!in_interrupt() &&
  3008. ((current->flags & PF_MEMALLOC) ||
  3009. unlikely(test_thread_flag(TIF_MEMDIE))))
  3010. return true;
  3011. return false;
  3012. }
  3013. /*
  3014. * Maximum number of reclaim retries without any progress before OOM killer
  3015. * is consider as the only way to move forward.
  3016. */
  3017. #define MAX_RECLAIM_RETRIES 16
  3018. /*
  3019. * Checks whether it makes sense to retry the reclaim to make a forward progress
  3020. * for the given allocation request.
  3021. * The reclaim feedback represented by did_some_progress (any progress during
  3022. * the last reclaim round) and no_progress_loops (number of reclaim rounds without
  3023. * any progress in a row) is considered as well as the reclaimable pages on the
  3024. * applicable zone list (with a backoff mechanism which is a function of
  3025. * no_progress_loops).
  3026. *
  3027. * Returns true if a retry is viable or false to enter the oom path.
  3028. */
  3029. static inline bool
  3030. should_reclaim_retry(gfp_t gfp_mask, unsigned order,
  3031. struct alloc_context *ac, int alloc_flags,
  3032. bool did_some_progress, int *no_progress_loops)
  3033. {
  3034. struct zone *zone;
  3035. struct zoneref *z;
  3036. /*
  3037. * Costly allocations might have made a progress but this doesn't mean
  3038. * their order will become available due to high fragmentation so
  3039. * always increment the no progress counter for them
  3040. */
  3041. if (did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER)
  3042. *no_progress_loops = 0;
  3043. else
  3044. (*no_progress_loops)++;
  3045. /*
  3046. * Make sure we converge to OOM if we cannot make any progress
  3047. * several times in the row.
  3048. */
  3049. if (*no_progress_loops > MAX_RECLAIM_RETRIES) {
  3050. /* Before OOM, exhaust highatomic_reserve */
  3051. return unreserve_highatomic_pageblock(ac, true);
  3052. }
  3053. /*
  3054. * Keep reclaiming pages while there is a chance this will lead
  3055. * somewhere. If none of the target zones can satisfy our allocation
  3056. * request even if all reclaimable pages are considered then we are
  3057. * screwed and have to go OOM.
  3058. */
  3059. for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
  3060. ac->nodemask) {
  3061. unsigned long available;
  3062. unsigned long reclaimable;
  3063. unsigned long min_wmark = min_wmark_pages(zone);
  3064. bool wmark;
  3065. available = reclaimable = zone_reclaimable_pages(zone);
  3066. available -= DIV_ROUND_UP((*no_progress_loops) * available,
  3067. MAX_RECLAIM_RETRIES);
  3068. available += zone_page_state_snapshot(zone, NR_FREE_PAGES);
  3069. /*
  3070. * Would the allocation succeed if we reclaimed the whole
  3071. * available?
  3072. */
  3073. wmark = __zone_watermark_ok(zone, order, min_wmark,
  3074. ac_classzone_idx(ac), alloc_flags, available);
  3075. trace_reclaim_retry_zone(z, order, reclaimable,
  3076. available, min_wmark, *no_progress_loops, wmark);
  3077. if (wmark) {
  3078. /*
  3079. * If we didn't make any progress and have a lot of
  3080. * dirty + writeback pages then we should wait for
  3081. * an IO to complete to slow down the reclaim and
  3082. * prevent from pre mature OOM
  3083. */
  3084. if (!did_some_progress) {
  3085. unsigned long write_pending;
  3086. write_pending = zone_page_state_snapshot(zone,
  3087. NR_ZONE_WRITE_PENDING);
  3088. if (2 * write_pending > reclaimable) {
  3089. congestion_wait(BLK_RW_ASYNC, HZ/10);
  3090. return true;
  3091. }
  3092. }
  3093. /*
  3094. * Memory allocation/reclaim might be called from a WQ
  3095. * context and the current implementation of the WQ
  3096. * concurrency control doesn't recognize that
  3097. * a particular WQ is congested if the worker thread is
  3098. * looping without ever sleeping. Therefore we have to
  3099. * do a short sleep here rather than calling
  3100. * cond_resched().
  3101. */
  3102. if (current->flags & PF_WQ_WORKER)
  3103. schedule_timeout_uninterruptible(1);
  3104. else
  3105. cond_resched();
  3106. return true;
  3107. }
  3108. }
  3109. return false;
  3110. }
  3111. static inline struct page *
  3112. __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
  3113. struct alloc_context *ac)
  3114. {
  3115. bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM;
  3116. struct page *page = NULL;
  3117. unsigned int alloc_flags;
  3118. unsigned long did_some_progress;
  3119. enum compact_priority compact_priority;
  3120. enum compact_result compact_result;
  3121. int compaction_retries;
  3122. int no_progress_loops;
  3123. unsigned long alloc_start = jiffies;
  3124. unsigned int stall_timeout = 10 * HZ;
  3125. unsigned int cpuset_mems_cookie;
  3126. /*
  3127. * In the slowpath, we sanity check order to avoid ever trying to
  3128. * reclaim >= MAX_ORDER areas which will never succeed. Callers may
  3129. * be using allocators in order of preference for an area that is
  3130. * too large.
  3131. */
  3132. if (order >= MAX_ORDER) {
  3133. WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
  3134. return NULL;
  3135. }
  3136. /*
  3137. * We also sanity check to catch abuse of atomic reserves being used by
  3138. * callers that are not in atomic context.
  3139. */
  3140. if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) ==
  3141. (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)))
  3142. gfp_mask &= ~__GFP_ATOMIC;
  3143. retry_cpuset:
  3144. compaction_retries = 0;
  3145. no_progress_loops = 0;
  3146. compact_priority = DEF_COMPACT_PRIORITY;
  3147. cpuset_mems_cookie = read_mems_allowed_begin();
  3148. /*
  3149. * The fast path uses conservative alloc_flags to succeed only until
  3150. * kswapd needs to be woken up, and to avoid the cost of setting up
  3151. * alloc_flags precisely. So we do that now.
  3152. */
  3153. alloc_flags = gfp_to_alloc_flags(gfp_mask);
  3154. /*
  3155. * We need to recalculate the starting point for the zonelist iterator
  3156. * because we might have used different nodemask in the fast path, or
  3157. * there was a cpuset modification and we are retrying - otherwise we
  3158. * could end up iterating over non-eligible zones endlessly.
  3159. */
  3160. ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
  3161. ac->high_zoneidx, ac->nodemask);
  3162. if (!ac->preferred_zoneref->zone)
  3163. goto nopage;
  3164. if (gfp_mask & __GFP_KSWAPD_RECLAIM)
  3165. wake_all_kswapds(order, ac);
  3166. /*
  3167. * The adjusted alloc_flags might result in immediate success, so try
  3168. * that first
  3169. */
  3170. page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
  3171. if (page)
  3172. goto got_pg;
  3173. /*
  3174. * For costly allocations, try direct compaction first, as it's likely
  3175. * that we have enough base pages and don't need to reclaim. Don't try
  3176. * that for allocations that are allowed to ignore watermarks, as the
  3177. * ALLOC_NO_WATERMARKS attempt didn't yet happen.
  3178. */
  3179. if (can_direct_reclaim && order > PAGE_ALLOC_COSTLY_ORDER &&
  3180. !gfp_pfmemalloc_allowed(gfp_mask)) {
  3181. page = __alloc_pages_direct_compact(gfp_mask, order,
  3182. alloc_flags, ac,
  3183. INIT_COMPACT_PRIORITY,
  3184. &compact_result);
  3185. if (page)
  3186. goto got_pg;
  3187. /*
  3188. * Checks for costly allocations with __GFP_NORETRY, which
  3189. * includes THP page fault allocations
  3190. */
  3191. if (gfp_mask & __GFP_NORETRY) {
  3192. /*
  3193. * If compaction is deferred for high-order allocations,
  3194. * it is because sync compaction recently failed. If
  3195. * this is the case and the caller requested a THP
  3196. * allocation, we do not want to heavily disrupt the
  3197. * system, so we fail the allocation instead of entering
  3198. * direct reclaim.
  3199. */
  3200. if (compact_result == COMPACT_DEFERRED)
  3201. goto nopage;
  3202. /*
  3203. * Looks like reclaim/compaction is worth trying, but
  3204. * sync compaction could be very expensive, so keep
  3205. * using async compaction.
  3206. */
  3207. compact_priority = INIT_COMPACT_PRIORITY;
  3208. }
  3209. }
  3210. retry:
  3211. /* Ensure kswapd doesn't accidentally go to sleep as long as we loop */
  3212. if (gfp_mask & __GFP_KSWAPD_RECLAIM)
  3213. wake_all_kswapds(order, ac);
  3214. if (gfp_pfmemalloc_allowed(gfp_mask))
  3215. alloc_flags = ALLOC_NO_WATERMARKS;
  3216. /*
  3217. * Reset the zonelist iterators if memory policies can be ignored.
  3218. * These allocations are high priority and system rather than user
  3219. * orientated.
  3220. */
  3221. if (!(alloc_flags & ALLOC_CPUSET) || (alloc_flags & ALLOC_NO_WATERMARKS)) {
  3222. ac->zonelist = node_zonelist(numa_node_id(), gfp_mask);
  3223. ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
  3224. ac->high_zoneidx, ac->nodemask);
  3225. }
  3226. /* Attempt with potentially adjusted zonelist and alloc_flags */
  3227. page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
  3228. if (page)
  3229. goto got_pg;
  3230. /* Caller is not willing to reclaim, we can't balance anything */
  3231. if (!can_direct_reclaim)
  3232. goto nopage;
  3233. /* Make sure we know about allocations which stall for too long */
  3234. if (time_after(jiffies, alloc_start + stall_timeout)) {
  3235. warn_alloc(gfp_mask, ac->nodemask,
  3236. "page allocation stalls for %ums, order:%u",
  3237. jiffies_to_msecs(jiffies-alloc_start), order);
  3238. stall_timeout += 10 * HZ;
  3239. }
  3240. /* Avoid recursion of direct reclaim */
  3241. if (current->flags & PF_MEMALLOC)
  3242. goto nopage;
  3243. /* Try direct reclaim and then allocating */
  3244. page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac,
  3245. &did_some_progress);
  3246. if (page)
  3247. goto got_pg;
  3248. /* Try direct compaction and then allocating */
  3249. page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac,
  3250. compact_priority, &compact_result);
  3251. if (page)
  3252. goto got_pg;
  3253. /* Do not loop if specifically requested */
  3254. if (gfp_mask & __GFP_NORETRY)
  3255. goto nopage;
  3256. /*
  3257. * Do not retry costly high order allocations unless they are
  3258. * __GFP_REPEAT
  3259. */
  3260. if (order > PAGE_ALLOC_COSTLY_ORDER && !(gfp_mask & __GFP_REPEAT))
  3261. goto nopage;
  3262. if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags,
  3263. did_some_progress > 0, &no_progress_loops))
  3264. goto retry;
  3265. /*
  3266. * It doesn't make any sense to retry for the compaction if the order-0
  3267. * reclaim is not able to make any progress because the current
  3268. * implementation of the compaction depends on the sufficient amount
  3269. * of free memory (see __compaction_suitable)
  3270. */
  3271. if (did_some_progress > 0 &&
  3272. should_compact_retry(ac, order, alloc_flags,
  3273. compact_result, &compact_priority,
  3274. &compaction_retries))
  3275. goto retry;
  3276. /*
  3277. * It's possible we raced with cpuset update so the OOM would be
  3278. * premature (see below the nopage: label for full explanation).
  3279. */
  3280. if (read_mems_allowed_retry(cpuset_mems_cookie))
  3281. goto retry_cpuset;
  3282. /* Reclaim has failed us, start killing things */
  3283. page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress);
  3284. if (page)
  3285. goto got_pg;
  3286. /* Avoid allocations with no watermarks from looping endlessly */
  3287. if (test_thread_flag(TIF_MEMDIE))
  3288. goto nopage;
  3289. /* Retry as long as the OOM killer is making progress */
  3290. if (did_some_progress) {
  3291. no_progress_loops = 0;
  3292. goto retry;
  3293. }
  3294. nopage:
  3295. /*
  3296. * When updating a task's mems_allowed or mempolicy nodemask, it is
  3297. * possible to race with parallel threads in such a way that our
  3298. * allocation can fail while the mask is being updated. If we are about
  3299. * to fail, check if the cpuset changed during allocation and if so,
  3300. * retry.
  3301. */
  3302. if (read_mems_allowed_retry(cpuset_mems_cookie))
  3303. goto retry_cpuset;
  3304. /*
  3305. * Make sure that __GFP_NOFAIL request doesn't leak out and make sure
  3306. * we always retry
  3307. */
  3308. if (gfp_mask & __GFP_NOFAIL) {
  3309. /*
  3310. * All existing users of the __GFP_NOFAIL are blockable, so warn
  3311. * of any new users that actually require GFP_NOWAIT
  3312. */
  3313. if (WARN_ON_ONCE(!can_direct_reclaim))
  3314. goto fail;
  3315. /*
  3316. * PF_MEMALLOC request from this context is rather bizarre
  3317. * because we cannot reclaim anything and only can loop waiting
  3318. * for somebody to do a work for us
  3319. */
  3320. WARN_ON_ONCE(current->flags & PF_MEMALLOC);
  3321. /*
  3322. * non failing costly orders are a hard requirement which we
  3323. * are not prepared for much so let's warn about these users
  3324. * so that we can identify them and convert them to something
  3325. * else.
  3326. */
  3327. WARN_ON_ONCE(order > PAGE_ALLOC_COSTLY_ORDER);
  3328. /*
  3329. * Help non-failing allocations by giving them access to memory
  3330. * reserves but do not use ALLOC_NO_WATERMARKS because this
  3331. * could deplete whole memory reserves which would just make
  3332. * the situation worse
  3333. */
  3334. page = __alloc_pages_cpuset_fallback(gfp_mask, order, ALLOC_HARDER, ac);
  3335. if (page)
  3336. goto got_pg;
  3337. cond_resched();
  3338. goto retry;
  3339. }
  3340. fail:
  3341. warn_alloc(gfp_mask, ac->nodemask,
  3342. "page allocation failure: order:%u", order);
  3343. got_pg:
  3344. return page;
  3345. }
  3346. static inline bool prepare_alloc_pages(gfp_t gfp_mask, unsigned int order,
  3347. struct zonelist *zonelist, nodemask_t *nodemask,
  3348. struct alloc_context *ac, gfp_t *alloc_mask,
  3349. unsigned int *alloc_flags)
  3350. {
  3351. ac->high_zoneidx = gfp_zone(gfp_mask);
  3352. ac->zonelist = zonelist;
  3353. ac->nodemask = nodemask;
  3354. ac->migratetype = gfpflags_to_migratetype(gfp_mask);
  3355. if (cpusets_enabled()) {
  3356. *alloc_mask |= __GFP_HARDWALL;
  3357. if (!ac->nodemask)
  3358. ac->nodemask = &cpuset_current_mems_allowed;
  3359. else
  3360. *alloc_flags |= ALLOC_CPUSET;
  3361. }
  3362. lockdep_trace_alloc(gfp_mask);
  3363. might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM);
  3364. if (should_fail_alloc_page(gfp_mask, order))
  3365. return false;
  3366. if (IS_ENABLED(CONFIG_CMA) && ac->migratetype == MIGRATE_MOVABLE)
  3367. *alloc_flags |= ALLOC_CMA;
  3368. return true;
  3369. }
  3370. /* Determine whether to spread dirty pages and what the first usable zone */
  3371. static inline void finalise_ac(gfp_t gfp_mask,
  3372. unsigned int order, struct alloc_context *ac)
  3373. {
  3374. /* Dirty zone balancing only done in the fast path */
  3375. ac->spread_dirty_pages = (gfp_mask & __GFP_WRITE);
  3376. /*
  3377. * The preferred zone is used for statistics but crucially it is
  3378. * also used as the starting point for the zonelist iterator. It
  3379. * may get reset for allocations that ignore memory policies.
  3380. */
  3381. ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
  3382. ac->high_zoneidx, ac->nodemask);
  3383. }
  3384. /*
  3385. * This is the 'heart' of the zoned buddy allocator.
  3386. */
  3387. struct page *
  3388. __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
  3389. struct zonelist *zonelist, nodemask_t *nodemask)
  3390. {
  3391. struct page *page;
  3392. unsigned int alloc_flags = ALLOC_WMARK_LOW;
  3393. gfp_t alloc_mask = gfp_mask; /* The gfp_t that was actually used for allocation */
  3394. struct alloc_context ac = { };
  3395. gfp_mask &= gfp_allowed_mask;
  3396. if (!prepare_alloc_pages(gfp_mask, order, zonelist, nodemask, &ac, &alloc_mask, &alloc_flags))
  3397. return NULL;
  3398. finalise_ac(gfp_mask, order, &ac);
  3399. /* First allocation attempt */
  3400. page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac);
  3401. if (likely(page))
  3402. goto out;
  3403. /*
  3404. * Runtime PM, block IO and its error handling path can deadlock
  3405. * because I/O on the device might not complete.
  3406. */
  3407. alloc_mask = memalloc_noio_flags(gfp_mask);
  3408. ac.spread_dirty_pages = false;
  3409. /*
  3410. * Restore the original nodemask if it was potentially replaced with
  3411. * &cpuset_current_mems_allowed to optimize the fast-path attempt.
  3412. */
  3413. if (unlikely(ac.nodemask != nodemask))
  3414. ac.nodemask = nodemask;
  3415. page = __alloc_pages_slowpath(alloc_mask, order, &ac);
  3416. out:
  3417. if (memcg_kmem_enabled() && (gfp_mask & __GFP_ACCOUNT) && page &&
  3418. unlikely(memcg_kmem_charge(page, gfp_mask, order) != 0)) {
  3419. __free_pages(page, order);
  3420. page = NULL;
  3421. }
  3422. if (kmemcheck_enabled && page)
  3423. kmemcheck_pagealloc_alloc(page, order, gfp_mask);
  3424. trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype);
  3425. return page;
  3426. }
  3427. EXPORT_SYMBOL(__alloc_pages_nodemask);
  3428. /*
  3429. * Common helper functions.
  3430. */
  3431. unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
  3432. {
  3433. struct page *page;
  3434. /*
  3435. * __get_free_pages() returns a 32-bit address, which cannot represent
  3436. * a highmem page
  3437. */
  3438. VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
  3439. page = alloc_pages(gfp_mask, order);
  3440. if (!page)
  3441. return 0;
  3442. return (unsigned long) page_address(page);
  3443. }
  3444. EXPORT_SYMBOL(__get_free_pages);
  3445. unsigned long get_zeroed_page(gfp_t gfp_mask)
  3446. {
  3447. return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
  3448. }
  3449. EXPORT_SYMBOL(get_zeroed_page);
  3450. void __free_pages(struct page *page, unsigned int order)
  3451. {
  3452. if (put_page_testzero(page)) {
  3453. if (order == 0)
  3454. free_hot_cold_page(page, false);
  3455. else
  3456. __free_pages_ok(page, order);
  3457. }
  3458. }
  3459. EXPORT_SYMBOL(__free_pages);
  3460. void free_pages(unsigned long addr, unsigned int order)
  3461. {
  3462. if (addr != 0) {
  3463. VM_BUG_ON(!virt_addr_valid((void *)addr));
  3464. __free_pages(virt_to_page((void *)addr), order);
  3465. }
  3466. }
  3467. EXPORT_SYMBOL(free_pages);
  3468. /*
  3469. * Page Fragment:
  3470. * An arbitrary-length arbitrary-offset area of memory which resides
  3471. * within a 0 or higher order page. Multiple fragments within that page
  3472. * are individually refcounted, in the page's reference counter.
  3473. *
  3474. * The page_frag functions below provide a simple allocation framework for
  3475. * page fragments. This is used by the network stack and network device
  3476. * drivers to provide a backing region of memory for use as either an
  3477. * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
  3478. */
  3479. static struct page *__page_frag_cache_refill(struct page_frag_cache *nc,
  3480. gfp_t gfp_mask)
  3481. {
  3482. struct page *page = NULL;
  3483. gfp_t gfp = gfp_mask;
  3484. #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
  3485. gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY |
  3486. __GFP_NOMEMALLOC;
  3487. page = alloc_pages_node(NUMA_NO_NODE, gfp_mask,
  3488. PAGE_FRAG_CACHE_MAX_ORDER);
  3489. nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE;
  3490. #endif
  3491. if (unlikely(!page))
  3492. page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
  3493. nc->va = page ? page_address(page) : NULL;
  3494. return page;
  3495. }
  3496. void __page_frag_cache_drain(struct page *page, unsigned int count)
  3497. {
  3498. VM_BUG_ON_PAGE(page_ref_count(page) == 0, page);
  3499. if (page_ref_sub_and_test(page, count)) {
  3500. unsigned int order = compound_order(page);
  3501. if (order == 0)
  3502. free_hot_cold_page(page, false);
  3503. else
  3504. __free_pages_ok(page, order);
  3505. }
  3506. }
  3507. EXPORT_SYMBOL(__page_frag_cache_drain);
  3508. void *page_frag_alloc(struct page_frag_cache *nc,
  3509. unsigned int fragsz, gfp_t gfp_mask)
  3510. {
  3511. unsigned int size = PAGE_SIZE;
  3512. struct page *page;
  3513. int offset;
  3514. if (unlikely(!nc->va)) {
  3515. refill:
  3516. page = __page_frag_cache_refill(nc, gfp_mask);
  3517. if (!page)
  3518. return NULL;
  3519. #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
  3520. /* if size can vary use size else just use PAGE_SIZE */
  3521. size = nc->size;
  3522. #endif
  3523. /* Even if we own the page, we do not use atomic_set().
  3524. * This would break get_page_unless_zero() users.
  3525. */
  3526. page_ref_add(page, size - 1);
  3527. /* reset page count bias and offset to start of new frag */
  3528. nc->pfmemalloc = page_is_pfmemalloc(page);
  3529. nc->pagecnt_bias = size;
  3530. nc->offset = size;
  3531. }
  3532. offset = nc->offset - fragsz;
  3533. if (unlikely(offset < 0)) {
  3534. page = virt_to_page(nc->va);
  3535. if (!page_ref_sub_and_test(page, nc->pagecnt_bias))
  3536. goto refill;
  3537. #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
  3538. /* if size can vary use size else just use PAGE_SIZE */
  3539. size = nc->size;
  3540. #endif
  3541. /* OK, page count is 0, we can safely set it */
  3542. set_page_count(page, size);
  3543. /* reset page count bias and offset to start of new frag */
  3544. nc->pagecnt_bias = size;
  3545. offset = size - fragsz;
  3546. }
  3547. nc->pagecnt_bias--;
  3548. nc->offset = offset;
  3549. return nc->va + offset;
  3550. }
  3551. EXPORT_SYMBOL(page_frag_alloc);
  3552. /*
  3553. * Frees a page fragment allocated out of either a compound or order 0 page.
  3554. */
  3555. void page_frag_free(void *addr)
  3556. {
  3557. struct page *page = virt_to_head_page(addr);
  3558. if (unlikely(put_page_testzero(page)))
  3559. __free_pages_ok(page, compound_order(page));
  3560. }
  3561. EXPORT_SYMBOL(page_frag_free);
  3562. static void *make_alloc_exact(unsigned long addr, unsigned int order,
  3563. size_t size)
  3564. {
  3565. if (addr) {
  3566. unsigned long alloc_end = addr + (PAGE_SIZE << order);
  3567. unsigned long used = addr + PAGE_ALIGN(size);
  3568. split_page(virt_to_page((void *)addr), order);
  3569. while (used < alloc_end) {
  3570. free_page(used);
  3571. used += PAGE_SIZE;
  3572. }
  3573. }
  3574. return (void *)addr;
  3575. }
  3576. /**
  3577. * alloc_pages_exact - allocate an exact number physically-contiguous pages.
  3578. * @size: the number of bytes to allocate
  3579. * @gfp_mask: GFP flags for the allocation
  3580. *
  3581. * This function is similar to alloc_pages(), except that it allocates the
  3582. * minimum number of pages to satisfy the request. alloc_pages() can only
  3583. * allocate memory in power-of-two pages.
  3584. *
  3585. * This function is also limited by MAX_ORDER.
  3586. *
  3587. * Memory allocated by this function must be released by free_pages_exact().
  3588. */
  3589. void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
  3590. {
  3591. unsigned int order = get_order(size);
  3592. unsigned long addr;
  3593. addr = __get_free_pages(gfp_mask, order);
  3594. return make_alloc_exact(addr, order, size);
  3595. }
  3596. EXPORT_SYMBOL(alloc_pages_exact);
  3597. /**
  3598. * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
  3599. * pages on a node.
  3600. * @nid: the preferred node ID where memory should be allocated
  3601. * @size: the number of bytes to allocate
  3602. * @gfp_mask: GFP flags for the allocation
  3603. *
  3604. * Like alloc_pages_exact(), but try to allocate on node nid first before falling
  3605. * back.
  3606. */
  3607. void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
  3608. {
  3609. unsigned int order = get_order(size);
  3610. struct page *p = alloc_pages_node(nid, gfp_mask, order);
  3611. if (!p)
  3612. return NULL;
  3613. return make_alloc_exact((unsigned long)page_address(p), order, size);
  3614. }
  3615. /**
  3616. * free_pages_exact - release memory allocated via alloc_pages_exact()
  3617. * @virt: the value returned by alloc_pages_exact.
  3618. * @size: size of allocation, same value as passed to alloc_pages_exact().
  3619. *
  3620. * Release the memory allocated by a previous call to alloc_pages_exact.
  3621. */
  3622. void free_pages_exact(void *virt, size_t size)
  3623. {
  3624. unsigned long addr = (unsigned long)virt;
  3625. unsigned long end = addr + PAGE_ALIGN(size);
  3626. while (addr < end) {
  3627. free_page(addr);
  3628. addr += PAGE_SIZE;
  3629. }
  3630. }
  3631. EXPORT_SYMBOL(free_pages_exact);
  3632. /**
  3633. * nr_free_zone_pages - count number of pages beyond high watermark
  3634. * @offset: The zone index of the highest zone
  3635. *
  3636. * nr_free_zone_pages() counts the number of counts pages which are beyond the
  3637. * high watermark within all zones at or below a given zone index. For each
  3638. * zone, the number of pages is calculated as:
  3639. *
  3640. * nr_free_zone_pages = managed_pages - high_pages
  3641. */
  3642. static unsigned long nr_free_zone_pages(int offset)
  3643. {
  3644. struct zoneref *z;
  3645. struct zone *zone;
  3646. /* Just pick one node, since fallback list is circular */
  3647. unsigned long sum = 0;
  3648. struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
  3649. for_each_zone_zonelist(zone, z, zonelist, offset) {
  3650. unsigned long size = zone->managed_pages;
  3651. unsigned long high = high_wmark_pages(zone);
  3652. if (size > high)
  3653. sum += size - high;
  3654. }
  3655. return sum;
  3656. }
  3657. /**
  3658. * nr_free_buffer_pages - count number of pages beyond high watermark
  3659. *
  3660. * nr_free_buffer_pages() counts the number of pages which are beyond the high
  3661. * watermark within ZONE_DMA and ZONE_NORMAL.
  3662. */
  3663. unsigned long nr_free_buffer_pages(void)
  3664. {
  3665. return nr_free_zone_pages(gfp_zone(GFP_USER));
  3666. }
  3667. EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
  3668. /**
  3669. * nr_free_pagecache_pages - count number of pages beyond high watermark
  3670. *
  3671. * nr_free_pagecache_pages() counts the number of pages which are beyond the
  3672. * high watermark within all zones.
  3673. */
  3674. unsigned long nr_free_pagecache_pages(void)
  3675. {
  3676. return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
  3677. }
  3678. static inline void show_node(struct zone *zone)
  3679. {
  3680. if (IS_ENABLED(CONFIG_NUMA))
  3681. printk("Node %d ", zone_to_nid(zone));
  3682. }
  3683. long si_mem_available(void)
  3684. {
  3685. long available;
  3686. unsigned long pagecache;
  3687. unsigned long wmark_low = 0;
  3688. unsigned long pages[NR_LRU_LISTS];
  3689. struct zone *zone;
  3690. int lru;
  3691. for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++)
  3692. pages[lru] = global_node_page_state(NR_LRU_BASE + lru);
  3693. for_each_zone(zone)
  3694. wmark_low += zone->watermark[WMARK_LOW];
  3695. /*
  3696. * Estimate the amount of memory available for userspace allocations,
  3697. * without causing swapping.
  3698. */
  3699. available = global_page_state(NR_FREE_PAGES) - totalreserve_pages;
  3700. /*
  3701. * Not all the page cache can be freed, otherwise the system will
  3702. * start swapping. Assume at least half of the page cache, or the
  3703. * low watermark worth of cache, needs to stay.
  3704. */
  3705. pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE];
  3706. pagecache -= min(pagecache / 2, wmark_low);
  3707. available += pagecache;
  3708. /*
  3709. * Part of the reclaimable slab consists of items that are in use,
  3710. * and cannot be freed. Cap this estimate at the low watermark.
  3711. */
  3712. available += global_page_state(NR_SLAB_RECLAIMABLE) -
  3713. min(global_page_state(NR_SLAB_RECLAIMABLE) / 2, wmark_low);
  3714. if (available < 0)
  3715. available = 0;
  3716. return available;
  3717. }
  3718. EXPORT_SYMBOL_GPL(si_mem_available);
  3719. void si_meminfo(struct sysinfo *val)
  3720. {
  3721. val->totalram = totalram_pages;
  3722. val->sharedram = global_node_page_state(NR_SHMEM);
  3723. val->freeram = global_page_state(NR_FREE_PAGES);
  3724. val->bufferram = nr_blockdev_pages();
  3725. val->totalhigh = totalhigh_pages;
  3726. val->freehigh = nr_free_highpages();
  3727. val->mem_unit = PAGE_SIZE;
  3728. }
  3729. EXPORT_SYMBOL(si_meminfo);
  3730. #ifdef CONFIG_NUMA
  3731. void si_meminfo_node(struct sysinfo *val, int nid)
  3732. {
  3733. int zone_type; /* needs to be signed */
  3734. unsigned long managed_pages = 0;
  3735. unsigned long managed_highpages = 0;
  3736. unsigned long free_highpages = 0;
  3737. pg_data_t *pgdat = NODE_DATA(nid);
  3738. for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
  3739. managed_pages += pgdat->node_zones[zone_type].managed_pages;
  3740. val->totalram = managed_pages;
  3741. val->sharedram = node_page_state(pgdat, NR_SHMEM);
  3742. val->freeram = sum_zone_node_page_state(nid, NR_FREE_PAGES);
  3743. #ifdef CONFIG_HIGHMEM
  3744. for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
  3745. struct zone *zone = &pgdat->node_zones[zone_type];
  3746. if (is_highmem(zone)) {
  3747. managed_highpages += zone->managed_pages;
  3748. free_highpages += zone_page_state(zone, NR_FREE_PAGES);
  3749. }
  3750. }
  3751. val->totalhigh = managed_highpages;
  3752. val->freehigh = free_highpages;
  3753. #else
  3754. val->totalhigh = managed_highpages;
  3755. val->freehigh = free_highpages;
  3756. #endif
  3757. val->mem_unit = PAGE_SIZE;
  3758. }
  3759. #endif
  3760. /*
  3761. * Determine whether the node should be displayed or not, depending on whether
  3762. * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
  3763. */
  3764. static bool show_mem_node_skip(unsigned int flags, int nid, nodemask_t *nodemask)
  3765. {
  3766. if (!(flags & SHOW_MEM_FILTER_NODES))
  3767. return false;
  3768. /*
  3769. * no node mask - aka implicit memory numa policy. Do not bother with
  3770. * the synchronization - read_mems_allowed_begin - because we do not
  3771. * have to be precise here.
  3772. */
  3773. if (!nodemask)
  3774. nodemask = &cpuset_current_mems_allowed;
  3775. return !node_isset(nid, *nodemask);
  3776. }
  3777. #define K(x) ((x) << (PAGE_SHIFT-10))
  3778. static void show_migration_types(unsigned char type)
  3779. {
  3780. static const char types[MIGRATE_TYPES] = {
  3781. [MIGRATE_UNMOVABLE] = 'U',
  3782. [MIGRATE_MOVABLE] = 'M',
  3783. [MIGRATE_RECLAIMABLE] = 'E',
  3784. [MIGRATE_HIGHATOMIC] = 'H',
  3785. #ifdef CONFIG_CMA
  3786. [MIGRATE_CMA] = 'C',
  3787. #endif
  3788. #ifdef CONFIG_MEMORY_ISOLATION
  3789. [MIGRATE_ISOLATE] = 'I',
  3790. #endif
  3791. };
  3792. char tmp[MIGRATE_TYPES + 1];
  3793. char *p = tmp;
  3794. int i;
  3795. for (i = 0; i < MIGRATE_TYPES; i++) {
  3796. if (type & (1 << i))
  3797. *p++ = types[i];
  3798. }
  3799. *p = '\0';
  3800. printk(KERN_CONT "(%s) ", tmp);
  3801. }
  3802. /*
  3803. * Show free area list (used inside shift_scroll-lock stuff)
  3804. * We also calculate the percentage fragmentation. We do this by counting the
  3805. * memory on each free list with the exception of the first item on the list.
  3806. *
  3807. * Bits in @filter:
  3808. * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
  3809. * cpuset.
  3810. */
  3811. void show_free_areas(unsigned int filter, nodemask_t *nodemask)
  3812. {
  3813. unsigned long free_pcp = 0;
  3814. int cpu;
  3815. struct zone *zone;
  3816. pg_data_t *pgdat;
  3817. for_each_populated_zone(zone) {
  3818. if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
  3819. continue;
  3820. for_each_online_cpu(cpu)
  3821. free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
  3822. }
  3823. printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
  3824. " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
  3825. " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
  3826. " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
  3827. " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
  3828. " free:%lu free_pcp:%lu free_cma:%lu\n",
  3829. global_node_page_state(NR_ACTIVE_ANON),
  3830. global_node_page_state(NR_INACTIVE_ANON),
  3831. global_node_page_state(NR_ISOLATED_ANON),
  3832. global_node_page_state(NR_ACTIVE_FILE),
  3833. global_node_page_state(NR_INACTIVE_FILE),
  3834. global_node_page_state(NR_ISOLATED_FILE),
  3835. global_node_page_state(NR_UNEVICTABLE),
  3836. global_node_page_state(NR_FILE_DIRTY),
  3837. global_node_page_state(NR_WRITEBACK),
  3838. global_node_page_state(NR_UNSTABLE_NFS),
  3839. global_page_state(NR_SLAB_RECLAIMABLE),
  3840. global_page_state(NR_SLAB_UNRECLAIMABLE),
  3841. global_node_page_state(NR_FILE_MAPPED),
  3842. global_node_page_state(NR_SHMEM),
  3843. global_page_state(NR_PAGETABLE),
  3844. global_page_state(NR_BOUNCE),
  3845. global_page_state(NR_FREE_PAGES),
  3846. free_pcp,
  3847. global_page_state(NR_FREE_CMA_PAGES));
  3848. for_each_online_pgdat(pgdat) {
  3849. if (show_mem_node_skip(filter, pgdat->node_id, nodemask))
  3850. continue;
  3851. printk("Node %d"
  3852. " active_anon:%lukB"
  3853. " inactive_anon:%lukB"
  3854. " active_file:%lukB"
  3855. " inactive_file:%lukB"
  3856. " unevictable:%lukB"
  3857. " isolated(anon):%lukB"
  3858. " isolated(file):%lukB"
  3859. " mapped:%lukB"
  3860. " dirty:%lukB"
  3861. " writeback:%lukB"
  3862. " shmem:%lukB"
  3863. #ifdef CONFIG_TRANSPARENT_HUGEPAGE
  3864. " shmem_thp: %lukB"
  3865. " shmem_pmdmapped: %lukB"
  3866. " anon_thp: %lukB"
  3867. #endif
  3868. " writeback_tmp:%lukB"
  3869. " unstable:%lukB"
  3870. " pages_scanned:%lu"
  3871. " all_unreclaimable? %s"
  3872. "\n",
  3873. pgdat->node_id,
  3874. K(node_page_state(pgdat, NR_ACTIVE_ANON)),
  3875. K(node_page_state(pgdat, NR_INACTIVE_ANON)),
  3876. K(node_page_state(pgdat, NR_ACTIVE_FILE)),
  3877. K(node_page_state(pgdat, NR_INACTIVE_FILE)),
  3878. K(node_page_state(pgdat, NR_UNEVICTABLE)),
  3879. K(node_page_state(pgdat, NR_ISOLATED_ANON)),
  3880. K(node_page_state(pgdat, NR_ISOLATED_FILE)),
  3881. K(node_page_state(pgdat, NR_FILE_MAPPED)),
  3882. K(node_page_state(pgdat, NR_FILE_DIRTY)),
  3883. K(node_page_state(pgdat, NR_WRITEBACK)),
  3884. #ifdef CONFIG_TRANSPARENT_HUGEPAGE
  3885. K(node_page_state(pgdat, NR_SHMEM_THPS) * HPAGE_PMD_NR),
  3886. K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED)
  3887. * HPAGE_PMD_NR),
  3888. K(node_page_state(pgdat, NR_ANON_THPS) * HPAGE_PMD_NR),
  3889. #endif
  3890. K(node_page_state(pgdat, NR_SHMEM)),
  3891. K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
  3892. K(node_page_state(pgdat, NR_UNSTABLE_NFS)),
  3893. node_page_state(pgdat, NR_PAGES_SCANNED),
  3894. !pgdat_reclaimable(pgdat) ? "yes" : "no");
  3895. }
  3896. for_each_populated_zone(zone) {
  3897. int i;
  3898. if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
  3899. continue;
  3900. free_pcp = 0;
  3901. for_each_online_cpu(cpu)
  3902. free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
  3903. show_node(zone);
  3904. printk(KERN_CONT
  3905. "%s"
  3906. " free:%lukB"
  3907. " min:%lukB"
  3908. " low:%lukB"
  3909. " high:%lukB"
  3910. " active_anon:%lukB"
  3911. " inactive_anon:%lukB"
  3912. " active_file:%lukB"
  3913. " inactive_file:%lukB"
  3914. " unevictable:%lukB"
  3915. " writepending:%lukB"
  3916. " present:%lukB"
  3917. " managed:%lukB"
  3918. " mlocked:%lukB"
  3919. " slab_reclaimable:%lukB"
  3920. " slab_unreclaimable:%lukB"
  3921. " kernel_stack:%lukB"
  3922. " pagetables:%lukB"
  3923. " bounce:%lukB"
  3924. " free_pcp:%lukB"
  3925. " local_pcp:%ukB"
  3926. " free_cma:%lukB"
  3927. "\n",
  3928. zone->name,
  3929. K(zone_page_state(zone, NR_FREE_PAGES)),
  3930. K(min_wmark_pages(zone)),
  3931. K(low_wmark_pages(zone)),
  3932. K(high_wmark_pages(zone)),
  3933. K(zone_page_state(zone, NR_ZONE_ACTIVE_ANON)),
  3934. K(zone_page_state(zone, NR_ZONE_INACTIVE_ANON)),
  3935. K(zone_page_state(zone, NR_ZONE_ACTIVE_FILE)),
  3936. K(zone_page_state(zone, NR_ZONE_INACTIVE_FILE)),
  3937. K(zone_page_state(zone, NR_ZONE_UNEVICTABLE)),
  3938. K(zone_page_state(zone, NR_ZONE_WRITE_PENDING)),
  3939. K(zone->present_pages),
  3940. K(zone->managed_pages),
  3941. K(zone_page_state(zone, NR_MLOCK)),
  3942. K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
  3943. K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
  3944. zone_page_state(zone, NR_KERNEL_STACK_KB),
  3945. K(zone_page_state(zone, NR_PAGETABLE)),
  3946. K(zone_page_state(zone, NR_BOUNCE)),
  3947. K(free_pcp),
  3948. K(this_cpu_read(zone->pageset->pcp.count)),
  3949. K(zone_page_state(zone, NR_FREE_CMA_PAGES)));
  3950. printk("lowmem_reserve[]:");
  3951. for (i = 0; i < MAX_NR_ZONES; i++)
  3952. printk(KERN_CONT " %ld", zone->lowmem_reserve[i]);
  3953. printk(KERN_CONT "\n");
  3954. }
  3955. for_each_populated_zone(zone) {
  3956. unsigned int order;
  3957. unsigned long nr[MAX_ORDER], flags, total = 0;
  3958. unsigned char types[MAX_ORDER];
  3959. if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
  3960. continue;
  3961. show_node(zone);
  3962. printk(KERN_CONT "%s: ", zone->name);
  3963. spin_lock_irqsave(&zone->lock, flags);
  3964. for (order = 0; order < MAX_ORDER; order++) {
  3965. struct free_area *area = &zone->free_area[order];
  3966. int type;
  3967. nr[order] = area->nr_free;
  3968. total += nr[order] << order;
  3969. types[order] = 0;
  3970. for (type = 0; type < MIGRATE_TYPES; type++) {
  3971. if (!list_empty(&area->free_list[type]))
  3972. types[order] |= 1 << type;
  3973. }
  3974. }
  3975. spin_unlock_irqrestore(&zone->lock, flags);
  3976. for (order = 0; order < MAX_ORDER; order++) {
  3977. printk(KERN_CONT "%lu*%lukB ",
  3978. nr[order], K(1UL) << order);
  3979. if (nr[order])
  3980. show_migration_types(types[order]);
  3981. }
  3982. printk(KERN_CONT "= %lukB\n", K(total));
  3983. }
  3984. hugetlb_show_meminfo();
  3985. printk("%ld total pagecache pages\n", global_node_page_state(NR_FILE_PAGES));
  3986. show_swap_cache_info();
  3987. }
  3988. static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
  3989. {
  3990. zoneref->zone = zone;
  3991. zoneref->zone_idx = zone_idx(zone);
  3992. }
  3993. /*
  3994. * Builds allocation fallback zone lists.
  3995. *
  3996. * Add all populated zones of a node to the zonelist.
  3997. */
  3998. static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
  3999. int nr_zones)
  4000. {
  4001. struct zone *zone;
  4002. enum zone_type zone_type = MAX_NR_ZONES;
  4003. do {
  4004. zone_type--;
  4005. zone = pgdat->node_zones + zone_type;
  4006. if (managed_zone(zone)) {
  4007. zoneref_set_zone(zone,
  4008. &zonelist->_zonerefs[nr_zones++]);
  4009. check_highest_zone(zone_type);
  4010. }
  4011. } while (zone_type);
  4012. return nr_zones;
  4013. }
  4014. /*
  4015. * zonelist_order:
  4016. * 0 = automatic detection of better ordering.
  4017. * 1 = order by ([node] distance, -zonetype)
  4018. * 2 = order by (-zonetype, [node] distance)
  4019. *
  4020. * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
  4021. * the same zonelist. So only NUMA can configure this param.
  4022. */
  4023. #define ZONELIST_ORDER_DEFAULT 0
  4024. #define ZONELIST_ORDER_NODE 1
  4025. #define ZONELIST_ORDER_ZONE 2
  4026. /* zonelist order in the kernel.
  4027. * set_zonelist_order() will set this to NODE or ZONE.
  4028. */
  4029. static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
  4030. static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
  4031. #ifdef CONFIG_NUMA
  4032. /* The value user specified ....changed by config */
  4033. static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
  4034. /* string for sysctl */
  4035. #define NUMA_ZONELIST_ORDER_LEN 16
  4036. char numa_zonelist_order[16] = "default";
  4037. /*
  4038. * interface for configure zonelist ordering.
  4039. * command line option "numa_zonelist_order"
  4040. * = "[dD]efault - default, automatic configuration.
  4041. * = "[nN]ode - order by node locality, then by zone within node
  4042. * = "[zZ]one - order by zone, then by locality within zone
  4043. */
  4044. static int __parse_numa_zonelist_order(char *s)
  4045. {
  4046. if (*s == 'd' || *s == 'D') {
  4047. user_zonelist_order = ZONELIST_ORDER_DEFAULT;
  4048. } else if (*s == 'n' || *s == 'N') {
  4049. user_zonelist_order = ZONELIST_ORDER_NODE;
  4050. } else if (*s == 'z' || *s == 'Z') {
  4051. user_zonelist_order = ZONELIST_ORDER_ZONE;
  4052. } else {
  4053. pr_warn("Ignoring invalid numa_zonelist_order value: %s\n", s);
  4054. return -EINVAL;
  4055. }
  4056. return 0;
  4057. }
  4058. static __init int setup_numa_zonelist_order(char *s)
  4059. {
  4060. int ret;
  4061. if (!s)
  4062. return 0;
  4063. ret = __parse_numa_zonelist_order(s);
  4064. if (ret == 0)
  4065. strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
  4066. return ret;
  4067. }
  4068. early_param("numa_zonelist_order", setup_numa_zonelist_order);
  4069. /*
  4070. * sysctl handler for numa_zonelist_order
  4071. */
  4072. int numa_zonelist_order_handler(struct ctl_table *table, int write,
  4073. void __user *buffer, size_t *length,
  4074. loff_t *ppos)
  4075. {
  4076. char saved_string[NUMA_ZONELIST_ORDER_LEN];
  4077. int ret;
  4078. static DEFINE_MUTEX(zl_order_mutex);
  4079. mutex_lock(&zl_order_mutex);
  4080. if (write) {
  4081. if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) {
  4082. ret = -EINVAL;
  4083. goto out;
  4084. }
  4085. strcpy(saved_string, (char *)table->data);
  4086. }
  4087. ret = proc_dostring(table, write, buffer, length, ppos);
  4088. if (ret)
  4089. goto out;
  4090. if (write) {
  4091. int oldval = user_zonelist_order;
  4092. ret = __parse_numa_zonelist_order((char *)table->data);
  4093. if (ret) {
  4094. /*
  4095. * bogus value. restore saved string
  4096. */
  4097. strncpy((char *)table->data, saved_string,
  4098. NUMA_ZONELIST_ORDER_LEN);
  4099. user_zonelist_order = oldval;
  4100. } else if (oldval != user_zonelist_order) {
  4101. mutex_lock(&zonelists_mutex);
  4102. build_all_zonelists(NULL, NULL);
  4103. mutex_unlock(&zonelists_mutex);
  4104. }
  4105. }
  4106. out:
  4107. mutex_unlock(&zl_order_mutex);
  4108. return ret;
  4109. }
  4110. #define MAX_NODE_LOAD (nr_online_nodes)
  4111. static int node_load[MAX_NUMNODES];
  4112. /**
  4113. * find_next_best_node - find the next node that should appear in a given node's fallback list
  4114. * @node: node whose fallback list we're appending
  4115. * @used_node_mask: nodemask_t of already used nodes
  4116. *
  4117. * We use a number of factors to determine which is the next node that should
  4118. * appear on a given node's fallback list. The node should not have appeared
  4119. * already in @node's fallback list, and it should be the next closest node
  4120. * according to the distance array (which contains arbitrary distance values
  4121. * from each node to each node in the system), and should also prefer nodes
  4122. * with no CPUs, since presumably they'll have very little allocation pressure
  4123. * on them otherwise.
  4124. * It returns -1 if no node is found.
  4125. */
  4126. static int find_next_best_node(int node, nodemask_t *used_node_mask)
  4127. {
  4128. int n, val;
  4129. int min_val = INT_MAX;
  4130. int best_node = NUMA_NO_NODE;
  4131. const struct cpumask *tmp = cpumask_of_node(0);
  4132. /* Use the local node if we haven't already */
  4133. if (!node_isset(node, *used_node_mask)) {
  4134. node_set(node, *used_node_mask);
  4135. return node;
  4136. }
  4137. for_each_node_state(n, N_MEMORY) {
  4138. /* Don't want a node to appear more than once */
  4139. if (node_isset(n, *used_node_mask))
  4140. continue;
  4141. /* Use the distance array to find the distance */
  4142. val = node_distance(node, n);
  4143. /* Penalize nodes under us ("prefer the next node") */
  4144. val += (n < node);
  4145. /* Give preference to headless and unused nodes */
  4146. tmp = cpumask_of_node(n);
  4147. if (!cpumask_empty(tmp))
  4148. val += PENALTY_FOR_NODE_WITH_CPUS;
  4149. /* Slight preference for less loaded node */
  4150. val *= (MAX_NODE_LOAD*MAX_NUMNODES);
  4151. val += node_load[n];
  4152. if (val < min_val) {
  4153. min_val = val;
  4154. best_node = n;
  4155. }
  4156. }
  4157. if (best_node >= 0)
  4158. node_set(best_node, *used_node_mask);
  4159. return best_node;
  4160. }
  4161. /*
  4162. * Build zonelists ordered by node and zones within node.
  4163. * This results in maximum locality--normal zone overflows into local
  4164. * DMA zone, if any--but risks exhausting DMA zone.
  4165. */
  4166. static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
  4167. {
  4168. int j;
  4169. struct zonelist *zonelist;
  4170. zonelist = &pgdat->node_zonelists[ZONELIST_FALLBACK];
  4171. for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
  4172. ;
  4173. j = build_zonelists_node(NODE_DATA(node), zonelist, j);
  4174. zonelist->_zonerefs[j].zone = NULL;
  4175. zonelist->_zonerefs[j].zone_idx = 0;
  4176. }
  4177. /*
  4178. * Build gfp_thisnode zonelists
  4179. */
  4180. static void build_thisnode_zonelists(pg_data_t *pgdat)
  4181. {
  4182. int j;
  4183. struct zonelist *zonelist;
  4184. zonelist = &pgdat->node_zonelists[ZONELIST_NOFALLBACK];
  4185. j = build_zonelists_node(pgdat, zonelist, 0);
  4186. zonelist->_zonerefs[j].zone = NULL;
  4187. zonelist->_zonerefs[j].zone_idx = 0;
  4188. }
  4189. /*
  4190. * Build zonelists ordered by zone and nodes within zones.
  4191. * This results in conserving DMA zone[s] until all Normal memory is
  4192. * exhausted, but results in overflowing to remote node while memory
  4193. * may still exist in local DMA zone.
  4194. */
  4195. static int node_order[MAX_NUMNODES];
  4196. static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
  4197. {
  4198. int pos, j, node;
  4199. int zone_type; /* needs to be signed */
  4200. struct zone *z;
  4201. struct zonelist *zonelist;
  4202. zonelist = &pgdat->node_zonelists[ZONELIST_FALLBACK];
  4203. pos = 0;
  4204. for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
  4205. for (j = 0; j < nr_nodes; j++) {
  4206. node = node_order[j];
  4207. z = &NODE_DATA(node)->node_zones[zone_type];
  4208. if (managed_zone(z)) {
  4209. zoneref_set_zone(z,
  4210. &zonelist->_zonerefs[pos++]);
  4211. check_highest_zone(zone_type);
  4212. }
  4213. }
  4214. }
  4215. zonelist->_zonerefs[pos].zone = NULL;
  4216. zonelist->_zonerefs[pos].zone_idx = 0;
  4217. }
  4218. #if defined(CONFIG_64BIT)
  4219. /*
  4220. * Devices that require DMA32/DMA are relatively rare and do not justify a
  4221. * penalty to every machine in case the specialised case applies. Default
  4222. * to Node-ordering on 64-bit NUMA machines
  4223. */
  4224. static int default_zonelist_order(void)
  4225. {
  4226. return ZONELIST_ORDER_NODE;
  4227. }
  4228. #else
  4229. /*
  4230. * On 32-bit, the Normal zone needs to be preserved for allocations accessible
  4231. * by the kernel. If processes running on node 0 deplete the low memory zone
  4232. * then reclaim will occur more frequency increasing stalls and potentially
  4233. * be easier to OOM if a large percentage of the zone is under writeback or
  4234. * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
  4235. * Hence, default to zone ordering on 32-bit.
  4236. */
  4237. static int default_zonelist_order(void)
  4238. {
  4239. return ZONELIST_ORDER_ZONE;
  4240. }
  4241. #endif /* CONFIG_64BIT */
  4242. static void set_zonelist_order(void)
  4243. {
  4244. if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
  4245. current_zonelist_order = default_zonelist_order();
  4246. else
  4247. current_zonelist_order = user_zonelist_order;
  4248. }
  4249. static void build_zonelists(pg_data_t *pgdat)
  4250. {
  4251. int i, node, load;
  4252. nodemask_t used_mask;
  4253. int local_node, prev_node;
  4254. struct zonelist *zonelist;
  4255. unsigned int order = current_zonelist_order;
  4256. /* initialize zonelists */
  4257. for (i = 0; i < MAX_ZONELISTS; i++) {
  4258. zonelist = pgdat->node_zonelists + i;
  4259. zonelist->_zonerefs[0].zone = NULL;
  4260. zonelist->_zonerefs[0].zone_idx = 0;
  4261. }
  4262. /* NUMA-aware ordering of nodes */
  4263. local_node = pgdat->node_id;
  4264. load = nr_online_nodes;
  4265. prev_node = local_node;
  4266. nodes_clear(used_mask);
  4267. memset(node_order, 0, sizeof(node_order));
  4268. i = 0;
  4269. while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
  4270. /*
  4271. * We don't want to pressure a particular node.
  4272. * So adding penalty to the first node in same
  4273. * distance group to make it round-robin.
  4274. */
  4275. if (node_distance(local_node, node) !=
  4276. node_distance(local_node, prev_node))
  4277. node_load[node] = load;
  4278. prev_node = node;
  4279. load--;
  4280. if (order == ZONELIST_ORDER_NODE)
  4281. build_zonelists_in_node_order(pgdat, node);
  4282. else
  4283. node_order[i++] = node; /* remember order */
  4284. }
  4285. if (order == ZONELIST_ORDER_ZONE) {
  4286. /* calculate node order -- i.e., DMA last! */
  4287. build_zonelists_in_zone_order(pgdat, i);
  4288. }
  4289. build_thisnode_zonelists(pgdat);
  4290. }
  4291. #ifdef CONFIG_HAVE_MEMORYLESS_NODES
  4292. /*
  4293. * Return node id of node used for "local" allocations.
  4294. * I.e., first node id of first zone in arg node's generic zonelist.
  4295. * Used for initializing percpu 'numa_mem', which is used primarily
  4296. * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
  4297. */
  4298. int local_memory_node(int node)
  4299. {
  4300. struct zoneref *z;
  4301. z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
  4302. gfp_zone(GFP_KERNEL),
  4303. NULL);
  4304. return z->zone->node;
  4305. }
  4306. #endif
  4307. static void setup_min_unmapped_ratio(void);
  4308. static void setup_min_slab_ratio(void);
  4309. #else /* CONFIG_NUMA */
  4310. static void set_zonelist_order(void)
  4311. {
  4312. current_zonelist_order = ZONELIST_ORDER_ZONE;
  4313. }
  4314. static void build_zonelists(pg_data_t *pgdat)
  4315. {
  4316. int node, local_node;
  4317. enum zone_type j;
  4318. struct zonelist *zonelist;
  4319. local_node = pgdat->node_id;
  4320. zonelist = &pgdat->node_zonelists[ZONELIST_FALLBACK];
  4321. j = build_zonelists_node(pgdat, zonelist, 0);
  4322. /*
  4323. * Now we build the zonelist so that it contains the zones
  4324. * of all the other nodes.
  4325. * We don't want to pressure a particular node, so when
  4326. * building the zones for node N, we make sure that the
  4327. * zones coming right after the local ones are those from
  4328. * node N+1 (modulo N)
  4329. */
  4330. for (node = local_node + 1; node < MAX_NUMNODES; node++) {
  4331. if (!node_online(node))
  4332. continue;
  4333. j = build_zonelists_node(NODE_DATA(node), zonelist, j);
  4334. }
  4335. for (node = 0; node < local_node; node++) {
  4336. if (!node_online(node))
  4337. continue;
  4338. j = build_zonelists_node(NODE_DATA(node), zonelist, j);
  4339. }
  4340. zonelist->_zonerefs[j].zone = NULL;
  4341. zonelist->_zonerefs[j].zone_idx = 0;
  4342. }
  4343. #endif /* CONFIG_NUMA */
  4344. /*
  4345. * Boot pageset table. One per cpu which is going to be used for all
  4346. * zones and all nodes. The parameters will be set in such a way
  4347. * that an item put on a list will immediately be handed over to
  4348. * the buddy list. This is safe since pageset manipulation is done
  4349. * with interrupts disabled.
  4350. *
  4351. * The boot_pagesets must be kept even after bootup is complete for
  4352. * unused processors and/or zones. They do play a role for bootstrapping
  4353. * hotplugged processors.
  4354. *
  4355. * zoneinfo_show() and maybe other functions do
  4356. * not check if the processor is online before following the pageset pointer.
  4357. * Other parts of the kernel may not check if the zone is available.
  4358. */
  4359. static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
  4360. static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
  4361. static void setup_zone_pageset(struct zone *zone);
  4362. /*
  4363. * Global mutex to protect against size modification of zonelists
  4364. * as well as to serialize pageset setup for the new populated zone.
  4365. */
  4366. DEFINE_MUTEX(zonelists_mutex);
  4367. /* return values int ....just for stop_machine() */
  4368. static int __build_all_zonelists(void *data)
  4369. {
  4370. int nid;
  4371. int cpu;
  4372. pg_data_t *self = data;
  4373. #ifdef CONFIG_NUMA
  4374. memset(node_load, 0, sizeof(node_load));
  4375. #endif
  4376. if (self && !node_online(self->node_id)) {
  4377. build_zonelists(self);
  4378. }
  4379. for_each_online_node(nid) {
  4380. pg_data_t *pgdat = NODE_DATA(nid);
  4381. build_zonelists(pgdat);
  4382. }
  4383. /*
  4384. * Initialize the boot_pagesets that are going to be used
  4385. * for bootstrapping processors. The real pagesets for
  4386. * each zone will be allocated later when the per cpu
  4387. * allocator is available.
  4388. *
  4389. * boot_pagesets are used also for bootstrapping offline
  4390. * cpus if the system is already booted because the pagesets
  4391. * are needed to initialize allocators on a specific cpu too.
  4392. * F.e. the percpu allocator needs the page allocator which
  4393. * needs the percpu allocator in order to allocate its pagesets
  4394. * (a chicken-egg dilemma).
  4395. */
  4396. for_each_possible_cpu(cpu) {
  4397. setup_pageset(&per_cpu(boot_pageset, cpu), 0);
  4398. #ifdef CONFIG_HAVE_MEMORYLESS_NODES
  4399. /*
  4400. * We now know the "local memory node" for each node--
  4401. * i.e., the node of the first zone in the generic zonelist.
  4402. * Set up numa_mem percpu variable for on-line cpus. During
  4403. * boot, only the boot cpu should be on-line; we'll init the
  4404. * secondary cpus' numa_mem as they come on-line. During
  4405. * node/memory hotplug, we'll fixup all on-line cpus.
  4406. */
  4407. if (cpu_online(cpu))
  4408. set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
  4409. #endif
  4410. }
  4411. return 0;
  4412. }
  4413. static noinline void __init
  4414. build_all_zonelists_init(void)
  4415. {
  4416. __build_all_zonelists(NULL);
  4417. mminit_verify_zonelist();
  4418. cpuset_init_current_mems_allowed();
  4419. }
  4420. /*
  4421. * Called with zonelists_mutex held always
  4422. * unless system_state == SYSTEM_BOOTING.
  4423. *
  4424. * __ref due to (1) call of __meminit annotated setup_zone_pageset
  4425. * [we're only called with non-NULL zone through __meminit paths] and
  4426. * (2) call of __init annotated helper build_all_zonelists_init
  4427. * [protected by SYSTEM_BOOTING].
  4428. */
  4429. void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone)
  4430. {
  4431. set_zonelist_order();
  4432. if (system_state == SYSTEM_BOOTING) {
  4433. build_all_zonelists_init();
  4434. } else {
  4435. #ifdef CONFIG_MEMORY_HOTPLUG
  4436. if (zone)
  4437. setup_zone_pageset(zone);
  4438. #endif
  4439. /* we have to stop all cpus to guarantee there is no user
  4440. of zonelist */
  4441. stop_machine(__build_all_zonelists, pgdat, NULL);
  4442. /* cpuset refresh routine should be here */
  4443. }
  4444. vm_total_pages = nr_free_pagecache_pages();
  4445. /*
  4446. * Disable grouping by mobility if the number of pages in the
  4447. * system is too low to allow the mechanism to work. It would be
  4448. * more accurate, but expensive to check per-zone. This check is
  4449. * made on memory-hotadd so a system can start with mobility
  4450. * disabled and enable it later
  4451. */
  4452. if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
  4453. page_group_by_mobility_disabled = 1;
  4454. else
  4455. page_group_by_mobility_disabled = 0;
  4456. pr_info("Built %i zonelists in %s order, mobility grouping %s. Total pages: %ld\n",
  4457. nr_online_nodes,
  4458. zonelist_order_name[current_zonelist_order],
  4459. page_group_by_mobility_disabled ? "off" : "on",
  4460. vm_total_pages);
  4461. #ifdef CONFIG_NUMA
  4462. pr_info("Policy zone: %s\n", zone_names[policy_zone]);
  4463. #endif
  4464. }
  4465. /*
  4466. * Initially all pages are reserved - free ones are freed
  4467. * up by free_all_bootmem() once the early boot process is
  4468. * done. Non-atomic initialization, single-pass.
  4469. */
  4470. void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
  4471. unsigned long start_pfn, enum memmap_context context)
  4472. {
  4473. struct vmem_altmap *altmap = to_vmem_altmap(__pfn_to_phys(start_pfn));
  4474. unsigned long end_pfn = start_pfn + size;
  4475. pg_data_t *pgdat = NODE_DATA(nid);
  4476. unsigned long pfn;
  4477. unsigned long nr_initialised = 0;
  4478. #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
  4479. struct memblock_region *r = NULL, *tmp;
  4480. #endif
  4481. if (highest_memmap_pfn < end_pfn - 1)
  4482. highest_memmap_pfn = end_pfn - 1;
  4483. /*
  4484. * Honor reservation requested by the driver for this ZONE_DEVICE
  4485. * memory
  4486. */
  4487. if (altmap && start_pfn == altmap->base_pfn)
  4488. start_pfn += altmap->reserve;
  4489. for (pfn = start_pfn; pfn < end_pfn; pfn++) {
  4490. /*
  4491. * There can be holes in boot-time mem_map[]s handed to this
  4492. * function. They do not exist on hotplugged memory.
  4493. */
  4494. if (context != MEMMAP_EARLY)
  4495. goto not_early;
  4496. if (!early_pfn_valid(pfn)) {
  4497. #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
  4498. /*
  4499. * Skip to the pfn preceding the next valid one (or
  4500. * end_pfn), such that we hit a valid pfn (or end_pfn)
  4501. * on our next iteration of the loop.
  4502. */
  4503. pfn = memblock_next_valid_pfn(pfn, end_pfn) - 1;
  4504. #endif
  4505. continue;
  4506. }
  4507. if (!early_pfn_in_nid(pfn, nid))
  4508. continue;
  4509. if (!update_defer_init(pgdat, pfn, end_pfn, &nr_initialised))
  4510. break;
  4511. #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
  4512. /*
  4513. * Check given memblock attribute by firmware which can affect
  4514. * kernel memory layout. If zone==ZONE_MOVABLE but memory is
  4515. * mirrored, it's an overlapped memmap init. skip it.
  4516. */
  4517. if (mirrored_kernelcore && zone == ZONE_MOVABLE) {
  4518. if (!r || pfn >= memblock_region_memory_end_pfn(r)) {
  4519. for_each_memblock(memory, tmp)
  4520. if (pfn < memblock_region_memory_end_pfn(tmp))
  4521. break;
  4522. r = tmp;
  4523. }
  4524. if (pfn >= memblock_region_memory_base_pfn(r) &&
  4525. memblock_is_mirror(r)) {
  4526. /* already initialized as NORMAL */
  4527. pfn = memblock_region_memory_end_pfn(r);
  4528. continue;
  4529. }
  4530. }
  4531. #endif
  4532. not_early:
  4533. /*
  4534. * Mark the block movable so that blocks are reserved for
  4535. * movable at startup. This will force kernel allocations
  4536. * to reserve their blocks rather than leaking throughout
  4537. * the address space during boot when many long-lived
  4538. * kernel allocations are made.
  4539. *
  4540. * bitmap is created for zone's valid pfn range. but memmap
  4541. * can be created for invalid pages (for alignment)
  4542. * check here not to call set_pageblock_migratetype() against
  4543. * pfn out of zone.
  4544. */
  4545. if (!(pfn & (pageblock_nr_pages - 1))) {
  4546. struct page *page = pfn_to_page(pfn);
  4547. __init_single_page(page, pfn, zone, nid);
  4548. set_pageblock_migratetype(page, MIGRATE_MOVABLE);
  4549. } else {
  4550. __init_single_pfn(pfn, zone, nid);
  4551. }
  4552. }
  4553. }
  4554. static void __meminit zone_init_free_lists(struct zone *zone)
  4555. {
  4556. unsigned int order, t;
  4557. for_each_migratetype_order(order, t) {
  4558. INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
  4559. zone->free_area[order].nr_free = 0;
  4560. }
  4561. }
  4562. #ifndef __HAVE_ARCH_MEMMAP_INIT
  4563. #define memmap_init(size, nid, zone, start_pfn) \
  4564. memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
  4565. #endif
  4566. static int zone_batchsize(struct zone *zone)
  4567. {
  4568. #ifdef CONFIG_MMU
  4569. int batch;
  4570. /*
  4571. * The per-cpu-pages pools are set to around 1000th of the
  4572. * size of the zone. But no more than 1/2 of a meg.
  4573. *
  4574. * OK, so we don't know how big the cache is. So guess.
  4575. */
  4576. batch = zone->managed_pages / 1024;
  4577. if (batch * PAGE_SIZE > 512 * 1024)
  4578. batch = (512 * 1024) / PAGE_SIZE;
  4579. batch /= 4; /* We effectively *= 4 below */
  4580. if (batch < 1)
  4581. batch = 1;
  4582. /*
  4583. * Clamp the batch to a 2^n - 1 value. Having a power
  4584. * of 2 value was found to be more likely to have
  4585. * suboptimal cache aliasing properties in some cases.
  4586. *
  4587. * For example if 2 tasks are alternately allocating
  4588. * batches of pages, one task can end up with a lot
  4589. * of pages of one half of the possible page colors
  4590. * and the other with pages of the other colors.
  4591. */
  4592. batch = rounddown_pow_of_two(batch + batch/2) - 1;
  4593. return batch;
  4594. #else
  4595. /* The deferral and batching of frees should be suppressed under NOMMU
  4596. * conditions.
  4597. *
  4598. * The problem is that NOMMU needs to be able to allocate large chunks
  4599. * of contiguous memory as there's no hardware page translation to
  4600. * assemble apparent contiguous memory from discontiguous pages.
  4601. *
  4602. * Queueing large contiguous runs of pages for batching, however,
  4603. * causes the pages to actually be freed in smaller chunks. As there
  4604. * can be a significant delay between the individual batches being
  4605. * recycled, this leads to the once large chunks of space being
  4606. * fragmented and becoming unavailable for high-order allocations.
  4607. */
  4608. return 0;
  4609. #endif
  4610. }
  4611. /*
  4612. * pcp->high and pcp->batch values are related and dependent on one another:
  4613. * ->batch must never be higher then ->high.
  4614. * The following function updates them in a safe manner without read side
  4615. * locking.
  4616. *
  4617. * Any new users of pcp->batch and pcp->high should ensure they can cope with
  4618. * those fields changing asynchronously (acording the the above rule).
  4619. *
  4620. * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
  4621. * outside of boot time (or some other assurance that no concurrent updaters
  4622. * exist).
  4623. */
  4624. static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
  4625. unsigned long batch)
  4626. {
  4627. /* start with a fail safe value for batch */
  4628. pcp->batch = 1;
  4629. smp_wmb();
  4630. /* Update high, then batch, in order */
  4631. pcp->high = high;
  4632. smp_wmb();
  4633. pcp->batch = batch;
  4634. }
  4635. /* a companion to pageset_set_high() */
  4636. static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
  4637. {
  4638. pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
  4639. }
  4640. static void pageset_init(struct per_cpu_pageset *p)
  4641. {
  4642. struct per_cpu_pages *pcp;
  4643. int migratetype;
  4644. memset(p, 0, sizeof(*p));
  4645. pcp = &p->pcp;
  4646. pcp->count = 0;
  4647. for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
  4648. INIT_LIST_HEAD(&pcp->lists[migratetype]);
  4649. }
  4650. static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
  4651. {
  4652. pageset_init(p);
  4653. pageset_set_batch(p, batch);
  4654. }
  4655. /*
  4656. * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
  4657. * to the value high for the pageset p.
  4658. */
  4659. static void pageset_set_high(struct per_cpu_pageset *p,
  4660. unsigned long high)
  4661. {
  4662. unsigned long batch = max(1UL, high / 4);
  4663. if ((high / 4) > (PAGE_SHIFT * 8))
  4664. batch = PAGE_SHIFT * 8;
  4665. pageset_update(&p->pcp, high, batch);
  4666. }
  4667. static void pageset_set_high_and_batch(struct zone *zone,
  4668. struct per_cpu_pageset *pcp)
  4669. {
  4670. if (percpu_pagelist_fraction)
  4671. pageset_set_high(pcp,
  4672. (zone->managed_pages /
  4673. percpu_pagelist_fraction));
  4674. else
  4675. pageset_set_batch(pcp, zone_batchsize(zone));
  4676. }
  4677. static void __meminit zone_pageset_init(struct zone *zone, int cpu)
  4678. {
  4679. struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
  4680. pageset_init(pcp);
  4681. pageset_set_high_and_batch(zone, pcp);
  4682. }
  4683. static void __meminit setup_zone_pageset(struct zone *zone)
  4684. {
  4685. int cpu;
  4686. zone->pageset = alloc_percpu(struct per_cpu_pageset);
  4687. for_each_possible_cpu(cpu)
  4688. zone_pageset_init(zone, cpu);
  4689. }
  4690. /*
  4691. * Allocate per cpu pagesets and initialize them.
  4692. * Before this call only boot pagesets were available.
  4693. */
  4694. void __init setup_per_cpu_pageset(void)
  4695. {
  4696. struct pglist_data *pgdat;
  4697. struct zone *zone;
  4698. for_each_populated_zone(zone)
  4699. setup_zone_pageset(zone);
  4700. for_each_online_pgdat(pgdat)
  4701. pgdat->per_cpu_nodestats =
  4702. alloc_percpu(struct per_cpu_nodestat);
  4703. }
  4704. static __meminit void zone_pcp_init(struct zone *zone)
  4705. {
  4706. /*
  4707. * per cpu subsystem is not up at this point. The following code
  4708. * relies on the ability of the linker to provide the
  4709. * offset of a (static) per cpu variable into the per cpu area.
  4710. */
  4711. zone->pageset = &boot_pageset;
  4712. if (populated_zone(zone))
  4713. printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
  4714. zone->name, zone->present_pages,
  4715. zone_batchsize(zone));
  4716. }
  4717. int __meminit init_currently_empty_zone(struct zone *zone,
  4718. unsigned long zone_start_pfn,
  4719. unsigned long size)
  4720. {
  4721. struct pglist_data *pgdat = zone->zone_pgdat;
  4722. pgdat->nr_zones = zone_idx(zone) + 1;
  4723. zone->zone_start_pfn = zone_start_pfn;
  4724. mminit_dprintk(MMINIT_TRACE, "memmap_init",
  4725. "Initialising map node %d zone %lu pfns %lu -> %lu\n",
  4726. pgdat->node_id,
  4727. (unsigned long)zone_idx(zone),
  4728. zone_start_pfn, (zone_start_pfn + size));
  4729. zone_init_free_lists(zone);
  4730. zone->initialized = 1;
  4731. return 0;
  4732. }
  4733. #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
  4734. #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
  4735. /*
  4736. * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
  4737. */
  4738. int __meminit __early_pfn_to_nid(unsigned long pfn,
  4739. struct mminit_pfnnid_cache *state)
  4740. {
  4741. unsigned long start_pfn, end_pfn;
  4742. int nid;
  4743. if (state->last_start <= pfn && pfn < state->last_end)
  4744. return state->last_nid;
  4745. nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
  4746. if (nid != -1) {
  4747. state->last_start = start_pfn;
  4748. state->last_end = end_pfn;
  4749. state->last_nid = nid;
  4750. }
  4751. return nid;
  4752. }
  4753. #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
  4754. /**
  4755. * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
  4756. * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
  4757. * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
  4758. *
  4759. * If an architecture guarantees that all ranges registered contain no holes
  4760. * and may be freed, this this function may be used instead of calling
  4761. * memblock_free_early_nid() manually.
  4762. */
  4763. void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
  4764. {
  4765. unsigned long start_pfn, end_pfn;
  4766. int i, this_nid;
  4767. for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
  4768. start_pfn = min(start_pfn, max_low_pfn);
  4769. end_pfn = min(end_pfn, max_low_pfn);
  4770. if (start_pfn < end_pfn)
  4771. memblock_free_early_nid(PFN_PHYS(start_pfn),
  4772. (end_pfn - start_pfn) << PAGE_SHIFT,
  4773. this_nid);
  4774. }
  4775. }
  4776. /**
  4777. * sparse_memory_present_with_active_regions - Call memory_present for each active range
  4778. * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
  4779. *
  4780. * If an architecture guarantees that all ranges registered contain no holes and may
  4781. * be freed, this function may be used instead of calling memory_present() manually.
  4782. */
  4783. void __init sparse_memory_present_with_active_regions(int nid)
  4784. {
  4785. unsigned long start_pfn, end_pfn;
  4786. int i, this_nid;
  4787. for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
  4788. memory_present(this_nid, start_pfn, end_pfn);
  4789. }
  4790. /**
  4791. * get_pfn_range_for_nid - Return the start and end page frames for a node
  4792. * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
  4793. * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
  4794. * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
  4795. *
  4796. * It returns the start and end page frame of a node based on information
  4797. * provided by memblock_set_node(). If called for a node
  4798. * with no available memory, a warning is printed and the start and end
  4799. * PFNs will be 0.
  4800. */
  4801. void __meminit get_pfn_range_for_nid(unsigned int nid,
  4802. unsigned long *start_pfn, unsigned long *end_pfn)
  4803. {
  4804. unsigned long this_start_pfn, this_end_pfn;
  4805. int i;
  4806. *start_pfn = -1UL;
  4807. *end_pfn = 0;
  4808. for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
  4809. *start_pfn = min(*start_pfn, this_start_pfn);
  4810. *end_pfn = max(*end_pfn, this_end_pfn);
  4811. }
  4812. if (*start_pfn == -1UL)
  4813. *start_pfn = 0;
  4814. }
  4815. /*
  4816. * This finds a zone that can be used for ZONE_MOVABLE pages. The
  4817. * assumption is made that zones within a node are ordered in monotonic
  4818. * increasing memory addresses so that the "highest" populated zone is used
  4819. */
  4820. static void __init find_usable_zone_for_movable(void)
  4821. {
  4822. int zone_index;
  4823. for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
  4824. if (zone_index == ZONE_MOVABLE)
  4825. continue;
  4826. if (arch_zone_highest_possible_pfn[zone_index] >
  4827. arch_zone_lowest_possible_pfn[zone_index])
  4828. break;
  4829. }
  4830. VM_BUG_ON(zone_index == -1);
  4831. movable_zone = zone_index;
  4832. }
  4833. /*
  4834. * The zone ranges provided by the architecture do not include ZONE_MOVABLE
  4835. * because it is sized independent of architecture. Unlike the other zones,
  4836. * the starting point for ZONE_MOVABLE is not fixed. It may be different
  4837. * in each node depending on the size of each node and how evenly kernelcore
  4838. * is distributed. This helper function adjusts the zone ranges
  4839. * provided by the architecture for a given node by using the end of the
  4840. * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
  4841. * zones within a node are in order of monotonic increases memory addresses
  4842. */
  4843. static void __meminit adjust_zone_range_for_zone_movable(int nid,
  4844. unsigned long zone_type,
  4845. unsigned long node_start_pfn,
  4846. unsigned long node_end_pfn,
  4847. unsigned long *zone_start_pfn,
  4848. unsigned long *zone_end_pfn)
  4849. {
  4850. /* Only adjust if ZONE_MOVABLE is on this node */
  4851. if (zone_movable_pfn[nid]) {
  4852. /* Size ZONE_MOVABLE */
  4853. if (zone_type == ZONE_MOVABLE) {
  4854. *zone_start_pfn = zone_movable_pfn[nid];
  4855. *zone_end_pfn = min(node_end_pfn,
  4856. arch_zone_highest_possible_pfn[movable_zone]);
  4857. /* Adjust for ZONE_MOVABLE starting within this range */
  4858. } else if (!mirrored_kernelcore &&
  4859. *zone_start_pfn < zone_movable_pfn[nid] &&
  4860. *zone_end_pfn > zone_movable_pfn[nid]) {
  4861. *zone_end_pfn = zone_movable_pfn[nid];
  4862. /* Check if this whole range is within ZONE_MOVABLE */
  4863. } else if (*zone_start_pfn >= zone_movable_pfn[nid])
  4864. *zone_start_pfn = *zone_end_pfn;
  4865. }
  4866. }
  4867. /*
  4868. * Return the number of pages a zone spans in a node, including holes
  4869. * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
  4870. */
  4871. static unsigned long __meminit zone_spanned_pages_in_node(int nid,
  4872. unsigned long zone_type,
  4873. unsigned long node_start_pfn,
  4874. unsigned long node_end_pfn,
  4875. unsigned long *zone_start_pfn,
  4876. unsigned long *zone_end_pfn,
  4877. unsigned long *ignored)
  4878. {
  4879. /* When hotadd a new node from cpu_up(), the node should be empty */
  4880. if (!node_start_pfn && !node_end_pfn)
  4881. return 0;
  4882. /* Get the start and end of the zone */
  4883. *zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
  4884. *zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
  4885. adjust_zone_range_for_zone_movable(nid, zone_type,
  4886. node_start_pfn, node_end_pfn,
  4887. zone_start_pfn, zone_end_pfn);
  4888. /* Check that this node has pages within the zone's required range */
  4889. if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn)
  4890. return 0;
  4891. /* Move the zone boundaries inside the node if necessary */
  4892. *zone_end_pfn = min(*zone_end_pfn, node_end_pfn);
  4893. *zone_start_pfn = max(*zone_start_pfn, node_start_pfn);
  4894. /* Return the spanned pages */
  4895. return *zone_end_pfn - *zone_start_pfn;
  4896. }
  4897. /*
  4898. * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
  4899. * then all holes in the requested range will be accounted for.
  4900. */
  4901. unsigned long __meminit __absent_pages_in_range(int nid,
  4902. unsigned long range_start_pfn,
  4903. unsigned long range_end_pfn)
  4904. {
  4905. unsigned long nr_absent = range_end_pfn - range_start_pfn;
  4906. unsigned long start_pfn, end_pfn;
  4907. int i;
  4908. for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
  4909. start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
  4910. end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
  4911. nr_absent -= end_pfn - start_pfn;
  4912. }
  4913. return nr_absent;
  4914. }
  4915. /**
  4916. * absent_pages_in_range - Return number of page frames in holes within a range
  4917. * @start_pfn: The start PFN to start searching for holes
  4918. * @end_pfn: The end PFN to stop searching for holes
  4919. *
  4920. * It returns the number of pages frames in memory holes within a range.
  4921. */
  4922. unsigned long __init absent_pages_in_range(unsigned long start_pfn,
  4923. unsigned long end_pfn)
  4924. {
  4925. return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
  4926. }
  4927. /* Return the number of page frames in holes in a zone on a node */
  4928. static unsigned long __meminit zone_absent_pages_in_node(int nid,
  4929. unsigned long zone_type,
  4930. unsigned long node_start_pfn,
  4931. unsigned long node_end_pfn,
  4932. unsigned long *ignored)
  4933. {
  4934. unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
  4935. unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
  4936. unsigned long zone_start_pfn, zone_end_pfn;
  4937. unsigned long nr_absent;
  4938. /* When hotadd a new node from cpu_up(), the node should be empty */
  4939. if (!node_start_pfn && !node_end_pfn)
  4940. return 0;
  4941. zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
  4942. zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
  4943. adjust_zone_range_for_zone_movable(nid, zone_type,
  4944. node_start_pfn, node_end_pfn,
  4945. &zone_start_pfn, &zone_end_pfn);
  4946. nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
  4947. /*
  4948. * ZONE_MOVABLE handling.
  4949. * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages
  4950. * and vice versa.
  4951. */
  4952. if (mirrored_kernelcore && zone_movable_pfn[nid]) {
  4953. unsigned long start_pfn, end_pfn;
  4954. struct memblock_region *r;
  4955. for_each_memblock(memory, r) {
  4956. start_pfn = clamp(memblock_region_memory_base_pfn(r),
  4957. zone_start_pfn, zone_end_pfn);
  4958. end_pfn = clamp(memblock_region_memory_end_pfn(r),
  4959. zone_start_pfn, zone_end_pfn);
  4960. if (zone_type == ZONE_MOVABLE &&
  4961. memblock_is_mirror(r))
  4962. nr_absent += end_pfn - start_pfn;
  4963. if (zone_type == ZONE_NORMAL &&
  4964. !memblock_is_mirror(r))
  4965. nr_absent += end_pfn - start_pfn;
  4966. }
  4967. }
  4968. return nr_absent;
  4969. }
  4970. #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
  4971. static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
  4972. unsigned long zone_type,
  4973. unsigned long node_start_pfn,
  4974. unsigned long node_end_pfn,
  4975. unsigned long *zone_start_pfn,
  4976. unsigned long *zone_end_pfn,
  4977. unsigned long *zones_size)
  4978. {
  4979. unsigned int zone;
  4980. *zone_start_pfn = node_start_pfn;
  4981. for (zone = 0; zone < zone_type; zone++)
  4982. *zone_start_pfn += zones_size[zone];
  4983. *zone_end_pfn = *zone_start_pfn + zones_size[zone_type];
  4984. return zones_size[zone_type];
  4985. }
  4986. static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
  4987. unsigned long zone_type,
  4988. unsigned long node_start_pfn,
  4989. unsigned long node_end_pfn,
  4990. unsigned long *zholes_size)
  4991. {
  4992. if (!zholes_size)
  4993. return 0;
  4994. return zholes_size[zone_type];
  4995. }
  4996. #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
  4997. static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
  4998. unsigned long node_start_pfn,
  4999. unsigned long node_end_pfn,
  5000. unsigned long *zones_size,
  5001. unsigned long *zholes_size)
  5002. {
  5003. unsigned long realtotalpages = 0, totalpages = 0;
  5004. enum zone_type i;
  5005. for (i = 0; i < MAX_NR_ZONES; i++) {
  5006. struct zone *zone = pgdat->node_zones + i;
  5007. unsigned long zone_start_pfn, zone_end_pfn;
  5008. unsigned long size, real_size;
  5009. size = zone_spanned_pages_in_node(pgdat->node_id, i,
  5010. node_start_pfn,
  5011. node_end_pfn,
  5012. &zone_start_pfn,
  5013. &zone_end_pfn,
  5014. zones_size);
  5015. real_size = size - zone_absent_pages_in_node(pgdat->node_id, i,
  5016. node_start_pfn, node_end_pfn,
  5017. zholes_size);
  5018. if (size)
  5019. zone->zone_start_pfn = zone_start_pfn;
  5020. else
  5021. zone->zone_start_pfn = 0;
  5022. zone->spanned_pages = size;
  5023. zone->present_pages = real_size;
  5024. totalpages += size;
  5025. realtotalpages += real_size;
  5026. }
  5027. pgdat->node_spanned_pages = totalpages;
  5028. pgdat->node_present_pages = realtotalpages;
  5029. printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
  5030. realtotalpages);
  5031. }
  5032. #ifndef CONFIG_SPARSEMEM
  5033. /*
  5034. * Calculate the size of the zone->blockflags rounded to an unsigned long
  5035. * Start by making sure zonesize is a multiple of pageblock_order by rounding
  5036. * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
  5037. * round what is now in bits to nearest long in bits, then return it in
  5038. * bytes.
  5039. */
  5040. static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
  5041. {
  5042. unsigned long usemapsize;
  5043. zonesize += zone_start_pfn & (pageblock_nr_pages-1);
  5044. usemapsize = roundup(zonesize, pageblock_nr_pages);
  5045. usemapsize = usemapsize >> pageblock_order;
  5046. usemapsize *= NR_PAGEBLOCK_BITS;
  5047. usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
  5048. return usemapsize / 8;
  5049. }
  5050. static void __init setup_usemap(struct pglist_data *pgdat,
  5051. struct zone *zone,
  5052. unsigned long zone_start_pfn,
  5053. unsigned long zonesize)
  5054. {
  5055. unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
  5056. zone->pageblock_flags = NULL;
  5057. if (usemapsize)
  5058. zone->pageblock_flags =
  5059. memblock_virt_alloc_node_nopanic(usemapsize,
  5060. pgdat->node_id);
  5061. }
  5062. #else
  5063. static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
  5064. unsigned long zone_start_pfn, unsigned long zonesize) {}
  5065. #endif /* CONFIG_SPARSEMEM */
  5066. #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
  5067. /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
  5068. void __paginginit set_pageblock_order(void)
  5069. {
  5070. unsigned int order;
  5071. /* Check that pageblock_nr_pages has not already been setup */
  5072. if (pageblock_order)
  5073. return;
  5074. if (HPAGE_SHIFT > PAGE_SHIFT)
  5075. order = HUGETLB_PAGE_ORDER;
  5076. else
  5077. order = MAX_ORDER - 1;
  5078. /*
  5079. * Assume the largest contiguous order of interest is a huge page.
  5080. * This value may be variable depending on boot parameters on IA64 and
  5081. * powerpc.
  5082. */
  5083. pageblock_order = order;
  5084. }
  5085. #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
  5086. /*
  5087. * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
  5088. * is unused as pageblock_order is set at compile-time. See
  5089. * include/linux/pageblock-flags.h for the values of pageblock_order based on
  5090. * the kernel config
  5091. */
  5092. void __paginginit set_pageblock_order(void)
  5093. {
  5094. }
  5095. #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
  5096. static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages,
  5097. unsigned long present_pages)
  5098. {
  5099. unsigned long pages = spanned_pages;
  5100. /*
  5101. * Provide a more accurate estimation if there are holes within
  5102. * the zone and SPARSEMEM is in use. If there are holes within the
  5103. * zone, each populated memory region may cost us one or two extra
  5104. * memmap pages due to alignment because memmap pages for each
  5105. * populated regions may not be naturally aligned on page boundary.
  5106. * So the (present_pages >> 4) heuristic is a tradeoff for that.
  5107. */
  5108. if (spanned_pages > present_pages + (present_pages >> 4) &&
  5109. IS_ENABLED(CONFIG_SPARSEMEM))
  5110. pages = present_pages;
  5111. return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
  5112. }
  5113. /*
  5114. * Set up the zone data structures:
  5115. * - mark all pages reserved
  5116. * - mark all memory queues empty
  5117. * - clear the memory bitmaps
  5118. *
  5119. * NOTE: pgdat should get zeroed by caller.
  5120. */
  5121. static void __paginginit free_area_init_core(struct pglist_data *pgdat)
  5122. {
  5123. enum zone_type j;
  5124. int nid = pgdat->node_id;
  5125. int ret;
  5126. pgdat_resize_init(pgdat);
  5127. #ifdef CONFIG_NUMA_BALANCING
  5128. spin_lock_init(&pgdat->numabalancing_migrate_lock);
  5129. pgdat->numabalancing_migrate_nr_pages = 0;
  5130. pgdat->numabalancing_migrate_next_window = jiffies;
  5131. #endif
  5132. #ifdef CONFIG_TRANSPARENT_HUGEPAGE
  5133. spin_lock_init(&pgdat->split_queue_lock);
  5134. INIT_LIST_HEAD(&pgdat->split_queue);
  5135. pgdat->split_queue_len = 0;
  5136. #endif
  5137. init_waitqueue_head(&pgdat->kswapd_wait);
  5138. init_waitqueue_head(&pgdat->pfmemalloc_wait);
  5139. #ifdef CONFIG_COMPACTION
  5140. init_waitqueue_head(&pgdat->kcompactd_wait);
  5141. #endif
  5142. pgdat_page_ext_init(pgdat);
  5143. spin_lock_init(&pgdat->lru_lock);
  5144. lruvec_init(node_lruvec(pgdat));
  5145. for (j = 0; j < MAX_NR_ZONES; j++) {
  5146. struct zone *zone = pgdat->node_zones + j;
  5147. unsigned long size, realsize, freesize, memmap_pages;
  5148. unsigned long zone_start_pfn = zone->zone_start_pfn;
  5149. size = zone->spanned_pages;
  5150. realsize = freesize = zone->present_pages;
  5151. /*
  5152. * Adjust freesize so that it accounts for how much memory
  5153. * is used by this zone for memmap. This affects the watermark
  5154. * and per-cpu initialisations
  5155. */
  5156. memmap_pages = calc_memmap_size(size, realsize);
  5157. if (!is_highmem_idx(j)) {
  5158. if (freesize >= memmap_pages) {
  5159. freesize -= memmap_pages;
  5160. if (memmap_pages)
  5161. printk(KERN_DEBUG
  5162. " %s zone: %lu pages used for memmap\n",
  5163. zone_names[j], memmap_pages);
  5164. } else
  5165. pr_warn(" %s zone: %lu pages exceeds freesize %lu\n",
  5166. zone_names[j], memmap_pages, freesize);
  5167. }
  5168. /* Account for reserved pages */
  5169. if (j == 0 && freesize > dma_reserve) {
  5170. freesize -= dma_reserve;
  5171. printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
  5172. zone_names[0], dma_reserve);
  5173. }
  5174. if (!is_highmem_idx(j))
  5175. nr_kernel_pages += freesize;
  5176. /* Charge for highmem memmap if there are enough kernel pages */
  5177. else if (nr_kernel_pages > memmap_pages * 2)
  5178. nr_kernel_pages -= memmap_pages;
  5179. nr_all_pages += freesize;
  5180. /*
  5181. * Set an approximate value for lowmem here, it will be adjusted
  5182. * when the bootmem allocator frees pages into the buddy system.
  5183. * And all highmem pages will be managed by the buddy system.
  5184. */
  5185. zone->managed_pages = is_highmem_idx(j) ? realsize : freesize;
  5186. #ifdef CONFIG_NUMA
  5187. zone->node = nid;
  5188. #endif
  5189. zone->name = zone_names[j];
  5190. zone->zone_pgdat = pgdat;
  5191. spin_lock_init(&zone->lock);
  5192. zone_seqlock_init(zone);
  5193. zone_pcp_init(zone);
  5194. if (!size)
  5195. continue;
  5196. set_pageblock_order();
  5197. setup_usemap(pgdat, zone, zone_start_pfn, size);
  5198. ret = init_currently_empty_zone(zone, zone_start_pfn, size);
  5199. BUG_ON(ret);
  5200. memmap_init(size, nid, j, zone_start_pfn);
  5201. }
  5202. }
  5203. static void __ref alloc_node_mem_map(struct pglist_data *pgdat)
  5204. {
  5205. unsigned long __maybe_unused start = 0;
  5206. unsigned long __maybe_unused offset = 0;
  5207. /* Skip empty nodes */
  5208. if (!pgdat->node_spanned_pages)
  5209. return;
  5210. #ifdef CONFIG_FLAT_NODE_MEM_MAP
  5211. start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
  5212. offset = pgdat->node_start_pfn - start;
  5213. /* ia64 gets its own node_mem_map, before this, without bootmem */
  5214. if (!pgdat->node_mem_map) {
  5215. unsigned long size, end;
  5216. struct page *map;
  5217. /*
  5218. * The zone's endpoints aren't required to be MAX_ORDER
  5219. * aligned but the node_mem_map endpoints must be in order
  5220. * for the buddy allocator to function correctly.
  5221. */
  5222. end = pgdat_end_pfn(pgdat);
  5223. end = ALIGN(end, MAX_ORDER_NR_PAGES);
  5224. size = (end - start) * sizeof(struct page);
  5225. map = alloc_remap(pgdat->node_id, size);
  5226. if (!map)
  5227. map = memblock_virt_alloc_node_nopanic(size,
  5228. pgdat->node_id);
  5229. pgdat->node_mem_map = map + offset;
  5230. }
  5231. #ifndef CONFIG_NEED_MULTIPLE_NODES
  5232. /*
  5233. * With no DISCONTIG, the global mem_map is just set as node 0's
  5234. */
  5235. if (pgdat == NODE_DATA(0)) {
  5236. mem_map = NODE_DATA(0)->node_mem_map;
  5237. #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM)
  5238. if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
  5239. mem_map -= offset;
  5240. #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
  5241. }
  5242. #endif
  5243. #endif /* CONFIG_FLAT_NODE_MEM_MAP */
  5244. }
  5245. void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
  5246. unsigned long node_start_pfn, unsigned long *zholes_size)
  5247. {
  5248. pg_data_t *pgdat = NODE_DATA(nid);
  5249. unsigned long start_pfn = 0;
  5250. unsigned long end_pfn = 0;
  5251. /* pg_data_t should be reset to zero when it's allocated */
  5252. WARN_ON(pgdat->nr_zones || pgdat->kswapd_classzone_idx);
  5253. reset_deferred_meminit(pgdat);
  5254. pgdat->node_id = nid;
  5255. pgdat->node_start_pfn = node_start_pfn;
  5256. pgdat->per_cpu_nodestats = NULL;
  5257. #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
  5258. get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
  5259. pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
  5260. (u64)start_pfn << PAGE_SHIFT,
  5261. end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0);
  5262. #else
  5263. start_pfn = node_start_pfn;
  5264. #endif
  5265. calculate_node_totalpages(pgdat, start_pfn, end_pfn,
  5266. zones_size, zholes_size);
  5267. alloc_node_mem_map(pgdat);
  5268. #ifdef CONFIG_FLAT_NODE_MEM_MAP
  5269. printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
  5270. nid, (unsigned long)pgdat,
  5271. (unsigned long)pgdat->node_mem_map);
  5272. #endif
  5273. free_area_init_core(pgdat);
  5274. }
  5275. #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
  5276. #if MAX_NUMNODES > 1
  5277. /*
  5278. * Figure out the number of possible node ids.
  5279. */
  5280. void __init setup_nr_node_ids(void)
  5281. {
  5282. unsigned int highest;
  5283. highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES);
  5284. nr_node_ids = highest + 1;
  5285. }
  5286. #endif
  5287. /**
  5288. * node_map_pfn_alignment - determine the maximum internode alignment
  5289. *
  5290. * This function should be called after node map is populated and sorted.
  5291. * It calculates the maximum power of two alignment which can distinguish
  5292. * all the nodes.
  5293. *
  5294. * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
  5295. * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
  5296. * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
  5297. * shifted, 1GiB is enough and this function will indicate so.
  5298. *
  5299. * This is used to test whether pfn -> nid mapping of the chosen memory
  5300. * model has fine enough granularity to avoid incorrect mapping for the
  5301. * populated node map.
  5302. *
  5303. * Returns the determined alignment in pfn's. 0 if there is no alignment
  5304. * requirement (single node).
  5305. */
  5306. unsigned long __init node_map_pfn_alignment(void)
  5307. {
  5308. unsigned long accl_mask = 0, last_end = 0;
  5309. unsigned long start, end, mask;
  5310. int last_nid = -1;
  5311. int i, nid;
  5312. for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
  5313. if (!start || last_nid < 0 || last_nid == nid) {
  5314. last_nid = nid;
  5315. last_end = end;
  5316. continue;
  5317. }
  5318. /*
  5319. * Start with a mask granular enough to pin-point to the
  5320. * start pfn and tick off bits one-by-one until it becomes
  5321. * too coarse to separate the current node from the last.
  5322. */
  5323. mask = ~((1 << __ffs(start)) - 1);
  5324. while (mask && last_end <= (start & (mask << 1)))
  5325. mask <<= 1;
  5326. /* accumulate all internode masks */
  5327. accl_mask |= mask;
  5328. }
  5329. /* convert mask to number of pages */
  5330. return ~accl_mask + 1;
  5331. }
  5332. /* Find the lowest pfn for a node */
  5333. static unsigned long __init find_min_pfn_for_node(int nid)
  5334. {
  5335. unsigned long min_pfn = ULONG_MAX;
  5336. unsigned long start_pfn;
  5337. int i;
  5338. for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
  5339. min_pfn = min(min_pfn, start_pfn);
  5340. if (min_pfn == ULONG_MAX) {
  5341. pr_warn("Could not find start_pfn for node %d\n", nid);
  5342. return 0;
  5343. }
  5344. return min_pfn;
  5345. }
  5346. /**
  5347. * find_min_pfn_with_active_regions - Find the minimum PFN registered
  5348. *
  5349. * It returns the minimum PFN based on information provided via
  5350. * memblock_set_node().
  5351. */
  5352. unsigned long __init find_min_pfn_with_active_regions(void)
  5353. {
  5354. return find_min_pfn_for_node(MAX_NUMNODES);
  5355. }
  5356. /*
  5357. * early_calculate_totalpages()
  5358. * Sum pages in active regions for movable zone.
  5359. * Populate N_MEMORY for calculating usable_nodes.
  5360. */
  5361. static unsigned long __init early_calculate_totalpages(void)
  5362. {
  5363. unsigned long totalpages = 0;
  5364. unsigned long start_pfn, end_pfn;
  5365. int i, nid;
  5366. for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
  5367. unsigned long pages = end_pfn - start_pfn;
  5368. totalpages += pages;
  5369. if (pages)
  5370. node_set_state(nid, N_MEMORY);
  5371. }
  5372. return totalpages;
  5373. }
  5374. /*
  5375. * Find the PFN the Movable zone begins in each node. Kernel memory
  5376. * is spread evenly between nodes as long as the nodes have enough
  5377. * memory. When they don't, some nodes will have more kernelcore than
  5378. * others
  5379. */
  5380. static void __init find_zone_movable_pfns_for_nodes(void)
  5381. {
  5382. int i, nid;
  5383. unsigned long usable_startpfn;
  5384. unsigned long kernelcore_node, kernelcore_remaining;
  5385. /* save the state before borrow the nodemask */
  5386. nodemask_t saved_node_state = node_states[N_MEMORY];
  5387. unsigned long totalpages = early_calculate_totalpages();
  5388. int usable_nodes = nodes_weight(node_states[N_MEMORY]);
  5389. struct memblock_region *r;
  5390. /* Need to find movable_zone earlier when movable_node is specified. */
  5391. find_usable_zone_for_movable();
  5392. /*
  5393. * If movable_node is specified, ignore kernelcore and movablecore
  5394. * options.
  5395. */
  5396. if (movable_node_is_enabled()) {
  5397. for_each_memblock(memory, r) {
  5398. if (!memblock_is_hotpluggable(r))
  5399. continue;
  5400. nid = r->nid;
  5401. usable_startpfn = PFN_DOWN(r->base);
  5402. zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
  5403. min(usable_startpfn, zone_movable_pfn[nid]) :
  5404. usable_startpfn;
  5405. }
  5406. goto out2;
  5407. }
  5408. /*
  5409. * If kernelcore=mirror is specified, ignore movablecore option
  5410. */
  5411. if (mirrored_kernelcore) {
  5412. bool mem_below_4gb_not_mirrored = false;
  5413. for_each_memblock(memory, r) {
  5414. if (memblock_is_mirror(r))
  5415. continue;
  5416. nid = r->nid;
  5417. usable_startpfn = memblock_region_memory_base_pfn(r);
  5418. if (usable_startpfn < 0x100000) {
  5419. mem_below_4gb_not_mirrored = true;
  5420. continue;
  5421. }
  5422. zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
  5423. min(usable_startpfn, zone_movable_pfn[nid]) :
  5424. usable_startpfn;
  5425. }
  5426. if (mem_below_4gb_not_mirrored)
  5427. pr_warn("This configuration results in unmirrored kernel memory.");
  5428. goto out2;
  5429. }
  5430. /*
  5431. * If movablecore=nn[KMG] was specified, calculate what size of
  5432. * kernelcore that corresponds so that memory usable for
  5433. * any allocation type is evenly spread. If both kernelcore
  5434. * and movablecore are specified, then the value of kernelcore
  5435. * will be used for required_kernelcore if it's greater than
  5436. * what movablecore would have allowed.
  5437. */
  5438. if (required_movablecore) {
  5439. unsigned long corepages;
  5440. /*
  5441. * Round-up so that ZONE_MOVABLE is at least as large as what
  5442. * was requested by the user
  5443. */
  5444. required_movablecore =
  5445. roundup(required_movablecore, MAX_ORDER_NR_PAGES);
  5446. required_movablecore = min(totalpages, required_movablecore);
  5447. corepages = totalpages - required_movablecore;
  5448. required_kernelcore = max(required_kernelcore, corepages);
  5449. }
  5450. /*
  5451. * If kernelcore was not specified or kernelcore size is larger
  5452. * than totalpages, there is no ZONE_MOVABLE.
  5453. */
  5454. if (!required_kernelcore || required_kernelcore >= totalpages)
  5455. goto out;
  5456. /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
  5457. usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
  5458. restart:
  5459. /* Spread kernelcore memory as evenly as possible throughout nodes */
  5460. kernelcore_node = required_kernelcore / usable_nodes;
  5461. for_each_node_state(nid, N_MEMORY) {
  5462. unsigned long start_pfn, end_pfn;
  5463. /*
  5464. * Recalculate kernelcore_node if the division per node
  5465. * now exceeds what is necessary to satisfy the requested
  5466. * amount of memory for the kernel
  5467. */
  5468. if (required_kernelcore < kernelcore_node)
  5469. kernelcore_node = required_kernelcore / usable_nodes;
  5470. /*
  5471. * As the map is walked, we track how much memory is usable
  5472. * by the kernel using kernelcore_remaining. When it is
  5473. * 0, the rest of the node is usable by ZONE_MOVABLE
  5474. */
  5475. kernelcore_remaining = kernelcore_node;
  5476. /* Go through each range of PFNs within this node */
  5477. for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
  5478. unsigned long size_pages;
  5479. start_pfn = max(start_pfn, zone_movable_pfn[nid]);
  5480. if (start_pfn >= end_pfn)
  5481. continue;
  5482. /* Account for what is only usable for kernelcore */
  5483. if (start_pfn < usable_startpfn) {
  5484. unsigned long kernel_pages;
  5485. kernel_pages = min(end_pfn, usable_startpfn)
  5486. - start_pfn;
  5487. kernelcore_remaining -= min(kernel_pages,
  5488. kernelcore_remaining);
  5489. required_kernelcore -= min(kernel_pages,
  5490. required_kernelcore);
  5491. /* Continue if range is now fully accounted */
  5492. if (end_pfn <= usable_startpfn) {
  5493. /*
  5494. * Push zone_movable_pfn to the end so
  5495. * that if we have to rebalance
  5496. * kernelcore across nodes, we will
  5497. * not double account here
  5498. */
  5499. zone_movable_pfn[nid] = end_pfn;
  5500. continue;
  5501. }
  5502. start_pfn = usable_startpfn;
  5503. }
  5504. /*
  5505. * The usable PFN range for ZONE_MOVABLE is from
  5506. * start_pfn->end_pfn. Calculate size_pages as the
  5507. * number of pages used as kernelcore
  5508. */
  5509. size_pages = end_pfn - start_pfn;
  5510. if (size_pages > kernelcore_remaining)
  5511. size_pages = kernelcore_remaining;
  5512. zone_movable_pfn[nid] = start_pfn + size_pages;
  5513. /*
  5514. * Some kernelcore has been met, update counts and
  5515. * break if the kernelcore for this node has been
  5516. * satisfied
  5517. */
  5518. required_kernelcore -= min(required_kernelcore,
  5519. size_pages);
  5520. kernelcore_remaining -= size_pages;
  5521. if (!kernelcore_remaining)
  5522. break;
  5523. }
  5524. }
  5525. /*
  5526. * If there is still required_kernelcore, we do another pass with one
  5527. * less node in the count. This will push zone_movable_pfn[nid] further
  5528. * along on the nodes that still have memory until kernelcore is
  5529. * satisfied
  5530. */
  5531. usable_nodes--;
  5532. if (usable_nodes && required_kernelcore > usable_nodes)
  5533. goto restart;
  5534. out2:
  5535. /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
  5536. for (nid = 0; nid < MAX_NUMNODES; nid++)
  5537. zone_movable_pfn[nid] =
  5538. roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
  5539. out:
  5540. /* restore the node_state */
  5541. node_states[N_MEMORY] = saved_node_state;
  5542. }
  5543. /* Any regular or high memory on that node ? */
  5544. static void check_for_memory(pg_data_t *pgdat, int nid)
  5545. {
  5546. enum zone_type zone_type;
  5547. if (N_MEMORY == N_NORMAL_MEMORY)
  5548. return;
  5549. for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
  5550. struct zone *zone = &pgdat->node_zones[zone_type];
  5551. if (populated_zone(zone)) {
  5552. node_set_state(nid, N_HIGH_MEMORY);
  5553. if (N_NORMAL_MEMORY != N_HIGH_MEMORY &&
  5554. zone_type <= ZONE_NORMAL)
  5555. node_set_state(nid, N_NORMAL_MEMORY);
  5556. break;
  5557. }
  5558. }
  5559. }
  5560. /**
  5561. * free_area_init_nodes - Initialise all pg_data_t and zone data
  5562. * @max_zone_pfn: an array of max PFNs for each zone
  5563. *
  5564. * This will call free_area_init_node() for each active node in the system.
  5565. * Using the page ranges provided by memblock_set_node(), the size of each
  5566. * zone in each node and their holes is calculated. If the maximum PFN
  5567. * between two adjacent zones match, it is assumed that the zone is empty.
  5568. * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
  5569. * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
  5570. * starts where the previous one ended. For example, ZONE_DMA32 starts
  5571. * at arch_max_dma_pfn.
  5572. */
  5573. void __init free_area_init_nodes(unsigned long *max_zone_pfn)
  5574. {
  5575. unsigned long start_pfn, end_pfn;
  5576. int i, nid;
  5577. /* Record where the zone boundaries are */
  5578. memset(arch_zone_lowest_possible_pfn, 0,
  5579. sizeof(arch_zone_lowest_possible_pfn));
  5580. memset(arch_zone_highest_possible_pfn, 0,
  5581. sizeof(arch_zone_highest_possible_pfn));
  5582. start_pfn = find_min_pfn_with_active_regions();
  5583. for (i = 0; i < MAX_NR_ZONES; i++) {
  5584. if (i == ZONE_MOVABLE)
  5585. continue;
  5586. end_pfn = max(max_zone_pfn[i], start_pfn);
  5587. arch_zone_lowest_possible_pfn[i] = start_pfn;
  5588. arch_zone_highest_possible_pfn[i] = end_pfn;
  5589. start_pfn = end_pfn;
  5590. }
  5591. /* Find the PFNs that ZONE_MOVABLE begins at in each node */
  5592. memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
  5593. find_zone_movable_pfns_for_nodes();
  5594. /* Print out the zone ranges */
  5595. pr_info("Zone ranges:\n");
  5596. for (i = 0; i < MAX_NR_ZONES; i++) {
  5597. if (i == ZONE_MOVABLE)
  5598. continue;
  5599. pr_info(" %-8s ", zone_names[i]);
  5600. if (arch_zone_lowest_possible_pfn[i] ==
  5601. arch_zone_highest_possible_pfn[i])
  5602. pr_cont("empty\n");
  5603. else
  5604. pr_cont("[mem %#018Lx-%#018Lx]\n",
  5605. (u64)arch_zone_lowest_possible_pfn[i]
  5606. << PAGE_SHIFT,
  5607. ((u64)arch_zone_highest_possible_pfn[i]
  5608. << PAGE_SHIFT) - 1);
  5609. }
  5610. /* Print out the PFNs ZONE_MOVABLE begins at in each node */
  5611. pr_info("Movable zone start for each node\n");
  5612. for (i = 0; i < MAX_NUMNODES; i++) {
  5613. if (zone_movable_pfn[i])
  5614. pr_info(" Node %d: %#018Lx\n", i,
  5615. (u64)zone_movable_pfn[i] << PAGE_SHIFT);
  5616. }
  5617. /* Print out the early node map */
  5618. pr_info("Early memory node ranges\n");
  5619. for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
  5620. pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid,
  5621. (u64)start_pfn << PAGE_SHIFT,
  5622. ((u64)end_pfn << PAGE_SHIFT) - 1);
  5623. /* Initialise every node */
  5624. mminit_verify_pageflags_layout();
  5625. setup_nr_node_ids();
  5626. for_each_online_node(nid) {
  5627. pg_data_t *pgdat = NODE_DATA(nid);
  5628. free_area_init_node(nid, NULL,
  5629. find_min_pfn_for_node(nid), NULL);
  5630. /* Any memory on that node */
  5631. if (pgdat->node_present_pages)
  5632. node_set_state(nid, N_MEMORY);
  5633. check_for_memory(pgdat, nid);
  5634. }
  5635. }
  5636. static int __init cmdline_parse_core(char *p, unsigned long *core)
  5637. {
  5638. unsigned long long coremem;
  5639. if (!p)
  5640. return -EINVAL;
  5641. coremem = memparse(p, &p);
  5642. *core = coremem >> PAGE_SHIFT;
  5643. /* Paranoid check that UL is enough for the coremem value */
  5644. WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
  5645. return 0;
  5646. }
  5647. /*
  5648. * kernelcore=size sets the amount of memory for use for allocations that
  5649. * cannot be reclaimed or migrated.
  5650. */
  5651. static int __init cmdline_parse_kernelcore(char *p)
  5652. {
  5653. /* parse kernelcore=mirror */
  5654. if (parse_option_str(p, "mirror")) {
  5655. mirrored_kernelcore = true;
  5656. return 0;
  5657. }
  5658. return cmdline_parse_core(p, &required_kernelcore);
  5659. }
  5660. /*
  5661. * movablecore=size sets the amount of memory for use for allocations that
  5662. * can be reclaimed or migrated.
  5663. */
  5664. static int __init cmdline_parse_movablecore(char *p)
  5665. {
  5666. return cmdline_parse_core(p, &required_movablecore);
  5667. }
  5668. early_param("kernelcore", cmdline_parse_kernelcore);
  5669. early_param("movablecore", cmdline_parse_movablecore);
  5670. #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
  5671. void adjust_managed_page_count(struct page *page, long count)
  5672. {
  5673. spin_lock(&managed_page_count_lock);
  5674. page_zone(page)->managed_pages += count;
  5675. totalram_pages += count;
  5676. #ifdef CONFIG_HIGHMEM
  5677. if (PageHighMem(page))
  5678. totalhigh_pages += count;
  5679. #endif
  5680. spin_unlock(&managed_page_count_lock);
  5681. }
  5682. EXPORT_SYMBOL(adjust_managed_page_count);
  5683. unsigned long free_reserved_area(void *start, void *end, int poison, char *s)
  5684. {
  5685. void *pos;
  5686. unsigned long pages = 0;
  5687. start = (void *)PAGE_ALIGN((unsigned long)start);
  5688. end = (void *)((unsigned long)end & PAGE_MASK);
  5689. for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
  5690. if ((unsigned int)poison <= 0xFF)
  5691. memset(pos, poison, PAGE_SIZE);
  5692. free_reserved_page(virt_to_page(pos));
  5693. }
  5694. if (pages && s)
  5695. pr_info("Freeing %s memory: %ldK\n",
  5696. s, pages << (PAGE_SHIFT - 10));
  5697. return pages;
  5698. }
  5699. EXPORT_SYMBOL(free_reserved_area);
  5700. #ifdef CONFIG_HIGHMEM
  5701. void free_highmem_page(struct page *page)
  5702. {
  5703. __free_reserved_page(page);
  5704. totalram_pages++;
  5705. page_zone(page)->managed_pages++;
  5706. totalhigh_pages++;
  5707. }
  5708. #endif
  5709. void __init mem_init_print_info(const char *str)
  5710. {
  5711. unsigned long physpages, codesize, datasize, rosize, bss_size;
  5712. unsigned long init_code_size, init_data_size;
  5713. physpages = get_num_physpages();
  5714. codesize = _etext - _stext;
  5715. datasize = _edata - _sdata;
  5716. rosize = __end_rodata - __start_rodata;
  5717. bss_size = __bss_stop - __bss_start;
  5718. init_data_size = __init_end - __init_begin;
  5719. init_code_size = _einittext - _sinittext;
  5720. /*
  5721. * Detect special cases and adjust section sizes accordingly:
  5722. * 1) .init.* may be embedded into .data sections
  5723. * 2) .init.text.* may be out of [__init_begin, __init_end],
  5724. * please refer to arch/tile/kernel/vmlinux.lds.S.
  5725. * 3) .rodata.* may be embedded into .text or .data sections.
  5726. */
  5727. #define adj_init_size(start, end, size, pos, adj) \
  5728. do { \
  5729. if (start <= pos && pos < end && size > adj) \
  5730. size -= adj; \
  5731. } while (0)
  5732. adj_init_size(__init_begin, __init_end, init_data_size,
  5733. _sinittext, init_code_size);
  5734. adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
  5735. adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
  5736. adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
  5737. adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
  5738. #undef adj_init_size
  5739. pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved"
  5740. #ifdef CONFIG_HIGHMEM
  5741. ", %luK highmem"
  5742. #endif
  5743. "%s%s)\n",
  5744. nr_free_pages() << (PAGE_SHIFT - 10),
  5745. physpages << (PAGE_SHIFT - 10),
  5746. codesize >> 10, datasize >> 10, rosize >> 10,
  5747. (init_data_size + init_code_size) >> 10, bss_size >> 10,
  5748. (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT - 10),
  5749. totalcma_pages << (PAGE_SHIFT - 10),
  5750. #ifdef CONFIG_HIGHMEM
  5751. totalhigh_pages << (PAGE_SHIFT - 10),
  5752. #endif
  5753. str ? ", " : "", str ? str : "");
  5754. }
  5755. /**
  5756. * set_dma_reserve - set the specified number of pages reserved in the first zone
  5757. * @new_dma_reserve: The number of pages to mark reserved
  5758. *
  5759. * The per-cpu batchsize and zone watermarks are determined by managed_pages.
  5760. * In the DMA zone, a significant percentage may be consumed by kernel image
  5761. * and other unfreeable allocations which can skew the watermarks badly. This
  5762. * function may optionally be used to account for unfreeable pages in the
  5763. * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
  5764. * smaller per-cpu batchsize.
  5765. */
  5766. void __init set_dma_reserve(unsigned long new_dma_reserve)
  5767. {
  5768. dma_reserve = new_dma_reserve;
  5769. }
  5770. void __init free_area_init(unsigned long *zones_size)
  5771. {
  5772. free_area_init_node(0, zones_size,
  5773. __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
  5774. }
  5775. static int page_alloc_cpu_dead(unsigned int cpu)
  5776. {
  5777. lru_add_drain_cpu(cpu);
  5778. drain_pages(cpu);
  5779. /*
  5780. * Spill the event counters of the dead processor
  5781. * into the current processors event counters.
  5782. * This artificially elevates the count of the current
  5783. * processor.
  5784. */
  5785. vm_events_fold_cpu(cpu);
  5786. /*
  5787. * Zero the differential counters of the dead processor
  5788. * so that the vm statistics are consistent.
  5789. *
  5790. * This is only okay since the processor is dead and cannot
  5791. * race with what we are doing.
  5792. */
  5793. cpu_vm_stats_fold(cpu);
  5794. return 0;
  5795. }
  5796. void __init page_alloc_init(void)
  5797. {
  5798. int ret;
  5799. ret = cpuhp_setup_state_nocalls(CPUHP_PAGE_ALLOC_DEAD,
  5800. "mm/page_alloc:dead", NULL,
  5801. page_alloc_cpu_dead);
  5802. WARN_ON(ret < 0);
  5803. }
  5804. /*
  5805. * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio
  5806. * or min_free_kbytes changes.
  5807. */
  5808. static void calculate_totalreserve_pages(void)
  5809. {
  5810. struct pglist_data *pgdat;
  5811. unsigned long reserve_pages = 0;
  5812. enum zone_type i, j;
  5813. for_each_online_pgdat(pgdat) {
  5814. pgdat->totalreserve_pages = 0;
  5815. for (i = 0; i < MAX_NR_ZONES; i++) {
  5816. struct zone *zone = pgdat->node_zones + i;
  5817. long max = 0;
  5818. /* Find valid and maximum lowmem_reserve in the zone */
  5819. for (j = i; j < MAX_NR_ZONES; j++) {
  5820. if (zone->lowmem_reserve[j] > max)
  5821. max = zone->lowmem_reserve[j];
  5822. }
  5823. /* we treat the high watermark as reserved pages. */
  5824. max += high_wmark_pages(zone);
  5825. if (max > zone->managed_pages)
  5826. max = zone->managed_pages;
  5827. pgdat->totalreserve_pages += max;
  5828. reserve_pages += max;
  5829. }
  5830. }
  5831. totalreserve_pages = reserve_pages;
  5832. }
  5833. /*
  5834. * setup_per_zone_lowmem_reserve - called whenever
  5835. * sysctl_lowmem_reserve_ratio changes. Ensures that each zone
  5836. * has a correct pages reserved value, so an adequate number of
  5837. * pages are left in the zone after a successful __alloc_pages().
  5838. */
  5839. static void setup_per_zone_lowmem_reserve(void)
  5840. {
  5841. struct pglist_data *pgdat;
  5842. enum zone_type j, idx;
  5843. for_each_online_pgdat(pgdat) {
  5844. for (j = 0; j < MAX_NR_ZONES; j++) {
  5845. struct zone *zone = pgdat->node_zones + j;
  5846. unsigned long managed_pages = zone->managed_pages;
  5847. zone->lowmem_reserve[j] = 0;
  5848. idx = j;
  5849. while (idx) {
  5850. struct zone *lower_zone;
  5851. idx--;
  5852. if (sysctl_lowmem_reserve_ratio[idx] < 1)
  5853. sysctl_lowmem_reserve_ratio[idx] = 1;
  5854. lower_zone = pgdat->node_zones + idx;
  5855. lower_zone->lowmem_reserve[j] = managed_pages /
  5856. sysctl_lowmem_reserve_ratio[idx];
  5857. managed_pages += lower_zone->managed_pages;
  5858. }
  5859. }
  5860. }
  5861. /* update totalreserve_pages */
  5862. calculate_totalreserve_pages();
  5863. }
  5864. static void __setup_per_zone_wmarks(void)
  5865. {
  5866. unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
  5867. unsigned long lowmem_pages = 0;
  5868. struct zone *zone;
  5869. unsigned long flags;
  5870. /* Calculate total number of !ZONE_HIGHMEM pages */
  5871. for_each_zone(zone) {
  5872. if (!is_highmem(zone))
  5873. lowmem_pages += zone->managed_pages;
  5874. }
  5875. for_each_zone(zone) {
  5876. u64 tmp;
  5877. spin_lock_irqsave(&zone->lock, flags);
  5878. tmp = (u64)pages_min * zone->managed_pages;
  5879. do_div(tmp, lowmem_pages);
  5880. if (is_highmem(zone)) {
  5881. /*
  5882. * __GFP_HIGH and PF_MEMALLOC allocations usually don't
  5883. * need highmem pages, so cap pages_min to a small
  5884. * value here.
  5885. *
  5886. * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
  5887. * deltas control asynch page reclaim, and so should
  5888. * not be capped for highmem.
  5889. */
  5890. unsigned long min_pages;
  5891. min_pages = zone->managed_pages / 1024;
  5892. min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
  5893. zone->watermark[WMARK_MIN] = min_pages;
  5894. } else {
  5895. /*
  5896. * If it's a lowmem zone, reserve a number of pages
  5897. * proportionate to the zone's size.
  5898. */
  5899. zone->watermark[WMARK_MIN] = tmp;
  5900. }
  5901. /*
  5902. * Set the kswapd watermarks distance according to the
  5903. * scale factor in proportion to available memory, but
  5904. * ensure a minimum size on small systems.
  5905. */
  5906. tmp = max_t(u64, tmp >> 2,
  5907. mult_frac(zone->managed_pages,
  5908. watermark_scale_factor, 10000));
  5909. zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp;
  5910. zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2;
  5911. spin_unlock_irqrestore(&zone->lock, flags);
  5912. }
  5913. /* update totalreserve_pages */
  5914. calculate_totalreserve_pages();
  5915. }
  5916. /**
  5917. * setup_per_zone_wmarks - called when min_free_kbytes changes
  5918. * or when memory is hot-{added|removed}
  5919. *
  5920. * Ensures that the watermark[min,low,high] values for each zone are set
  5921. * correctly with respect to min_free_kbytes.
  5922. */
  5923. void setup_per_zone_wmarks(void)
  5924. {
  5925. mutex_lock(&zonelists_mutex);
  5926. __setup_per_zone_wmarks();
  5927. mutex_unlock(&zonelists_mutex);
  5928. }
  5929. /*
  5930. * Initialise min_free_kbytes.
  5931. *
  5932. * For small machines we want it small (128k min). For large machines
  5933. * we want it large (64MB max). But it is not linear, because network
  5934. * bandwidth does not increase linearly with machine size. We use
  5935. *
  5936. * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
  5937. * min_free_kbytes = sqrt(lowmem_kbytes * 16)
  5938. *
  5939. * which yields
  5940. *
  5941. * 16MB: 512k
  5942. * 32MB: 724k
  5943. * 64MB: 1024k
  5944. * 128MB: 1448k
  5945. * 256MB: 2048k
  5946. * 512MB: 2896k
  5947. * 1024MB: 4096k
  5948. * 2048MB: 5792k
  5949. * 4096MB: 8192k
  5950. * 8192MB: 11584k
  5951. * 16384MB: 16384k
  5952. */
  5953. int __meminit init_per_zone_wmark_min(void)
  5954. {
  5955. unsigned long lowmem_kbytes;
  5956. int new_min_free_kbytes;
  5957. lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
  5958. new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
  5959. if (new_min_free_kbytes > user_min_free_kbytes) {
  5960. min_free_kbytes = new_min_free_kbytes;
  5961. if (min_free_kbytes < 128)
  5962. min_free_kbytes = 128;
  5963. if (min_free_kbytes > 65536)
  5964. min_free_kbytes = 65536;
  5965. } else {
  5966. pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
  5967. new_min_free_kbytes, user_min_free_kbytes);
  5968. }
  5969. setup_per_zone_wmarks();
  5970. refresh_zone_stat_thresholds();
  5971. setup_per_zone_lowmem_reserve();
  5972. #ifdef CONFIG_NUMA
  5973. setup_min_unmapped_ratio();
  5974. setup_min_slab_ratio();
  5975. #endif
  5976. return 0;
  5977. }
  5978. core_initcall(init_per_zone_wmark_min)
  5979. /*
  5980. * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
  5981. * that we can call two helper functions whenever min_free_kbytes
  5982. * changes.
  5983. */
  5984. int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write,
  5985. void __user *buffer, size_t *length, loff_t *ppos)
  5986. {
  5987. int rc;
  5988. rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
  5989. if (rc)
  5990. return rc;
  5991. if (write) {
  5992. user_min_free_kbytes = min_free_kbytes;
  5993. setup_per_zone_wmarks();
  5994. }
  5995. return 0;
  5996. }
  5997. int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write,
  5998. void __user *buffer, size_t *length, loff_t *ppos)
  5999. {
  6000. int rc;
  6001. rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
  6002. if (rc)
  6003. return rc;
  6004. if (write)
  6005. setup_per_zone_wmarks();
  6006. return 0;
  6007. }
  6008. #ifdef CONFIG_NUMA
  6009. static void setup_min_unmapped_ratio(void)
  6010. {
  6011. pg_data_t *pgdat;
  6012. struct zone *zone;
  6013. for_each_online_pgdat(pgdat)
  6014. pgdat->min_unmapped_pages = 0;
  6015. for_each_zone(zone)
  6016. zone->zone_pgdat->min_unmapped_pages += (zone->managed_pages *
  6017. sysctl_min_unmapped_ratio) / 100;
  6018. }
  6019. int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write,
  6020. void __user *buffer, size_t *length, loff_t *ppos)
  6021. {
  6022. int rc;
  6023. rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
  6024. if (rc)
  6025. return rc;
  6026. setup_min_unmapped_ratio();
  6027. return 0;
  6028. }
  6029. static void setup_min_slab_ratio(void)
  6030. {
  6031. pg_data_t *pgdat;
  6032. struct zone *zone;
  6033. for_each_online_pgdat(pgdat)
  6034. pgdat->min_slab_pages = 0;
  6035. for_each_zone(zone)
  6036. zone->zone_pgdat->min_slab_pages += (zone->managed_pages *
  6037. sysctl_min_slab_ratio) / 100;
  6038. }
  6039. int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write,
  6040. void __user *buffer, size_t *length, loff_t *ppos)
  6041. {
  6042. int rc;
  6043. rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
  6044. if (rc)
  6045. return rc;
  6046. setup_min_slab_ratio();
  6047. return 0;
  6048. }
  6049. #endif
  6050. /*
  6051. * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
  6052. * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
  6053. * whenever sysctl_lowmem_reserve_ratio changes.
  6054. *
  6055. * The reserve ratio obviously has absolutely no relation with the
  6056. * minimum watermarks. The lowmem reserve ratio can only make sense
  6057. * if in function of the boot time zone sizes.
  6058. */
  6059. int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write,
  6060. void __user *buffer, size_t *length, loff_t *ppos)
  6061. {
  6062. proc_dointvec_minmax(table, write, buffer, length, ppos);
  6063. setup_per_zone_lowmem_reserve();
  6064. return 0;
  6065. }
  6066. /*
  6067. * percpu_pagelist_fraction - changes the pcp->high for each zone on each
  6068. * cpu. It is the fraction of total pages in each zone that a hot per cpu
  6069. * pagelist can have before it gets flushed back to buddy allocator.
  6070. */
  6071. int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write,
  6072. void __user *buffer, size_t *length, loff_t *ppos)
  6073. {
  6074. struct zone *zone;
  6075. int old_percpu_pagelist_fraction;
  6076. int ret;
  6077. mutex_lock(&pcp_batch_high_lock);
  6078. old_percpu_pagelist_fraction = percpu_pagelist_fraction;
  6079. ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
  6080. if (!write || ret < 0)
  6081. goto out;
  6082. /* Sanity checking to avoid pcp imbalance */
  6083. if (percpu_pagelist_fraction &&
  6084. percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) {
  6085. percpu_pagelist_fraction = old_percpu_pagelist_fraction;
  6086. ret = -EINVAL;
  6087. goto out;
  6088. }
  6089. /* No change? */
  6090. if (percpu_pagelist_fraction == old_percpu_pagelist_fraction)
  6091. goto out;
  6092. for_each_populated_zone(zone) {
  6093. unsigned int cpu;
  6094. for_each_possible_cpu(cpu)
  6095. pageset_set_high_and_batch(zone,
  6096. per_cpu_ptr(zone->pageset, cpu));
  6097. }
  6098. out:
  6099. mutex_unlock(&pcp_batch_high_lock);
  6100. return ret;
  6101. }
  6102. #ifdef CONFIG_NUMA
  6103. int hashdist = HASHDIST_DEFAULT;
  6104. static int __init set_hashdist(char *str)
  6105. {
  6106. if (!str)
  6107. return 0;
  6108. hashdist = simple_strtoul(str, &str, 0);
  6109. return 1;
  6110. }
  6111. __setup("hashdist=", set_hashdist);
  6112. #endif
  6113. #ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES
  6114. /*
  6115. * Returns the number of pages that arch has reserved but
  6116. * is not known to alloc_large_system_hash().
  6117. */
  6118. static unsigned long __init arch_reserved_kernel_pages(void)
  6119. {
  6120. return 0;
  6121. }
  6122. #endif
  6123. /*
  6124. * allocate a large system hash table from bootmem
  6125. * - it is assumed that the hash table must contain an exact power-of-2
  6126. * quantity of entries
  6127. * - limit is the number of hash buckets, not the total allocation size
  6128. */
  6129. void *__init alloc_large_system_hash(const char *tablename,
  6130. unsigned long bucketsize,
  6131. unsigned long numentries,
  6132. int scale,
  6133. int flags,
  6134. unsigned int *_hash_shift,
  6135. unsigned int *_hash_mask,
  6136. unsigned long low_limit,
  6137. unsigned long high_limit)
  6138. {
  6139. unsigned long long max = high_limit;
  6140. unsigned long log2qty, size;
  6141. void *table = NULL;
  6142. /* allow the kernel cmdline to have a say */
  6143. if (!numentries) {
  6144. /* round applicable memory size up to nearest megabyte */
  6145. numentries = nr_kernel_pages;
  6146. numentries -= arch_reserved_kernel_pages();
  6147. /* It isn't necessary when PAGE_SIZE >= 1MB */
  6148. if (PAGE_SHIFT < 20)
  6149. numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
  6150. /* limit to 1 bucket per 2^scale bytes of low memory */
  6151. if (scale > PAGE_SHIFT)
  6152. numentries >>= (scale - PAGE_SHIFT);
  6153. else
  6154. numentries <<= (PAGE_SHIFT - scale);
  6155. /* Make sure we've got at least a 0-order allocation.. */
  6156. if (unlikely(flags & HASH_SMALL)) {
  6157. /* Makes no sense without HASH_EARLY */
  6158. WARN_ON(!(flags & HASH_EARLY));
  6159. if (!(numentries >> *_hash_shift)) {
  6160. numentries = 1UL << *_hash_shift;
  6161. BUG_ON(!numentries);
  6162. }
  6163. } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
  6164. numentries = PAGE_SIZE / bucketsize;
  6165. }
  6166. numentries = roundup_pow_of_two(numentries);
  6167. /* limit allocation size to 1/16 total memory by default */
  6168. if (max == 0) {
  6169. max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
  6170. do_div(max, bucketsize);
  6171. }
  6172. max = min(max, 0x80000000ULL);
  6173. if (numentries < low_limit)
  6174. numentries = low_limit;
  6175. if (numentries > max)
  6176. numentries = max;
  6177. log2qty = ilog2(numentries);
  6178. do {
  6179. size = bucketsize << log2qty;
  6180. if (flags & HASH_EARLY)
  6181. table = memblock_virt_alloc_nopanic(size, 0);
  6182. else if (hashdist)
  6183. table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
  6184. else {
  6185. /*
  6186. * If bucketsize is not a power-of-two, we may free
  6187. * some pages at the end of hash table which
  6188. * alloc_pages_exact() automatically does
  6189. */
  6190. if (get_order(size) < MAX_ORDER) {
  6191. table = alloc_pages_exact(size, GFP_ATOMIC);
  6192. kmemleak_alloc(table, size, 1, GFP_ATOMIC);
  6193. }
  6194. }
  6195. } while (!table && size > PAGE_SIZE && --log2qty);
  6196. if (!table)
  6197. panic("Failed to allocate %s hash table\n", tablename);
  6198. pr_info("%s hash table entries: %ld (order: %d, %lu bytes)\n",
  6199. tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size);
  6200. if (_hash_shift)
  6201. *_hash_shift = log2qty;
  6202. if (_hash_mask)
  6203. *_hash_mask = (1 << log2qty) - 1;
  6204. return table;
  6205. }
  6206. /*
  6207. * This function checks whether pageblock includes unmovable pages or not.
  6208. * If @count is not zero, it is okay to include less @count unmovable pages
  6209. *
  6210. * PageLRU check without isolation or lru_lock could race so that
  6211. * MIGRATE_MOVABLE block might include unmovable pages. And __PageMovable
  6212. * check without lock_page also may miss some movable non-lru pages at
  6213. * race condition. So you can't expect this function should be exact.
  6214. */
  6215. bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
  6216. bool skip_hwpoisoned_pages)
  6217. {
  6218. unsigned long pfn, iter, found;
  6219. int mt;
  6220. /*
  6221. * For avoiding noise data, lru_add_drain_all() should be called
  6222. * If ZONE_MOVABLE, the zone never contains unmovable pages
  6223. */
  6224. if (zone_idx(zone) == ZONE_MOVABLE)
  6225. return false;
  6226. mt = get_pageblock_migratetype(page);
  6227. if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt))
  6228. return false;
  6229. pfn = page_to_pfn(page);
  6230. for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
  6231. unsigned long check = pfn + iter;
  6232. if (!pfn_valid_within(check))
  6233. continue;
  6234. page = pfn_to_page(check);
  6235. /*
  6236. * Hugepages are not in LRU lists, but they're movable.
  6237. * We need not scan over tail pages bacause we don't
  6238. * handle each tail page individually in migration.
  6239. */
  6240. if (PageHuge(page)) {
  6241. iter = round_up(iter + 1, 1<<compound_order(page)) - 1;
  6242. continue;
  6243. }
  6244. /*
  6245. * We can't use page_count without pin a page
  6246. * because another CPU can free compound page.
  6247. * This check already skips compound tails of THP
  6248. * because their page->_refcount is zero at all time.
  6249. */
  6250. if (!page_ref_count(page)) {
  6251. if (PageBuddy(page))
  6252. iter += (1 << page_order(page)) - 1;
  6253. continue;
  6254. }
  6255. /*
  6256. * The HWPoisoned page may be not in buddy system, and
  6257. * page_count() is not 0.
  6258. */
  6259. if (skip_hwpoisoned_pages && PageHWPoison(page))
  6260. continue;
  6261. if (__PageMovable(page))
  6262. continue;
  6263. if (!PageLRU(page))
  6264. found++;
  6265. /*
  6266. * If there are RECLAIMABLE pages, we need to check
  6267. * it. But now, memory offline itself doesn't call
  6268. * shrink_node_slabs() and it still to be fixed.
  6269. */
  6270. /*
  6271. * If the page is not RAM, page_count()should be 0.
  6272. * we don't need more check. This is an _used_ not-movable page.
  6273. *
  6274. * The problematic thing here is PG_reserved pages. PG_reserved
  6275. * is set to both of a memory hole page and a _used_ kernel
  6276. * page at boot.
  6277. */
  6278. if (found > count)
  6279. return true;
  6280. }
  6281. return false;
  6282. }
  6283. bool is_pageblock_removable_nolock(struct page *page)
  6284. {
  6285. struct zone *zone;
  6286. unsigned long pfn;
  6287. /*
  6288. * We have to be careful here because we are iterating over memory
  6289. * sections which are not zone aware so we might end up outside of
  6290. * the zone but still within the section.
  6291. * We have to take care about the node as well. If the node is offline
  6292. * its NODE_DATA will be NULL - see page_zone.
  6293. */
  6294. if (!node_online(page_to_nid(page)))
  6295. return false;
  6296. zone = page_zone(page);
  6297. pfn = page_to_pfn(page);
  6298. if (!zone_spans_pfn(zone, pfn))
  6299. return false;
  6300. return !has_unmovable_pages(zone, page, 0, true);
  6301. }
  6302. #if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA)
  6303. static unsigned long pfn_max_align_down(unsigned long pfn)
  6304. {
  6305. return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
  6306. pageblock_nr_pages) - 1);
  6307. }
  6308. static unsigned long pfn_max_align_up(unsigned long pfn)
  6309. {
  6310. return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
  6311. pageblock_nr_pages));
  6312. }
  6313. /* [start, end) must belong to a single zone. */
  6314. static int __alloc_contig_migrate_range(struct compact_control *cc,
  6315. unsigned long start, unsigned long end)
  6316. {
  6317. /* This function is based on compact_zone() from compaction.c. */
  6318. unsigned long nr_reclaimed;
  6319. unsigned long pfn = start;
  6320. unsigned int tries = 0;
  6321. int ret = 0;
  6322. migrate_prep();
  6323. while (pfn < end || !list_empty(&cc->migratepages)) {
  6324. if (fatal_signal_pending(current)) {
  6325. ret = -EINTR;
  6326. break;
  6327. }
  6328. if (list_empty(&cc->migratepages)) {
  6329. cc->nr_migratepages = 0;
  6330. pfn = isolate_migratepages_range(cc, pfn, end);
  6331. if (!pfn) {
  6332. ret = -EINTR;
  6333. break;
  6334. }
  6335. tries = 0;
  6336. } else if (++tries == 5) {
  6337. ret = ret < 0 ? ret : -EBUSY;
  6338. break;
  6339. }
  6340. nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
  6341. &cc->migratepages);
  6342. cc->nr_migratepages -= nr_reclaimed;
  6343. ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
  6344. NULL, 0, cc->mode, MR_CMA);
  6345. }
  6346. if (ret < 0) {
  6347. putback_movable_pages(&cc->migratepages);
  6348. return ret;
  6349. }
  6350. return 0;
  6351. }
  6352. /**
  6353. * alloc_contig_range() -- tries to allocate given range of pages
  6354. * @start: start PFN to allocate
  6355. * @end: one-past-the-last PFN to allocate
  6356. * @migratetype: migratetype of the underlaying pageblocks (either
  6357. * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
  6358. * in range must have the same migratetype and it must
  6359. * be either of the two.
  6360. * @gfp_mask: GFP mask to use during compaction
  6361. *
  6362. * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
  6363. * aligned, however it's the caller's responsibility to guarantee that
  6364. * we are the only thread that changes migrate type of pageblocks the
  6365. * pages fall in.
  6366. *
  6367. * The PFN range must belong to a single zone.
  6368. *
  6369. * Returns zero on success or negative error code. On success all
  6370. * pages which PFN is in [start, end) are allocated for the caller and
  6371. * need to be freed with free_contig_range().
  6372. */
  6373. int alloc_contig_range(unsigned long start, unsigned long end,
  6374. unsigned migratetype, gfp_t gfp_mask)
  6375. {
  6376. unsigned long outer_start, outer_end;
  6377. unsigned int order;
  6378. int ret = 0;
  6379. struct compact_control cc = {
  6380. .nr_migratepages = 0,
  6381. .order = -1,
  6382. .zone = page_zone(pfn_to_page(start)),
  6383. .mode = MIGRATE_SYNC,
  6384. .ignore_skip_hint = true,
  6385. .gfp_mask = memalloc_noio_flags(gfp_mask),
  6386. };
  6387. INIT_LIST_HEAD(&cc.migratepages);
  6388. /*
  6389. * What we do here is we mark all pageblocks in range as
  6390. * MIGRATE_ISOLATE. Because pageblock and max order pages may
  6391. * have different sizes, and due to the way page allocator
  6392. * work, we align the range to biggest of the two pages so
  6393. * that page allocator won't try to merge buddies from
  6394. * different pageblocks and change MIGRATE_ISOLATE to some
  6395. * other migration type.
  6396. *
  6397. * Once the pageblocks are marked as MIGRATE_ISOLATE, we
  6398. * migrate the pages from an unaligned range (ie. pages that
  6399. * we are interested in). This will put all the pages in
  6400. * range back to page allocator as MIGRATE_ISOLATE.
  6401. *
  6402. * When this is done, we take the pages in range from page
  6403. * allocator removing them from the buddy system. This way
  6404. * page allocator will never consider using them.
  6405. *
  6406. * This lets us mark the pageblocks back as
  6407. * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
  6408. * aligned range but not in the unaligned, original range are
  6409. * put back to page allocator so that buddy can use them.
  6410. */
  6411. ret = start_isolate_page_range(pfn_max_align_down(start),
  6412. pfn_max_align_up(end), migratetype,
  6413. false);
  6414. if (ret)
  6415. return ret;
  6416. /*
  6417. * In case of -EBUSY, we'd like to know which page causes problem.
  6418. * So, just fall through. We will check it in test_pages_isolated().
  6419. */
  6420. ret = __alloc_contig_migrate_range(&cc, start, end);
  6421. if (ret && ret != -EBUSY)
  6422. goto done;
  6423. /*
  6424. * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
  6425. * aligned blocks that are marked as MIGRATE_ISOLATE. What's
  6426. * more, all pages in [start, end) are free in page allocator.
  6427. * What we are going to do is to allocate all pages from
  6428. * [start, end) (that is remove them from page allocator).
  6429. *
  6430. * The only problem is that pages at the beginning and at the
  6431. * end of interesting range may be not aligned with pages that
  6432. * page allocator holds, ie. they can be part of higher order
  6433. * pages. Because of this, we reserve the bigger range and
  6434. * once this is done free the pages we are not interested in.
  6435. *
  6436. * We don't have to hold zone->lock here because the pages are
  6437. * isolated thus they won't get removed from buddy.
  6438. */
  6439. lru_add_drain_all();
  6440. drain_all_pages(cc.zone);
  6441. order = 0;
  6442. outer_start = start;
  6443. while (!PageBuddy(pfn_to_page(outer_start))) {
  6444. if (++order >= MAX_ORDER) {
  6445. outer_start = start;
  6446. break;
  6447. }
  6448. outer_start &= ~0UL << order;
  6449. }
  6450. if (outer_start != start) {
  6451. order = page_order(pfn_to_page(outer_start));
  6452. /*
  6453. * outer_start page could be small order buddy page and
  6454. * it doesn't include start page. Adjust outer_start
  6455. * in this case to report failed page properly
  6456. * on tracepoint in test_pages_isolated()
  6457. */
  6458. if (outer_start + (1UL << order) <= start)
  6459. outer_start = start;
  6460. }
  6461. /* Make sure the range is really isolated. */
  6462. if (test_pages_isolated(outer_start, end, false)) {
  6463. pr_info("%s: [%lx, %lx) PFNs busy\n",
  6464. __func__, outer_start, end);
  6465. ret = -EBUSY;
  6466. goto done;
  6467. }
  6468. /* Grab isolated pages from freelists. */
  6469. outer_end = isolate_freepages_range(&cc, outer_start, end);
  6470. if (!outer_end) {
  6471. ret = -EBUSY;
  6472. goto done;
  6473. }
  6474. /* Free head and tail (if any) */
  6475. if (start != outer_start)
  6476. free_contig_range(outer_start, start - outer_start);
  6477. if (end != outer_end)
  6478. free_contig_range(end, outer_end - end);
  6479. done:
  6480. undo_isolate_page_range(pfn_max_align_down(start),
  6481. pfn_max_align_up(end), migratetype);
  6482. return ret;
  6483. }
  6484. void free_contig_range(unsigned long pfn, unsigned nr_pages)
  6485. {
  6486. unsigned int count = 0;
  6487. for (; nr_pages--; pfn++) {
  6488. struct page *page = pfn_to_page(pfn);
  6489. count += page_count(page) != 1;
  6490. __free_page(page);
  6491. }
  6492. WARN(count != 0, "%d pages are still in use!\n", count);
  6493. }
  6494. #endif
  6495. #ifdef CONFIG_MEMORY_HOTPLUG
  6496. /*
  6497. * The zone indicated has a new number of managed_pages; batch sizes and percpu
  6498. * page high values need to be recalulated.
  6499. */
  6500. void __meminit zone_pcp_update(struct zone *zone)
  6501. {
  6502. unsigned cpu;
  6503. mutex_lock(&pcp_batch_high_lock);
  6504. for_each_possible_cpu(cpu)
  6505. pageset_set_high_and_batch(zone,
  6506. per_cpu_ptr(zone->pageset, cpu));
  6507. mutex_unlock(&pcp_batch_high_lock);
  6508. }
  6509. #endif
  6510. void zone_pcp_reset(struct zone *zone)
  6511. {
  6512. unsigned long flags;
  6513. int cpu;
  6514. struct per_cpu_pageset *pset;
  6515. /* avoid races with drain_pages() */
  6516. local_irq_save(flags);
  6517. if (zone->pageset != &boot_pageset) {
  6518. for_each_online_cpu(cpu) {
  6519. pset = per_cpu_ptr(zone->pageset, cpu);
  6520. drain_zonestat(zone, pset);
  6521. }
  6522. free_percpu(zone->pageset);
  6523. zone->pageset = &boot_pageset;
  6524. }
  6525. local_irq_restore(flags);
  6526. }
  6527. #ifdef CONFIG_MEMORY_HOTREMOVE
  6528. /*
  6529. * All pages in the range must be in a single zone and isolated
  6530. * before calling this.
  6531. */
  6532. void
  6533. __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
  6534. {
  6535. struct page *page;
  6536. struct zone *zone;
  6537. unsigned int order, i;
  6538. unsigned long pfn;
  6539. unsigned long flags;
  6540. /* find the first valid pfn */
  6541. for (pfn = start_pfn; pfn < end_pfn; pfn++)
  6542. if (pfn_valid(pfn))
  6543. break;
  6544. if (pfn == end_pfn)
  6545. return;
  6546. zone = page_zone(pfn_to_page(pfn));
  6547. spin_lock_irqsave(&zone->lock, flags);
  6548. pfn = start_pfn;
  6549. while (pfn < end_pfn) {
  6550. if (!pfn_valid(pfn)) {
  6551. pfn++;
  6552. continue;
  6553. }
  6554. page = pfn_to_page(pfn);
  6555. /*
  6556. * The HWPoisoned page may be not in buddy system, and
  6557. * page_count() is not 0.
  6558. */
  6559. if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
  6560. pfn++;
  6561. SetPageReserved(page);
  6562. continue;
  6563. }
  6564. BUG_ON(page_count(page));
  6565. BUG_ON(!PageBuddy(page));
  6566. order = page_order(page);
  6567. #ifdef CONFIG_DEBUG_VM
  6568. pr_info("remove from free list %lx %d %lx\n",
  6569. pfn, 1 << order, end_pfn);
  6570. #endif
  6571. list_del(&page->lru);
  6572. rmv_page_order(page);
  6573. zone->free_area[order].nr_free--;
  6574. for (i = 0; i < (1 << order); i++)
  6575. SetPageReserved((page+i));
  6576. pfn += (1 << order);
  6577. }
  6578. spin_unlock_irqrestore(&zone->lock, flags);
  6579. }
  6580. #endif
  6581. bool is_free_buddy_page(struct page *page)
  6582. {
  6583. struct zone *zone = page_zone(page);
  6584. unsigned long pfn = page_to_pfn(page);
  6585. unsigned long flags;
  6586. unsigned int order;
  6587. spin_lock_irqsave(&zone->lock, flags);
  6588. for (order = 0; order < MAX_ORDER; order++) {
  6589. struct page *page_head = page - (pfn & ((1 << order) - 1));
  6590. if (PageBuddy(page_head) && page_order(page_head) >= order)
  6591. break;
  6592. }
  6593. spin_unlock_irqrestore(&zone->lock, flags);
  6594. return order < MAX_ORDER;
  6595. }