sched.c 160 KB

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385138613871388138913901391139213931394139513961397139813991400140114021403140414051406140714081409141014111412141314141415141614171418141914201421142214231424142514261427142814291430143114321433143414351436143714381439144014411442144314441445144614471448144914501451145214531454145514561457145814591460146114621463146414651466146714681469147014711472147314741475147614771478147914801481148214831484148514861487148814891490149114921493149414951496149714981499150015011502150315041505150615071508150915101511151215131514151515161517151815191520152115221523152415251526152715281529153015311532153315341535153615371538153915401541154215431544154515461547154815491550155115521553155415551556155715581559156015611562156315641565156615671568156915701571157215731574157515761577157815791580158115821583158415851586158715881589159015911592159315941595159615971598159916001601160216031604160516061607160816091610161116121613161416151616161716181619162016211622162316241625162616271628162916301631163216331634163516361637163816391640164116421643164416451646164716481649165016511652165316541655165616571658165916601661166216631664166516661667166816691670167116721673167416751676167716781679168016811682168316841685168616871688168916901691169216931694169516961697169816991700170117021703170417051706170717081709171017111712171317141715171617171718171917201721172217231724172517261727172817291730173117321733173417351736173717381739174017411742174317441745174617471748174917501751175217531754175517561757175817591760176117621763176417651766176717681769177017711772177317741775177617771778177917801781178217831784178517861787178817891790179117921793179417951796179717981799180018011802180318041805180618071808180918101811181218131814181518161817181818191820182118221823182418251826182718281829183018311832183318341835183618371838183918401841184218431844184518461847184818491850185118521853185418551856185718581859186018611862186318641865186618671868186918701871187218731874187518761877187818791880188118821883188418851886188718881889189018911892189318941895189618971898189919001901190219031904190519061907190819091910191119121913191419151916191719181919192019211922192319241925192619271928192919301931193219331934193519361937193819391940194119421943194419451946194719481949195019511952195319541955195619571958195919601961196219631964196519661967196819691970197119721973197419751976197719781979198019811982198319841985198619871988198919901991199219931994199519961997199819992000200120022003200420052006200720082009201020112012201320142015201620172018201920202021202220232024202520262027202820292030203120322033203420352036203720382039204020412042204320442045204620472048204920502051205220532054205520562057205820592060206120622063206420652066206720682069207020712072207320742075207620772078207920802081208220832084208520862087208820892090209120922093209420952096209720982099210021012102210321042105210621072108210921102111211221132114211521162117211821192120212121222123212421252126212721282129213021312132213321342135213621372138213921402141214221432144214521462147214821492150215121522153215421552156215721582159216021612162216321642165216621672168216921702171217221732174217521762177217821792180218121822183218421852186218721882189219021912192219321942195219621972198219922002201220222032204220522062207220822092210221122122213221422152216221722182219222022212222222322242225222622272228222922302231223222332234223522362237223822392240224122422243224422452246224722482249225022512252225322542255225622572258225922602261226222632264226522662267226822692270227122722273227422752276227722782279228022812282228322842285228622872288228922902291229222932294229522962297229822992300230123022303230423052306230723082309231023112312231323142315231623172318231923202321232223232324232523262327232823292330233123322333233423352336233723382339234023412342234323442345234623472348234923502351235223532354235523562357235823592360236123622363236423652366236723682369237023712372237323742375237623772378237923802381238223832384238523862387238823892390239123922393239423952396239723982399240024012402240324042405240624072408240924102411241224132414241524162417241824192420242124222423242424252426242724282429243024312432243324342435243624372438243924402441244224432444244524462447244824492450245124522453245424552456245724582459246024612462246324642465246624672468246924702471247224732474247524762477247824792480248124822483248424852486248724882489249024912492249324942495249624972498249925002501250225032504250525062507250825092510251125122513251425152516251725182519252025212522252325242525252625272528252925302531253225332534253525362537253825392540254125422543254425452546254725482549255025512552255325542555255625572558255925602561256225632564256525662567256825692570257125722573257425752576257725782579258025812582258325842585258625872588258925902591259225932594259525962597259825992600260126022603260426052606260726082609261026112612261326142615261626172618261926202621262226232624262526262627262826292630263126322633263426352636263726382639264026412642264326442645264626472648264926502651265226532654265526562657265826592660266126622663266426652666266726682669267026712672267326742675267626772678267926802681268226832684268526862687268826892690269126922693269426952696269726982699270027012702270327042705270627072708270927102711271227132714271527162717271827192720272127222723272427252726272727282729273027312732273327342735273627372738273927402741274227432744274527462747274827492750275127522753275427552756275727582759276027612762276327642765276627672768276927702771277227732774277527762777277827792780278127822783278427852786278727882789279027912792279327942795279627972798279928002801280228032804280528062807280828092810281128122813281428152816281728182819282028212822282328242825282628272828282928302831283228332834283528362837283828392840284128422843284428452846284728482849285028512852285328542855285628572858285928602861286228632864286528662867286828692870287128722873287428752876287728782879288028812882288328842885288628872888288928902891289228932894289528962897289828992900290129022903290429052906290729082909291029112912291329142915291629172918291929202921292229232924292529262927292829292930293129322933293429352936293729382939294029412942294329442945294629472948294929502951295229532954295529562957295829592960296129622963296429652966296729682969297029712972297329742975297629772978297929802981298229832984298529862987298829892990299129922993299429952996299729982999300030013002300330043005300630073008300930103011301230133014301530163017301830193020302130223023302430253026302730283029303030313032303330343035303630373038303930403041304230433044304530463047304830493050305130523053305430553056305730583059306030613062306330643065306630673068306930703071307230733074307530763077307830793080308130823083308430853086308730883089309030913092309330943095309630973098309931003101310231033104310531063107310831093110311131123113311431153116311731183119312031213122312331243125312631273128312931303131313231333134313531363137313831393140314131423143314431453146314731483149315031513152315331543155315631573158315931603161316231633164316531663167316831693170317131723173317431753176317731783179318031813182318331843185318631873188318931903191319231933194319531963197319831993200320132023203320432053206320732083209321032113212321332143215321632173218321932203221322232233224322532263227322832293230323132323233323432353236323732383239324032413242324332443245324632473248324932503251325232533254325532563257325832593260326132623263326432653266326732683269327032713272327332743275327632773278327932803281328232833284328532863287328832893290329132923293329432953296329732983299330033013302330333043305330633073308330933103311331233133314331533163317331833193320332133223323332433253326332733283329333033313332333333343335333633373338333933403341334233433344334533463347334833493350335133523353335433553356335733583359336033613362336333643365336633673368336933703371337233733374337533763377337833793380338133823383338433853386338733883389339033913392339333943395339633973398339934003401340234033404340534063407340834093410341134123413341434153416341734183419342034213422342334243425342634273428342934303431343234333434343534363437343834393440344134423443344434453446344734483449345034513452345334543455345634573458345934603461346234633464346534663467346834693470347134723473347434753476347734783479348034813482348334843485348634873488348934903491349234933494349534963497349834993500350135023503350435053506350735083509351035113512351335143515351635173518351935203521352235233524352535263527352835293530353135323533353435353536353735383539354035413542354335443545354635473548354935503551355235533554355535563557355835593560356135623563356435653566356735683569357035713572357335743575357635773578357935803581358235833584358535863587358835893590359135923593359435953596359735983599360036013602360336043605360636073608360936103611361236133614361536163617361836193620362136223623362436253626362736283629363036313632363336343635363636373638363936403641364236433644364536463647364836493650365136523653365436553656365736583659366036613662366336643665366636673668366936703671367236733674367536763677367836793680368136823683368436853686368736883689369036913692369336943695369636973698369937003701370237033704370537063707370837093710371137123713371437153716371737183719372037213722372337243725372637273728372937303731373237333734373537363737373837393740374137423743374437453746374737483749375037513752375337543755375637573758375937603761376237633764376537663767376837693770377137723773377437753776377737783779378037813782378337843785378637873788378937903791379237933794379537963797379837993800380138023803380438053806380738083809381038113812381338143815381638173818381938203821382238233824382538263827382838293830383138323833383438353836383738383839384038413842384338443845384638473848384938503851385238533854385538563857385838593860386138623863386438653866386738683869387038713872387338743875387638773878387938803881388238833884388538863887388838893890389138923893389438953896389738983899390039013902390339043905390639073908390939103911391239133914391539163917391839193920392139223923392439253926392739283929393039313932393339343935393639373938393939403941394239433944394539463947394839493950395139523953395439553956395739583959396039613962396339643965396639673968396939703971397239733974397539763977397839793980398139823983398439853986398739883989399039913992399339943995399639973998399940004001400240034004400540064007400840094010401140124013401440154016401740184019402040214022402340244025402640274028402940304031403240334034403540364037403840394040404140424043404440454046404740484049405040514052405340544055405640574058405940604061406240634064406540664067406840694070407140724073407440754076407740784079408040814082408340844085408640874088408940904091409240934094409540964097409840994100410141024103410441054106410741084109411041114112411341144115411641174118411941204121412241234124412541264127412841294130413141324133413441354136413741384139414041414142414341444145414641474148414941504151415241534154415541564157415841594160416141624163416441654166416741684169417041714172417341744175417641774178417941804181418241834184418541864187418841894190419141924193419441954196419741984199420042014202420342044205420642074208420942104211421242134214421542164217421842194220422142224223422442254226422742284229423042314232423342344235423642374238423942404241424242434244424542464247424842494250425142524253425442554256425742584259426042614262426342644265426642674268426942704271427242734274427542764277427842794280428142824283428442854286428742884289429042914292429342944295429642974298429943004301430243034304430543064307430843094310431143124313431443154316431743184319432043214322432343244325432643274328432943304331433243334334433543364337433843394340434143424343434443454346434743484349435043514352435343544355435643574358435943604361436243634364436543664367436843694370437143724373437443754376437743784379438043814382438343844385438643874388438943904391439243934394439543964397439843994400440144024403440444054406440744084409441044114412441344144415441644174418441944204421442244234424442544264427442844294430443144324433443444354436443744384439444044414442444344444445444644474448444944504451445244534454445544564457445844594460446144624463446444654466446744684469447044714472447344744475447644774478447944804481448244834484448544864487448844894490449144924493449444954496449744984499450045014502450345044505450645074508450945104511451245134514451545164517451845194520452145224523452445254526452745284529453045314532453345344535453645374538453945404541454245434544454545464547454845494550455145524553455445554556455745584559456045614562456345644565456645674568456945704571457245734574457545764577457845794580458145824583458445854586458745884589459045914592459345944595459645974598459946004601460246034604460546064607460846094610461146124613461446154616461746184619462046214622462346244625462646274628462946304631463246334634463546364637463846394640464146424643464446454646464746484649465046514652465346544655465646574658465946604661466246634664466546664667466846694670467146724673467446754676467746784679468046814682468346844685468646874688468946904691469246934694469546964697469846994700470147024703470447054706470747084709471047114712471347144715471647174718471947204721472247234724472547264727472847294730473147324733473447354736473747384739474047414742474347444745474647474748474947504751475247534754475547564757475847594760476147624763476447654766476747684769477047714772477347744775477647774778477947804781478247834784478547864787478847894790479147924793479447954796479747984799480048014802480348044805480648074808480948104811481248134814481548164817481848194820482148224823482448254826482748284829483048314832483348344835483648374838483948404841484248434844484548464847484848494850485148524853485448554856485748584859486048614862486348644865486648674868486948704871487248734874487548764877487848794880488148824883488448854886488748884889489048914892489348944895489648974898489949004901490249034904490549064907490849094910491149124913491449154916491749184919492049214922492349244925492649274928492949304931493249334934493549364937493849394940494149424943494449454946494749484949495049514952495349544955495649574958495949604961496249634964496549664967496849694970497149724973497449754976497749784979498049814982498349844985498649874988498949904991499249934994499549964997499849995000500150025003500450055006500750085009501050115012501350145015501650175018501950205021502250235024502550265027502850295030503150325033503450355036503750385039504050415042504350445045504650475048504950505051505250535054505550565057505850595060506150625063506450655066506750685069507050715072507350745075507650775078507950805081508250835084508550865087508850895090509150925093509450955096509750985099510051015102510351045105510651075108510951105111511251135114511551165117511851195120512151225123512451255126512751285129513051315132513351345135513651375138513951405141514251435144514551465147514851495150515151525153515451555156515751585159516051615162516351645165516651675168516951705171517251735174517551765177517851795180518151825183518451855186518751885189519051915192519351945195519651975198519952005201520252035204520552065207520852095210521152125213521452155216521752185219522052215222522352245225522652275228522952305231523252335234523552365237523852395240524152425243524452455246524752485249525052515252525352545255525652575258525952605261526252635264526552665267526852695270527152725273527452755276527752785279528052815282528352845285528652875288528952905291529252935294529552965297529852995300530153025303530453055306530753085309531053115312531353145315531653175318531953205321532253235324532553265327532853295330533153325333533453355336533753385339534053415342534353445345534653475348534953505351535253535354535553565357535853595360536153625363536453655366536753685369537053715372537353745375537653775378537953805381538253835384538553865387538853895390539153925393539453955396539753985399540054015402540354045405540654075408540954105411541254135414541554165417541854195420542154225423542454255426542754285429543054315432543354345435543654375438543954405441544254435444544554465447544854495450545154525453545454555456545754585459546054615462546354645465546654675468546954705471547254735474547554765477547854795480548154825483548454855486548754885489549054915492549354945495549654975498549955005501550255035504550555065507550855095510551155125513551455155516551755185519552055215522552355245525552655275528552955305531553255335534553555365537553855395540554155425543554455455546554755485549555055515552555355545555555655575558555955605561556255635564556555665567556855695570557155725573557455755576557755785579558055815582558355845585558655875588558955905591559255935594559555965597559855995600560156025603560456055606560756085609561056115612561356145615561656175618561956205621562256235624562556265627562856295630563156325633563456355636563756385639564056415642564356445645564656475648564956505651565256535654565556565657565856595660566156625663566456655666566756685669567056715672567356745675567656775678567956805681568256835684568556865687568856895690569156925693569456955696569756985699570057015702570357045705570657075708570957105711571257135714571557165717571857195720572157225723572457255726572757285729573057315732573357345735573657375738573957405741574257435744574557465747574857495750575157525753575457555756575757585759576057615762576357645765576657675768576957705771577257735774577557765777577857795780578157825783578457855786578757885789579057915792579357945795579657975798579958005801580258035804580558065807580858095810581158125813581458155816581758185819582058215822582358245825582658275828582958305831583258335834583558365837583858395840584158425843584458455846584758485849585058515852585358545855585658575858585958605861586258635864586558665867586858695870587158725873587458755876587758785879588058815882588358845885588658875888588958905891589258935894589558965897589858995900590159025903590459055906590759085909591059115912591359145915591659175918591959205921592259235924592559265927592859295930593159325933593459355936593759385939594059415942594359445945594659475948594959505951595259535954595559565957595859595960596159625963596459655966596759685969597059715972597359745975597659775978597959805981598259835984598559865987598859895990599159925993599459955996599759985999600060016002600360046005600660076008600960106011601260136014601560166017601860196020602160226023602460256026602760286029603060316032603360346035603660376038603960406041604260436044604560466047604860496050605160526053605460556056605760586059606060616062606360646065606660676068606960706071607260736074607560766077607860796080608160826083608460856086608760886089609060916092609360946095609660976098609961006101610261036104610561066107610861096110611161126113611461156116611761186119612061216122612361246125612661276128612961306131613261336134613561366137613861396140614161426143614461456146614761486149615061516152615361546155615661576158615961606161616261636164616561666167616861696170617161726173617461756176617761786179618061816182618361846185618661876188618961906191619261936194619561966197619861996200620162026203620462056206620762086209621062116212621362146215621662176218621962206221622262236224622562266227622862296230623162326233623462356236623762386239624062416242624362446245624662476248624962506251625262536254625562566257625862596260626162626263626462656266626762686269627062716272627362746275627662776278627962806281628262836284628562866287628862896290629162926293629462956296629762986299630063016302630363046305630663076308630963106311631263136314631563166317631863196320632163226323632463256326632763286329633063316332633363346335633663376338633963406341634263436344634563466347634863496350635163526353635463556356635763586359636063616362636363646365636663676368636963706371637263736374637563766377637863796380638163826383638463856386638763886389639063916392639363946395639663976398639964006401640264036404640564066407640864096410641164126413641464156416641764186419642064216422642364246425642664276428642964306431643264336434643564366437643864396440644164426443644464456446644764486449645064516452645364546455645664576458645964606461646264636464646564666467646864696470647164726473647464756476647764786479648064816482648364846485648664876488648964906491649264936494
  1. /*
  2. * kernel/sched.c
  3. *
  4. * Kernel scheduler and related syscalls
  5. *
  6. * Copyright (C) 1991-2002 Linus Torvalds
  7. *
  8. * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and
  9. * make semaphores SMP safe
  10. * 1998-11-19 Implemented schedule_timeout() and related stuff
  11. * by Andrea Arcangeli
  12. * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar:
  13. * hybrid priority-list and round-robin design with
  14. * an array-switch method of distributing timeslices
  15. * and per-CPU runqueues. Cleanups and useful suggestions
  16. * by Davide Libenzi, preemptible kernel bits by Robert Love.
  17. * 2003-09-03 Interactivity tuning by Con Kolivas.
  18. * 2004-04-02 Scheduler domains code by Nick Piggin
  19. * 2007-04-15 Work begun on replacing all interactivity tuning with a
  20. * fair scheduling design by Con Kolivas.
  21. * 2007-05-05 Load balancing (smp-nice) and other improvements
  22. * by Peter Williams
  23. * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith
  24. * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri
  25. */
  26. #include <linux/mm.h>
  27. #include <linux/module.h>
  28. #include <linux/nmi.h>
  29. #include <linux/init.h>
  30. #include <linux/uaccess.h>
  31. #include <linux/highmem.h>
  32. #include <linux/smp_lock.h>
  33. #include <asm/mmu_context.h>
  34. #include <linux/interrupt.h>
  35. #include <linux/capability.h>
  36. #include <linux/completion.h>
  37. #include <linux/kernel_stat.h>
  38. #include <linux/debug_locks.h>
  39. #include <linux/security.h>
  40. #include <linux/notifier.h>
  41. #include <linux/profile.h>
  42. #include <linux/freezer.h>
  43. #include <linux/vmalloc.h>
  44. #include <linux/blkdev.h>
  45. #include <linux/delay.h>
  46. #include <linux/smp.h>
  47. #include <linux/threads.h>
  48. #include <linux/timer.h>
  49. #include <linux/rcupdate.h>
  50. #include <linux/cpu.h>
  51. #include <linux/cpuset.h>
  52. #include <linux/percpu.h>
  53. #include <linux/kthread.h>
  54. #include <linux/seq_file.h>
  55. #include <linux/syscalls.h>
  56. #include <linux/times.h>
  57. #include <linux/tsacct_kern.h>
  58. #include <linux/kprobes.h>
  59. #include <linux/delayacct.h>
  60. #include <linux/reciprocal_div.h>
  61. #include <linux/unistd.h>
  62. #include <asm/tlb.h>
  63. /*
  64. * Scheduler clock - returns current time in nanosec units.
  65. * This is default implementation.
  66. * Architectures and sub-architectures can override this.
  67. */
  68. unsigned long long __attribute__((weak)) sched_clock(void)
  69. {
  70. return (unsigned long long)jiffies * (1000000000 / HZ);
  71. }
  72. /*
  73. * Convert user-nice values [ -20 ... 0 ... 19 ]
  74. * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ],
  75. * and back.
  76. */
  77. #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20)
  78. #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20)
  79. #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio)
  80. /*
  81. * 'User priority' is the nice value converted to something we
  82. * can work with better when scaling various scheduler parameters,
  83. * it's a [ 0 ... 39 ] range.
  84. */
  85. #define USER_PRIO(p) ((p)-MAX_RT_PRIO)
  86. #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio)
  87. #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO))
  88. /*
  89. * Some helpers for converting nanosecond timing to jiffy resolution
  90. */
  91. #define NS_TO_JIFFIES(TIME) ((TIME) / (1000000000 / HZ))
  92. #define JIFFIES_TO_NS(TIME) ((TIME) * (1000000000 / HZ))
  93. #define NICE_0_LOAD SCHED_LOAD_SCALE
  94. #define NICE_0_SHIFT SCHED_LOAD_SHIFT
  95. /*
  96. * These are the 'tuning knobs' of the scheduler:
  97. *
  98. * Minimum timeslice is 5 msecs (or 1 jiffy, whichever is larger),
  99. * default timeslice is 100 msecs, maximum timeslice is 800 msecs.
  100. * Timeslices get refilled after they expire.
  101. */
  102. #define MIN_TIMESLICE max(5 * HZ / 1000, 1)
  103. #define DEF_TIMESLICE (100 * HZ / 1000)
  104. #ifdef CONFIG_SMP
  105. /*
  106. * Divide a load by a sched group cpu_power : (load / sg->__cpu_power)
  107. * Since cpu_power is a 'constant', we can use a reciprocal divide.
  108. */
  109. static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load)
  110. {
  111. return reciprocal_divide(load, sg->reciprocal_cpu_power);
  112. }
  113. /*
  114. * Each time a sched group cpu_power is changed,
  115. * we must compute its reciprocal value
  116. */
  117. static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val)
  118. {
  119. sg->__cpu_power += val;
  120. sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power);
  121. }
  122. #endif
  123. #define SCALE_PRIO(x, prio) \
  124. max(x * (MAX_PRIO - prio) / (MAX_USER_PRIO / 2), MIN_TIMESLICE)
  125. /*
  126. * static_prio_timeslice() scales user-nice values [ -20 ... 0 ... 19 ]
  127. * to time slice values: [800ms ... 100ms ... 5ms]
  128. */
  129. static unsigned int static_prio_timeslice(int static_prio)
  130. {
  131. if (static_prio == NICE_TO_PRIO(19))
  132. return 1;
  133. if (static_prio < NICE_TO_PRIO(0))
  134. return SCALE_PRIO(DEF_TIMESLICE * 4, static_prio);
  135. else
  136. return SCALE_PRIO(DEF_TIMESLICE, static_prio);
  137. }
  138. static inline int rt_policy(int policy)
  139. {
  140. if (unlikely(policy == SCHED_FIFO) || unlikely(policy == SCHED_RR))
  141. return 1;
  142. return 0;
  143. }
  144. static inline int task_has_rt_policy(struct task_struct *p)
  145. {
  146. return rt_policy(p->policy);
  147. }
  148. /*
  149. * This is the priority-queue data structure of the RT scheduling class:
  150. */
  151. struct rt_prio_array {
  152. DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */
  153. struct list_head queue[MAX_RT_PRIO];
  154. };
  155. struct load_stat {
  156. struct load_weight load;
  157. u64 load_update_start, load_update_last;
  158. unsigned long delta_fair, delta_exec, delta_stat;
  159. };
  160. /* CFS-related fields in a runqueue */
  161. struct cfs_rq {
  162. struct load_weight load;
  163. unsigned long nr_running;
  164. s64 fair_clock;
  165. u64 exec_clock;
  166. s64 wait_runtime;
  167. u64 sleeper_bonus;
  168. unsigned long wait_runtime_overruns, wait_runtime_underruns;
  169. struct rb_root tasks_timeline;
  170. struct rb_node *rb_leftmost;
  171. struct rb_node *rb_load_balance_curr;
  172. #ifdef CONFIG_FAIR_GROUP_SCHED
  173. /* 'curr' points to currently running entity on this cfs_rq.
  174. * It is set to NULL otherwise (i.e when none are currently running).
  175. */
  176. struct sched_entity *curr;
  177. struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */
  178. /* leaf cfs_rqs are those that hold tasks (lowest schedulable entity in
  179. * a hierarchy). Non-leaf lrqs hold other higher schedulable entities
  180. * (like users, containers etc.)
  181. *
  182. * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This
  183. * list is used during load balance.
  184. */
  185. struct list_head leaf_cfs_rq_list; /* Better name : task_cfs_rq_list? */
  186. #endif
  187. };
  188. /* Real-Time classes' related field in a runqueue: */
  189. struct rt_rq {
  190. struct rt_prio_array active;
  191. int rt_load_balance_idx;
  192. struct list_head *rt_load_balance_head, *rt_load_balance_curr;
  193. };
  194. /*
  195. * This is the main, per-CPU runqueue data structure.
  196. *
  197. * Locking rule: those places that want to lock multiple runqueues
  198. * (such as the load balancing or the thread migration code), lock
  199. * acquire operations must be ordered by ascending &runqueue.
  200. */
  201. struct rq {
  202. spinlock_t lock; /* runqueue lock */
  203. /*
  204. * nr_running and cpu_load should be in the same cacheline because
  205. * remote CPUs use both these fields when doing load calculation.
  206. */
  207. unsigned long nr_running;
  208. #define CPU_LOAD_IDX_MAX 5
  209. unsigned long cpu_load[CPU_LOAD_IDX_MAX];
  210. unsigned char idle_at_tick;
  211. #ifdef CONFIG_NO_HZ
  212. unsigned char in_nohz_recently;
  213. #endif
  214. struct load_stat ls; /* capture load from *all* tasks on this cpu */
  215. unsigned long nr_load_updates;
  216. u64 nr_switches;
  217. struct cfs_rq cfs;
  218. #ifdef CONFIG_FAIR_GROUP_SCHED
  219. struct list_head leaf_cfs_rq_list; /* list of leaf cfs_rq on this cpu */
  220. #endif
  221. struct rt_rq rt;
  222. /*
  223. * This is part of a global counter where only the total sum
  224. * over all CPUs matters. A task can increase this counter on
  225. * one CPU and if it got migrated afterwards it may decrease
  226. * it on another CPU. Always updated under the runqueue lock:
  227. */
  228. unsigned long nr_uninterruptible;
  229. struct task_struct *curr, *idle;
  230. unsigned long next_balance;
  231. struct mm_struct *prev_mm;
  232. u64 clock, prev_clock_raw;
  233. s64 clock_max_delta;
  234. unsigned int clock_warps, clock_overflows;
  235. unsigned int clock_unstable_events;
  236. struct sched_class *load_balance_class;
  237. atomic_t nr_iowait;
  238. #ifdef CONFIG_SMP
  239. struct sched_domain *sd;
  240. /* For active balancing */
  241. int active_balance;
  242. int push_cpu;
  243. int cpu; /* cpu of this runqueue */
  244. struct task_struct *migration_thread;
  245. struct list_head migration_queue;
  246. #endif
  247. #ifdef CONFIG_SCHEDSTATS
  248. /* latency stats */
  249. struct sched_info rq_sched_info;
  250. /* sys_sched_yield() stats */
  251. unsigned long yld_exp_empty;
  252. unsigned long yld_act_empty;
  253. unsigned long yld_both_empty;
  254. unsigned long yld_cnt;
  255. /* schedule() stats */
  256. unsigned long sched_switch;
  257. unsigned long sched_cnt;
  258. unsigned long sched_goidle;
  259. /* try_to_wake_up() stats */
  260. unsigned long ttwu_cnt;
  261. unsigned long ttwu_local;
  262. #endif
  263. struct lock_class_key rq_lock_key;
  264. };
  265. static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
  266. static DEFINE_MUTEX(sched_hotcpu_mutex);
  267. static inline void check_preempt_curr(struct rq *rq, struct task_struct *p)
  268. {
  269. rq->curr->sched_class->check_preempt_curr(rq, p);
  270. }
  271. static inline int cpu_of(struct rq *rq)
  272. {
  273. #ifdef CONFIG_SMP
  274. return rq->cpu;
  275. #else
  276. return 0;
  277. #endif
  278. }
  279. /*
  280. * Per-runqueue clock, as finegrained as the platform can give us:
  281. */
  282. static unsigned long long __rq_clock(struct rq *rq)
  283. {
  284. u64 prev_raw = rq->prev_clock_raw;
  285. u64 now = sched_clock();
  286. s64 delta = now - prev_raw;
  287. u64 clock = rq->clock;
  288. /*
  289. * Protect against sched_clock() occasionally going backwards:
  290. */
  291. if (unlikely(delta < 0)) {
  292. clock++;
  293. rq->clock_warps++;
  294. } else {
  295. /*
  296. * Catch too large forward jumps too:
  297. */
  298. if (unlikely(delta > 2*TICK_NSEC)) {
  299. clock++;
  300. rq->clock_overflows++;
  301. } else {
  302. if (unlikely(delta > rq->clock_max_delta))
  303. rq->clock_max_delta = delta;
  304. clock += delta;
  305. }
  306. }
  307. rq->prev_clock_raw = now;
  308. rq->clock = clock;
  309. return clock;
  310. }
  311. static inline unsigned long long rq_clock(struct rq *rq)
  312. {
  313. int this_cpu = smp_processor_id();
  314. if (this_cpu == cpu_of(rq))
  315. return __rq_clock(rq);
  316. return rq->clock;
  317. }
  318. /*
  319. * The domain tree (rq->sd) is protected by RCU's quiescent state transition.
  320. * See detach_destroy_domains: synchronize_sched for details.
  321. *
  322. * The domain tree of any CPU may only be accessed from within
  323. * preempt-disabled sections.
  324. */
  325. #define for_each_domain(cpu, __sd) \
  326. for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent)
  327. #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu)))
  328. #define this_rq() (&__get_cpu_var(runqueues))
  329. #define task_rq(p) cpu_rq(task_cpu(p))
  330. #define cpu_curr(cpu) (cpu_rq(cpu)->curr)
  331. /*
  332. * For kernel-internal use: high-speed (but slightly incorrect) per-cpu
  333. * clock constructed from sched_clock():
  334. */
  335. unsigned long long cpu_clock(int cpu)
  336. {
  337. struct rq *rq = cpu_rq(cpu);
  338. unsigned long long now;
  339. unsigned long flags;
  340. spin_lock_irqsave(&rq->lock, flags);
  341. now = rq_clock(rq);
  342. spin_unlock_irqrestore(&rq->lock, flags);
  343. return now;
  344. }
  345. #ifdef CONFIG_FAIR_GROUP_SCHED
  346. /* Change a task's ->cfs_rq if it moves across CPUs */
  347. static inline void set_task_cfs_rq(struct task_struct *p)
  348. {
  349. p->se.cfs_rq = &task_rq(p)->cfs;
  350. }
  351. #else
  352. static inline void set_task_cfs_rq(struct task_struct *p)
  353. {
  354. }
  355. #endif
  356. #ifndef prepare_arch_switch
  357. # define prepare_arch_switch(next) do { } while (0)
  358. #endif
  359. #ifndef finish_arch_switch
  360. # define finish_arch_switch(prev) do { } while (0)
  361. #endif
  362. #ifndef __ARCH_WANT_UNLOCKED_CTXSW
  363. static inline int task_running(struct rq *rq, struct task_struct *p)
  364. {
  365. return rq->curr == p;
  366. }
  367. static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
  368. {
  369. }
  370. static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
  371. {
  372. #ifdef CONFIG_DEBUG_SPINLOCK
  373. /* this is a valid case when another task releases the spinlock */
  374. rq->lock.owner = current;
  375. #endif
  376. /*
  377. * If we are tracking spinlock dependencies then we have to
  378. * fix up the runqueue lock - which gets 'carried over' from
  379. * prev into current:
  380. */
  381. spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);
  382. spin_unlock_irq(&rq->lock);
  383. }
  384. #else /* __ARCH_WANT_UNLOCKED_CTXSW */
  385. static inline int task_running(struct rq *rq, struct task_struct *p)
  386. {
  387. #ifdef CONFIG_SMP
  388. return p->oncpu;
  389. #else
  390. return rq->curr == p;
  391. #endif
  392. }
  393. static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
  394. {
  395. #ifdef CONFIG_SMP
  396. /*
  397. * We can optimise this out completely for !SMP, because the
  398. * SMP rebalancing from interrupt is the only thing that cares
  399. * here.
  400. */
  401. next->oncpu = 1;
  402. #endif
  403. #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
  404. spin_unlock_irq(&rq->lock);
  405. #else
  406. spin_unlock(&rq->lock);
  407. #endif
  408. }
  409. static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
  410. {
  411. #ifdef CONFIG_SMP
  412. /*
  413. * After ->oncpu is cleared, the task can be moved to a different CPU.
  414. * We must ensure this doesn't happen until the switch is completely
  415. * finished.
  416. */
  417. smp_wmb();
  418. prev->oncpu = 0;
  419. #endif
  420. #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW
  421. local_irq_enable();
  422. #endif
  423. }
  424. #endif /* __ARCH_WANT_UNLOCKED_CTXSW */
  425. /*
  426. * __task_rq_lock - lock the runqueue a given task resides on.
  427. * Must be called interrupts disabled.
  428. */
  429. static inline struct rq *__task_rq_lock(struct task_struct *p)
  430. __acquires(rq->lock)
  431. {
  432. struct rq *rq;
  433. repeat_lock_task:
  434. rq = task_rq(p);
  435. spin_lock(&rq->lock);
  436. if (unlikely(rq != task_rq(p))) {
  437. spin_unlock(&rq->lock);
  438. goto repeat_lock_task;
  439. }
  440. return rq;
  441. }
  442. /*
  443. * task_rq_lock - lock the runqueue a given task resides on and disable
  444. * interrupts. Note the ordering: we can safely lookup the task_rq without
  445. * explicitly disabling preemption.
  446. */
  447. static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags)
  448. __acquires(rq->lock)
  449. {
  450. struct rq *rq;
  451. repeat_lock_task:
  452. local_irq_save(*flags);
  453. rq = task_rq(p);
  454. spin_lock(&rq->lock);
  455. if (unlikely(rq != task_rq(p))) {
  456. spin_unlock_irqrestore(&rq->lock, *flags);
  457. goto repeat_lock_task;
  458. }
  459. return rq;
  460. }
  461. static inline void __task_rq_unlock(struct rq *rq)
  462. __releases(rq->lock)
  463. {
  464. spin_unlock(&rq->lock);
  465. }
  466. static inline void task_rq_unlock(struct rq *rq, unsigned long *flags)
  467. __releases(rq->lock)
  468. {
  469. spin_unlock_irqrestore(&rq->lock, *flags);
  470. }
  471. /*
  472. * this_rq_lock - lock this runqueue and disable interrupts.
  473. */
  474. static inline struct rq *this_rq_lock(void)
  475. __acquires(rq->lock)
  476. {
  477. struct rq *rq;
  478. local_irq_disable();
  479. rq = this_rq();
  480. spin_lock(&rq->lock);
  481. return rq;
  482. }
  483. /*
  484. * CPU frequency is/was unstable - start new by setting prev_clock_raw:
  485. */
  486. void sched_clock_unstable_event(void)
  487. {
  488. unsigned long flags;
  489. struct rq *rq;
  490. rq = task_rq_lock(current, &flags);
  491. rq->prev_clock_raw = sched_clock();
  492. rq->clock_unstable_events++;
  493. task_rq_unlock(rq, &flags);
  494. }
  495. /*
  496. * resched_task - mark a task 'to be rescheduled now'.
  497. *
  498. * On UP this means the setting of the need_resched flag, on SMP it
  499. * might also involve a cross-CPU call to trigger the scheduler on
  500. * the target CPU.
  501. */
  502. #ifdef CONFIG_SMP
  503. #ifndef tsk_is_polling
  504. #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG)
  505. #endif
  506. static void resched_task(struct task_struct *p)
  507. {
  508. int cpu;
  509. assert_spin_locked(&task_rq(p)->lock);
  510. if (unlikely(test_tsk_thread_flag(p, TIF_NEED_RESCHED)))
  511. return;
  512. set_tsk_thread_flag(p, TIF_NEED_RESCHED);
  513. cpu = task_cpu(p);
  514. if (cpu == smp_processor_id())
  515. return;
  516. /* NEED_RESCHED must be visible before we test polling */
  517. smp_mb();
  518. if (!tsk_is_polling(p))
  519. smp_send_reschedule(cpu);
  520. }
  521. static void resched_cpu(int cpu)
  522. {
  523. struct rq *rq = cpu_rq(cpu);
  524. unsigned long flags;
  525. if (!spin_trylock_irqsave(&rq->lock, flags))
  526. return;
  527. resched_task(cpu_curr(cpu));
  528. spin_unlock_irqrestore(&rq->lock, flags);
  529. }
  530. #else
  531. static inline void resched_task(struct task_struct *p)
  532. {
  533. assert_spin_locked(&task_rq(p)->lock);
  534. set_tsk_need_resched(p);
  535. }
  536. #endif
  537. static u64 div64_likely32(u64 divident, unsigned long divisor)
  538. {
  539. #if BITS_PER_LONG == 32
  540. if (likely(divident <= 0xffffffffULL))
  541. return (u32)divident / divisor;
  542. do_div(divident, divisor);
  543. return divident;
  544. #else
  545. return divident / divisor;
  546. #endif
  547. }
  548. #if BITS_PER_LONG == 32
  549. # define WMULT_CONST (~0UL)
  550. #else
  551. # define WMULT_CONST (1UL << 32)
  552. #endif
  553. #define WMULT_SHIFT 32
  554. static inline unsigned long
  555. calc_delta_mine(unsigned long delta_exec, unsigned long weight,
  556. struct load_weight *lw)
  557. {
  558. u64 tmp;
  559. if (unlikely(!lw->inv_weight))
  560. lw->inv_weight = WMULT_CONST / lw->weight;
  561. tmp = (u64)delta_exec * weight;
  562. /*
  563. * Check whether we'd overflow the 64-bit multiplication:
  564. */
  565. if (unlikely(tmp > WMULT_CONST)) {
  566. tmp = ((tmp >> WMULT_SHIFT/2) * lw->inv_weight)
  567. >> (WMULT_SHIFT/2);
  568. } else {
  569. tmp = (tmp * lw->inv_weight) >> WMULT_SHIFT;
  570. }
  571. return (unsigned long)min(tmp, (u64)sysctl_sched_runtime_limit);
  572. }
  573. static inline unsigned long
  574. calc_delta_fair(unsigned long delta_exec, struct load_weight *lw)
  575. {
  576. return calc_delta_mine(delta_exec, NICE_0_LOAD, lw);
  577. }
  578. static void update_load_add(struct load_weight *lw, unsigned long inc)
  579. {
  580. lw->weight += inc;
  581. lw->inv_weight = 0;
  582. }
  583. static void update_load_sub(struct load_weight *lw, unsigned long dec)
  584. {
  585. lw->weight -= dec;
  586. lw->inv_weight = 0;
  587. }
  588. static void __update_curr_load(struct rq *rq, struct load_stat *ls)
  589. {
  590. if (rq->curr != rq->idle && ls->load.weight) {
  591. ls->delta_exec += ls->delta_stat;
  592. ls->delta_fair += calc_delta_fair(ls->delta_stat, &ls->load);
  593. ls->delta_stat = 0;
  594. }
  595. }
  596. /*
  597. * Update delta_exec, delta_fair fields for rq.
  598. *
  599. * delta_fair clock advances at a rate inversely proportional to
  600. * total load (rq->ls.load.weight) on the runqueue, while
  601. * delta_exec advances at the same rate as wall-clock (provided
  602. * cpu is not idle).
  603. *
  604. * delta_exec / delta_fair is a measure of the (smoothened) load on this
  605. * runqueue over any given interval. This (smoothened) load is used
  606. * during load balance.
  607. *
  608. * This function is called /before/ updating rq->ls.load
  609. * and when switching tasks.
  610. */
  611. static void update_curr_load(struct rq *rq, u64 now)
  612. {
  613. struct load_stat *ls = &rq->ls;
  614. u64 start;
  615. start = ls->load_update_start;
  616. ls->load_update_start = now;
  617. ls->delta_stat += now - start;
  618. /*
  619. * Stagger updates to ls->delta_fair. Very frequent updates
  620. * can be expensive.
  621. */
  622. if (ls->delta_stat >= sysctl_sched_stat_granularity)
  623. __update_curr_load(rq, ls);
  624. }
  625. /*
  626. * To aid in avoiding the subversion of "niceness" due to uneven distribution
  627. * of tasks with abnormal "nice" values across CPUs the contribution that
  628. * each task makes to its run queue's load is weighted according to its
  629. * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
  630. * scaled version of the new time slice allocation that they receive on time
  631. * slice expiry etc.
  632. */
  633. /*
  634. * Assume: static_prio_timeslice(NICE_TO_PRIO(0)) == DEF_TIMESLICE
  635. * If static_prio_timeslice() is ever changed to break this assumption then
  636. * this code will need modification
  637. */
  638. #define TIME_SLICE_NICE_ZERO DEF_TIMESLICE
  639. #define load_weight(lp) \
  640. (((lp) * SCHED_LOAD_SCALE) / TIME_SLICE_NICE_ZERO)
  641. #define PRIO_TO_LOAD_WEIGHT(prio) \
  642. load_weight(static_prio_timeslice(prio))
  643. #define RTPRIO_TO_LOAD_WEIGHT(rp) \
  644. (PRIO_TO_LOAD_WEIGHT(MAX_RT_PRIO) + load_weight(rp))
  645. #define WEIGHT_IDLEPRIO 2
  646. #define WMULT_IDLEPRIO (1 << 31)
  647. /*
  648. * Nice levels are multiplicative, with a gentle 10% change for every
  649. * nice level changed. I.e. when a CPU-bound task goes from nice 0 to
  650. * nice 1, it will get ~10% less CPU time than another CPU-bound task
  651. * that remained on nice 0.
  652. *
  653. * The "10% effect" is relative and cumulative: from _any_ nice level,
  654. * if you go up 1 level, it's -10% CPU usage, if you go down 1 level
  655. * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25.
  656. * If a task goes up by ~10% and another task goes down by ~10% then
  657. * the relative distance between them is ~25%.)
  658. */
  659. static const int prio_to_weight[40] = {
  660. /* -20 */ 88818, 71054, 56843, 45475, 36380, 29104, 23283, 18626, 14901, 11921,
  661. /* -10 */ 9537, 7629, 6103, 4883, 3906, 3125, 2500, 2000, 1600, 1280,
  662. /* 0 */ NICE_0_LOAD /* 1024 */,
  663. /* 1 */ 819, 655, 524, 419, 336, 268, 215, 172, 137,
  664. /* 10 */ 110, 87, 70, 56, 45, 36, 29, 23, 18, 15,
  665. };
  666. /*
  667. * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated.
  668. *
  669. * In cases where the weight does not change often, we can use the
  670. * precalculated inverse to speed up arithmetics by turning divisions
  671. * into multiplications:
  672. */
  673. static const u32 prio_to_wmult[40] = {
  674. /* -20 */ 48356, 60446, 75558, 94446, 118058,
  675. /* -15 */ 147573, 184467, 230589, 288233, 360285,
  676. /* -10 */ 450347, 562979, 703746, 879575, 1099582,
  677. /* -5 */ 1374389, 1717986, 2147483, 2684354, 3355443,
  678. /* 0 */ 4194304, 5244160, 6557201, 8196502, 10250518,
  679. /* 5 */ 12782640, 16025997, 19976592, 24970740, 31350126,
  680. /* 10 */ 39045157, 49367440, 61356675, 76695844, 95443717,
  681. /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153,
  682. };
  683. static inline void
  684. inc_load(struct rq *rq, const struct task_struct *p, u64 now)
  685. {
  686. update_curr_load(rq, now);
  687. update_load_add(&rq->ls.load, p->se.load.weight);
  688. }
  689. static inline void
  690. dec_load(struct rq *rq, const struct task_struct *p, u64 now)
  691. {
  692. update_curr_load(rq, now);
  693. update_load_sub(&rq->ls.load, p->se.load.weight);
  694. }
  695. static inline void inc_nr_running(struct task_struct *p, struct rq *rq, u64 now)
  696. {
  697. rq->nr_running++;
  698. inc_load(rq, p, now);
  699. }
  700. static inline void dec_nr_running(struct task_struct *p, struct rq *rq, u64 now)
  701. {
  702. rq->nr_running--;
  703. dec_load(rq, p, now);
  704. }
  705. static void activate_task(struct rq *rq, struct task_struct *p, int wakeup);
  706. /*
  707. * runqueue iterator, to support SMP load-balancing between different
  708. * scheduling classes, without having to expose their internal data
  709. * structures to the load-balancing proper:
  710. */
  711. struct rq_iterator {
  712. void *arg;
  713. struct task_struct *(*start)(void *);
  714. struct task_struct *(*next)(void *);
  715. };
  716. static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest,
  717. unsigned long max_nr_move, unsigned long max_load_move,
  718. struct sched_domain *sd, enum cpu_idle_type idle,
  719. int *all_pinned, unsigned long *load_moved,
  720. int this_best_prio, int best_prio, int best_prio_seen,
  721. struct rq_iterator *iterator);
  722. #include "sched_stats.h"
  723. #include "sched_rt.c"
  724. #include "sched_fair.c"
  725. #include "sched_idletask.c"
  726. #ifdef CONFIG_SCHED_DEBUG
  727. # include "sched_debug.c"
  728. #endif
  729. #define sched_class_highest (&rt_sched_class)
  730. static void set_load_weight(struct task_struct *p)
  731. {
  732. task_rq(p)->cfs.wait_runtime -= p->se.wait_runtime;
  733. p->se.wait_runtime = 0;
  734. if (task_has_rt_policy(p)) {
  735. p->se.load.weight = prio_to_weight[0] * 2;
  736. p->se.load.inv_weight = prio_to_wmult[0] >> 1;
  737. return;
  738. }
  739. /*
  740. * SCHED_IDLE tasks get minimal weight:
  741. */
  742. if (p->policy == SCHED_IDLE) {
  743. p->se.load.weight = WEIGHT_IDLEPRIO;
  744. p->se.load.inv_weight = WMULT_IDLEPRIO;
  745. return;
  746. }
  747. p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO];
  748. p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO];
  749. }
  750. static void
  751. enqueue_task(struct rq *rq, struct task_struct *p, int wakeup, u64 now)
  752. {
  753. sched_info_queued(p);
  754. p->sched_class->enqueue_task(rq, p, wakeup, now);
  755. p->se.on_rq = 1;
  756. }
  757. static void
  758. dequeue_task(struct rq *rq, struct task_struct *p, int sleep, u64 now)
  759. {
  760. p->sched_class->dequeue_task(rq, p, sleep, now);
  761. p->se.on_rq = 0;
  762. }
  763. /*
  764. * __normal_prio - return the priority that is based on the static prio
  765. */
  766. static inline int __normal_prio(struct task_struct *p)
  767. {
  768. return p->static_prio;
  769. }
  770. /*
  771. * Calculate the expected normal priority: i.e. priority
  772. * without taking RT-inheritance into account. Might be
  773. * boosted by interactivity modifiers. Changes upon fork,
  774. * setprio syscalls, and whenever the interactivity
  775. * estimator recalculates.
  776. */
  777. static inline int normal_prio(struct task_struct *p)
  778. {
  779. int prio;
  780. if (task_has_rt_policy(p))
  781. prio = MAX_RT_PRIO-1 - p->rt_priority;
  782. else
  783. prio = __normal_prio(p);
  784. return prio;
  785. }
  786. /*
  787. * Calculate the current priority, i.e. the priority
  788. * taken into account by the scheduler. This value might
  789. * be boosted by RT tasks, or might be boosted by
  790. * interactivity modifiers. Will be RT if the task got
  791. * RT-boosted. If not then it returns p->normal_prio.
  792. */
  793. static int effective_prio(struct task_struct *p)
  794. {
  795. p->normal_prio = normal_prio(p);
  796. /*
  797. * If we are RT tasks or we were boosted to RT priority,
  798. * keep the priority unchanged. Otherwise, update priority
  799. * to the normal priority:
  800. */
  801. if (!rt_prio(p->prio))
  802. return p->normal_prio;
  803. return p->prio;
  804. }
  805. /*
  806. * activate_task - move a task to the runqueue.
  807. */
  808. static void activate_task(struct rq *rq, struct task_struct *p, int wakeup)
  809. {
  810. u64 now = rq_clock(rq);
  811. if (p->state == TASK_UNINTERRUPTIBLE)
  812. rq->nr_uninterruptible--;
  813. enqueue_task(rq, p, wakeup, now);
  814. inc_nr_running(p, rq, now);
  815. }
  816. /*
  817. * activate_idle_task - move idle task to the _front_ of runqueue.
  818. */
  819. static inline void activate_idle_task(struct task_struct *p, struct rq *rq)
  820. {
  821. u64 now = rq_clock(rq);
  822. if (p->state == TASK_UNINTERRUPTIBLE)
  823. rq->nr_uninterruptible--;
  824. enqueue_task(rq, p, 0, now);
  825. inc_nr_running(p, rq, now);
  826. }
  827. /*
  828. * deactivate_task - remove a task from the runqueue.
  829. */
  830. static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep)
  831. {
  832. u64 now = rq_clock(rq);
  833. if (p->state == TASK_UNINTERRUPTIBLE)
  834. rq->nr_uninterruptible++;
  835. dequeue_task(rq, p, sleep, now);
  836. dec_nr_running(p, rq, now);
  837. }
  838. /**
  839. * task_curr - is this task currently executing on a CPU?
  840. * @p: the task in question.
  841. */
  842. inline int task_curr(const struct task_struct *p)
  843. {
  844. return cpu_curr(task_cpu(p)) == p;
  845. }
  846. /* Used instead of source_load when we know the type == 0 */
  847. unsigned long weighted_cpuload(const int cpu)
  848. {
  849. return cpu_rq(cpu)->ls.load.weight;
  850. }
  851. static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
  852. {
  853. #ifdef CONFIG_SMP
  854. task_thread_info(p)->cpu = cpu;
  855. set_task_cfs_rq(p);
  856. #endif
  857. }
  858. #ifdef CONFIG_SMP
  859. void set_task_cpu(struct task_struct *p, unsigned int new_cpu)
  860. {
  861. int old_cpu = task_cpu(p);
  862. struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu);
  863. u64 clock_offset, fair_clock_offset;
  864. clock_offset = old_rq->clock - new_rq->clock;
  865. fair_clock_offset = old_rq->cfs.fair_clock -
  866. new_rq->cfs.fair_clock;
  867. if (p->se.wait_start)
  868. p->se.wait_start -= clock_offset;
  869. if (p->se.wait_start_fair)
  870. p->se.wait_start_fair -= fair_clock_offset;
  871. if (p->se.sleep_start)
  872. p->se.sleep_start -= clock_offset;
  873. if (p->se.block_start)
  874. p->se.block_start -= clock_offset;
  875. if (p->se.sleep_start_fair)
  876. p->se.sleep_start_fair -= fair_clock_offset;
  877. __set_task_cpu(p, new_cpu);
  878. }
  879. struct migration_req {
  880. struct list_head list;
  881. struct task_struct *task;
  882. int dest_cpu;
  883. struct completion done;
  884. };
  885. /*
  886. * The task's runqueue lock must be held.
  887. * Returns true if you have to wait for migration thread.
  888. */
  889. static int
  890. migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req)
  891. {
  892. struct rq *rq = task_rq(p);
  893. /*
  894. * If the task is not on a runqueue (and not running), then
  895. * it is sufficient to simply update the task's cpu field.
  896. */
  897. if (!p->se.on_rq && !task_running(rq, p)) {
  898. set_task_cpu(p, dest_cpu);
  899. return 0;
  900. }
  901. init_completion(&req->done);
  902. req->task = p;
  903. req->dest_cpu = dest_cpu;
  904. list_add(&req->list, &rq->migration_queue);
  905. return 1;
  906. }
  907. /*
  908. * wait_task_inactive - wait for a thread to unschedule.
  909. *
  910. * The caller must ensure that the task *will* unschedule sometime soon,
  911. * else this function might spin for a *long* time. This function can't
  912. * be called with interrupts off, or it may introduce deadlock with
  913. * smp_call_function() if an IPI is sent by the same process we are
  914. * waiting to become inactive.
  915. */
  916. void wait_task_inactive(struct task_struct *p)
  917. {
  918. unsigned long flags;
  919. int running, on_rq;
  920. struct rq *rq;
  921. repeat:
  922. /*
  923. * We do the initial early heuristics without holding
  924. * any task-queue locks at all. We'll only try to get
  925. * the runqueue lock when things look like they will
  926. * work out!
  927. */
  928. rq = task_rq(p);
  929. /*
  930. * If the task is actively running on another CPU
  931. * still, just relax and busy-wait without holding
  932. * any locks.
  933. *
  934. * NOTE! Since we don't hold any locks, it's not
  935. * even sure that "rq" stays as the right runqueue!
  936. * But we don't care, since "task_running()" will
  937. * return false if the runqueue has changed and p
  938. * is actually now running somewhere else!
  939. */
  940. while (task_running(rq, p))
  941. cpu_relax();
  942. /*
  943. * Ok, time to look more closely! We need the rq
  944. * lock now, to be *sure*. If we're wrong, we'll
  945. * just go back and repeat.
  946. */
  947. rq = task_rq_lock(p, &flags);
  948. running = task_running(rq, p);
  949. on_rq = p->se.on_rq;
  950. task_rq_unlock(rq, &flags);
  951. /*
  952. * Was it really running after all now that we
  953. * checked with the proper locks actually held?
  954. *
  955. * Oops. Go back and try again..
  956. */
  957. if (unlikely(running)) {
  958. cpu_relax();
  959. goto repeat;
  960. }
  961. /*
  962. * It's not enough that it's not actively running,
  963. * it must be off the runqueue _entirely_, and not
  964. * preempted!
  965. *
  966. * So if it wa still runnable (but just not actively
  967. * running right now), it's preempted, and we should
  968. * yield - it could be a while.
  969. */
  970. if (unlikely(on_rq)) {
  971. yield();
  972. goto repeat;
  973. }
  974. /*
  975. * Ahh, all good. It wasn't running, and it wasn't
  976. * runnable, which means that it will never become
  977. * running in the future either. We're all done!
  978. */
  979. }
  980. /***
  981. * kick_process - kick a running thread to enter/exit the kernel
  982. * @p: the to-be-kicked thread
  983. *
  984. * Cause a process which is running on another CPU to enter
  985. * kernel-mode, without any delay. (to get signals handled.)
  986. *
  987. * NOTE: this function doesnt have to take the runqueue lock,
  988. * because all it wants to ensure is that the remote task enters
  989. * the kernel. If the IPI races and the task has been migrated
  990. * to another CPU then no harm is done and the purpose has been
  991. * achieved as well.
  992. */
  993. void kick_process(struct task_struct *p)
  994. {
  995. int cpu;
  996. preempt_disable();
  997. cpu = task_cpu(p);
  998. if ((cpu != smp_processor_id()) && task_curr(p))
  999. smp_send_reschedule(cpu);
  1000. preempt_enable();
  1001. }
  1002. /*
  1003. * Return a low guess at the load of a migration-source cpu weighted
  1004. * according to the scheduling class and "nice" value.
  1005. *
  1006. * We want to under-estimate the load of migration sources, to
  1007. * balance conservatively.
  1008. */
  1009. static inline unsigned long source_load(int cpu, int type)
  1010. {
  1011. struct rq *rq = cpu_rq(cpu);
  1012. unsigned long total = weighted_cpuload(cpu);
  1013. if (type == 0)
  1014. return total;
  1015. return min(rq->cpu_load[type-1], total);
  1016. }
  1017. /*
  1018. * Return a high guess at the load of a migration-target cpu weighted
  1019. * according to the scheduling class and "nice" value.
  1020. */
  1021. static inline unsigned long target_load(int cpu, int type)
  1022. {
  1023. struct rq *rq = cpu_rq(cpu);
  1024. unsigned long total = weighted_cpuload(cpu);
  1025. if (type == 0)
  1026. return total;
  1027. return max(rq->cpu_load[type-1], total);
  1028. }
  1029. /*
  1030. * Return the average load per task on the cpu's run queue
  1031. */
  1032. static inline unsigned long cpu_avg_load_per_task(int cpu)
  1033. {
  1034. struct rq *rq = cpu_rq(cpu);
  1035. unsigned long total = weighted_cpuload(cpu);
  1036. unsigned long n = rq->nr_running;
  1037. return n ? total / n : SCHED_LOAD_SCALE;
  1038. }
  1039. /*
  1040. * find_idlest_group finds and returns the least busy CPU group within the
  1041. * domain.
  1042. */
  1043. static struct sched_group *
  1044. find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu)
  1045. {
  1046. struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups;
  1047. unsigned long min_load = ULONG_MAX, this_load = 0;
  1048. int load_idx = sd->forkexec_idx;
  1049. int imbalance = 100 + (sd->imbalance_pct-100)/2;
  1050. do {
  1051. unsigned long load, avg_load;
  1052. int local_group;
  1053. int i;
  1054. /* Skip over this group if it has no CPUs allowed */
  1055. if (!cpus_intersects(group->cpumask, p->cpus_allowed))
  1056. goto nextgroup;
  1057. local_group = cpu_isset(this_cpu, group->cpumask);
  1058. /* Tally up the load of all CPUs in the group */
  1059. avg_load = 0;
  1060. for_each_cpu_mask(i, group->cpumask) {
  1061. /* Bias balancing toward cpus of our domain */
  1062. if (local_group)
  1063. load = source_load(i, load_idx);
  1064. else
  1065. load = target_load(i, load_idx);
  1066. avg_load += load;
  1067. }
  1068. /* Adjust by relative CPU power of the group */
  1069. avg_load = sg_div_cpu_power(group,
  1070. avg_load * SCHED_LOAD_SCALE);
  1071. if (local_group) {
  1072. this_load = avg_load;
  1073. this = group;
  1074. } else if (avg_load < min_load) {
  1075. min_load = avg_load;
  1076. idlest = group;
  1077. }
  1078. nextgroup:
  1079. group = group->next;
  1080. } while (group != sd->groups);
  1081. if (!idlest || 100*this_load < imbalance*min_load)
  1082. return NULL;
  1083. return idlest;
  1084. }
  1085. /*
  1086. * find_idlest_cpu - find the idlest cpu among the cpus in group.
  1087. */
  1088. static int
  1089. find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu)
  1090. {
  1091. cpumask_t tmp;
  1092. unsigned long load, min_load = ULONG_MAX;
  1093. int idlest = -1;
  1094. int i;
  1095. /* Traverse only the allowed CPUs */
  1096. cpus_and(tmp, group->cpumask, p->cpus_allowed);
  1097. for_each_cpu_mask(i, tmp) {
  1098. load = weighted_cpuload(i);
  1099. if (load < min_load || (load == min_load && i == this_cpu)) {
  1100. min_load = load;
  1101. idlest = i;
  1102. }
  1103. }
  1104. return idlest;
  1105. }
  1106. /*
  1107. * sched_balance_self: balance the current task (running on cpu) in domains
  1108. * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and
  1109. * SD_BALANCE_EXEC.
  1110. *
  1111. * Balance, ie. select the least loaded group.
  1112. *
  1113. * Returns the target CPU number, or the same CPU if no balancing is needed.
  1114. *
  1115. * preempt must be disabled.
  1116. */
  1117. static int sched_balance_self(int cpu, int flag)
  1118. {
  1119. struct task_struct *t = current;
  1120. struct sched_domain *tmp, *sd = NULL;
  1121. for_each_domain(cpu, tmp) {
  1122. /*
  1123. * If power savings logic is enabled for a domain, stop there.
  1124. */
  1125. if (tmp->flags & SD_POWERSAVINGS_BALANCE)
  1126. break;
  1127. if (tmp->flags & flag)
  1128. sd = tmp;
  1129. }
  1130. while (sd) {
  1131. cpumask_t span;
  1132. struct sched_group *group;
  1133. int new_cpu, weight;
  1134. if (!(sd->flags & flag)) {
  1135. sd = sd->child;
  1136. continue;
  1137. }
  1138. span = sd->span;
  1139. group = find_idlest_group(sd, t, cpu);
  1140. if (!group) {
  1141. sd = sd->child;
  1142. continue;
  1143. }
  1144. new_cpu = find_idlest_cpu(group, t, cpu);
  1145. if (new_cpu == -1 || new_cpu == cpu) {
  1146. /* Now try balancing at a lower domain level of cpu */
  1147. sd = sd->child;
  1148. continue;
  1149. }
  1150. /* Now try balancing at a lower domain level of new_cpu */
  1151. cpu = new_cpu;
  1152. sd = NULL;
  1153. weight = cpus_weight(span);
  1154. for_each_domain(cpu, tmp) {
  1155. if (weight <= cpus_weight(tmp->span))
  1156. break;
  1157. if (tmp->flags & flag)
  1158. sd = tmp;
  1159. }
  1160. /* while loop will break here if sd == NULL */
  1161. }
  1162. return cpu;
  1163. }
  1164. #endif /* CONFIG_SMP */
  1165. /*
  1166. * wake_idle() will wake a task on an idle cpu if task->cpu is
  1167. * not idle and an idle cpu is available. The span of cpus to
  1168. * search starts with cpus closest then further out as needed,
  1169. * so we always favor a closer, idle cpu.
  1170. *
  1171. * Returns the CPU we should wake onto.
  1172. */
  1173. #if defined(ARCH_HAS_SCHED_WAKE_IDLE)
  1174. static int wake_idle(int cpu, struct task_struct *p)
  1175. {
  1176. cpumask_t tmp;
  1177. struct sched_domain *sd;
  1178. int i;
  1179. /*
  1180. * If it is idle, then it is the best cpu to run this task.
  1181. *
  1182. * This cpu is also the best, if it has more than one task already.
  1183. * Siblings must be also busy(in most cases) as they didn't already
  1184. * pickup the extra load from this cpu and hence we need not check
  1185. * sibling runqueue info. This will avoid the checks and cache miss
  1186. * penalities associated with that.
  1187. */
  1188. if (idle_cpu(cpu) || cpu_rq(cpu)->nr_running > 1)
  1189. return cpu;
  1190. for_each_domain(cpu, sd) {
  1191. if (sd->flags & SD_WAKE_IDLE) {
  1192. cpus_and(tmp, sd->span, p->cpus_allowed);
  1193. for_each_cpu_mask(i, tmp) {
  1194. if (idle_cpu(i))
  1195. return i;
  1196. }
  1197. } else {
  1198. break;
  1199. }
  1200. }
  1201. return cpu;
  1202. }
  1203. #else
  1204. static inline int wake_idle(int cpu, struct task_struct *p)
  1205. {
  1206. return cpu;
  1207. }
  1208. #endif
  1209. /***
  1210. * try_to_wake_up - wake up a thread
  1211. * @p: the to-be-woken-up thread
  1212. * @state: the mask of task states that can be woken
  1213. * @sync: do a synchronous wakeup?
  1214. *
  1215. * Put it on the run-queue if it's not already there. The "current"
  1216. * thread is always on the run-queue (except when the actual
  1217. * re-schedule is in progress), and as such you're allowed to do
  1218. * the simpler "current->state = TASK_RUNNING" to mark yourself
  1219. * runnable without the overhead of this.
  1220. *
  1221. * returns failure only if the task is already active.
  1222. */
  1223. static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync)
  1224. {
  1225. int cpu, this_cpu, success = 0;
  1226. unsigned long flags;
  1227. long old_state;
  1228. struct rq *rq;
  1229. #ifdef CONFIG_SMP
  1230. struct sched_domain *sd, *this_sd = NULL;
  1231. unsigned long load, this_load;
  1232. int new_cpu;
  1233. #endif
  1234. rq = task_rq_lock(p, &flags);
  1235. old_state = p->state;
  1236. if (!(old_state & state))
  1237. goto out;
  1238. if (p->se.on_rq)
  1239. goto out_running;
  1240. cpu = task_cpu(p);
  1241. this_cpu = smp_processor_id();
  1242. #ifdef CONFIG_SMP
  1243. if (unlikely(task_running(rq, p)))
  1244. goto out_activate;
  1245. new_cpu = cpu;
  1246. schedstat_inc(rq, ttwu_cnt);
  1247. if (cpu == this_cpu) {
  1248. schedstat_inc(rq, ttwu_local);
  1249. goto out_set_cpu;
  1250. }
  1251. for_each_domain(this_cpu, sd) {
  1252. if (cpu_isset(cpu, sd->span)) {
  1253. schedstat_inc(sd, ttwu_wake_remote);
  1254. this_sd = sd;
  1255. break;
  1256. }
  1257. }
  1258. if (unlikely(!cpu_isset(this_cpu, p->cpus_allowed)))
  1259. goto out_set_cpu;
  1260. /*
  1261. * Check for affine wakeup and passive balancing possibilities.
  1262. */
  1263. if (this_sd) {
  1264. int idx = this_sd->wake_idx;
  1265. unsigned int imbalance;
  1266. imbalance = 100 + (this_sd->imbalance_pct - 100) / 2;
  1267. load = source_load(cpu, idx);
  1268. this_load = target_load(this_cpu, idx);
  1269. new_cpu = this_cpu; /* Wake to this CPU if we can */
  1270. if (this_sd->flags & SD_WAKE_AFFINE) {
  1271. unsigned long tl = this_load;
  1272. unsigned long tl_per_task;
  1273. tl_per_task = cpu_avg_load_per_task(this_cpu);
  1274. /*
  1275. * If sync wakeup then subtract the (maximum possible)
  1276. * effect of the currently running task from the load
  1277. * of the current CPU:
  1278. */
  1279. if (sync)
  1280. tl -= current->se.load.weight;
  1281. if ((tl <= load &&
  1282. tl + target_load(cpu, idx) <= tl_per_task) ||
  1283. 100*(tl + p->se.load.weight) <= imbalance*load) {
  1284. /*
  1285. * This domain has SD_WAKE_AFFINE and
  1286. * p is cache cold in this domain, and
  1287. * there is no bad imbalance.
  1288. */
  1289. schedstat_inc(this_sd, ttwu_move_affine);
  1290. goto out_set_cpu;
  1291. }
  1292. }
  1293. /*
  1294. * Start passive balancing when half the imbalance_pct
  1295. * limit is reached.
  1296. */
  1297. if (this_sd->flags & SD_WAKE_BALANCE) {
  1298. if (imbalance*this_load <= 100*load) {
  1299. schedstat_inc(this_sd, ttwu_move_balance);
  1300. goto out_set_cpu;
  1301. }
  1302. }
  1303. }
  1304. new_cpu = cpu; /* Could not wake to this_cpu. Wake to cpu instead */
  1305. out_set_cpu:
  1306. new_cpu = wake_idle(new_cpu, p);
  1307. if (new_cpu != cpu) {
  1308. set_task_cpu(p, new_cpu);
  1309. task_rq_unlock(rq, &flags);
  1310. /* might preempt at this point */
  1311. rq = task_rq_lock(p, &flags);
  1312. old_state = p->state;
  1313. if (!(old_state & state))
  1314. goto out;
  1315. if (p->se.on_rq)
  1316. goto out_running;
  1317. this_cpu = smp_processor_id();
  1318. cpu = task_cpu(p);
  1319. }
  1320. out_activate:
  1321. #endif /* CONFIG_SMP */
  1322. activate_task(rq, p, 1);
  1323. /*
  1324. * Sync wakeups (i.e. those types of wakeups where the waker
  1325. * has indicated that it will leave the CPU in short order)
  1326. * don't trigger a preemption, if the woken up task will run on
  1327. * this cpu. (in this case the 'I will reschedule' promise of
  1328. * the waker guarantees that the freshly woken up task is going
  1329. * to be considered on this CPU.)
  1330. */
  1331. if (!sync || cpu != this_cpu)
  1332. check_preempt_curr(rq, p);
  1333. success = 1;
  1334. out_running:
  1335. p->state = TASK_RUNNING;
  1336. out:
  1337. task_rq_unlock(rq, &flags);
  1338. return success;
  1339. }
  1340. int fastcall wake_up_process(struct task_struct *p)
  1341. {
  1342. return try_to_wake_up(p, TASK_STOPPED | TASK_TRACED |
  1343. TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE, 0);
  1344. }
  1345. EXPORT_SYMBOL(wake_up_process);
  1346. int fastcall wake_up_state(struct task_struct *p, unsigned int state)
  1347. {
  1348. return try_to_wake_up(p, state, 0);
  1349. }
  1350. /*
  1351. * Perform scheduler related setup for a newly forked process p.
  1352. * p is forked by current.
  1353. *
  1354. * __sched_fork() is basic setup used by init_idle() too:
  1355. */
  1356. static void __sched_fork(struct task_struct *p)
  1357. {
  1358. p->se.wait_start_fair = 0;
  1359. p->se.wait_start = 0;
  1360. p->se.exec_start = 0;
  1361. p->se.sum_exec_runtime = 0;
  1362. p->se.delta_exec = 0;
  1363. p->se.delta_fair_run = 0;
  1364. p->se.delta_fair_sleep = 0;
  1365. p->se.wait_runtime = 0;
  1366. p->se.sum_wait_runtime = 0;
  1367. p->se.sum_sleep_runtime = 0;
  1368. p->se.sleep_start = 0;
  1369. p->se.sleep_start_fair = 0;
  1370. p->se.block_start = 0;
  1371. p->se.sleep_max = 0;
  1372. p->se.block_max = 0;
  1373. p->se.exec_max = 0;
  1374. p->se.wait_max = 0;
  1375. p->se.wait_runtime_overruns = 0;
  1376. p->se.wait_runtime_underruns = 0;
  1377. INIT_LIST_HEAD(&p->run_list);
  1378. p->se.on_rq = 0;
  1379. /*
  1380. * We mark the process as running here, but have not actually
  1381. * inserted it onto the runqueue yet. This guarantees that
  1382. * nobody will actually run it, and a signal or other external
  1383. * event cannot wake it up and insert it on the runqueue either.
  1384. */
  1385. p->state = TASK_RUNNING;
  1386. }
  1387. /*
  1388. * fork()/clone()-time setup:
  1389. */
  1390. void sched_fork(struct task_struct *p, int clone_flags)
  1391. {
  1392. int cpu = get_cpu();
  1393. __sched_fork(p);
  1394. #ifdef CONFIG_SMP
  1395. cpu = sched_balance_self(cpu, SD_BALANCE_FORK);
  1396. #endif
  1397. __set_task_cpu(p, cpu);
  1398. /*
  1399. * Make sure we do not leak PI boosting priority to the child:
  1400. */
  1401. p->prio = current->normal_prio;
  1402. #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT)
  1403. if (likely(sched_info_on()))
  1404. memset(&p->sched_info, 0, sizeof(p->sched_info));
  1405. #endif
  1406. #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW)
  1407. p->oncpu = 0;
  1408. #endif
  1409. #ifdef CONFIG_PREEMPT
  1410. /* Want to start with kernel preemption disabled. */
  1411. task_thread_info(p)->preempt_count = 1;
  1412. #endif
  1413. put_cpu();
  1414. }
  1415. /*
  1416. * After fork, child runs first. (default) If set to 0 then
  1417. * parent will (try to) run first.
  1418. */
  1419. unsigned int __read_mostly sysctl_sched_child_runs_first = 1;
  1420. /*
  1421. * wake_up_new_task - wake up a newly created task for the first time.
  1422. *
  1423. * This function will do some initial scheduler statistics housekeeping
  1424. * that must be done for every newly created context, then puts the task
  1425. * on the runqueue and wakes it.
  1426. */
  1427. void fastcall wake_up_new_task(struct task_struct *p, unsigned long clone_flags)
  1428. {
  1429. unsigned long flags;
  1430. struct rq *rq;
  1431. int this_cpu;
  1432. rq = task_rq_lock(p, &flags);
  1433. BUG_ON(p->state != TASK_RUNNING);
  1434. this_cpu = smp_processor_id(); /* parent's CPU */
  1435. p->prio = effective_prio(p);
  1436. if (!sysctl_sched_child_runs_first || (clone_flags & CLONE_VM) ||
  1437. task_cpu(p) != this_cpu || !current->se.on_rq) {
  1438. activate_task(rq, p, 0);
  1439. } else {
  1440. /*
  1441. * Let the scheduling class do new task startup
  1442. * management (if any):
  1443. */
  1444. p->sched_class->task_new(rq, p);
  1445. }
  1446. check_preempt_curr(rq, p);
  1447. task_rq_unlock(rq, &flags);
  1448. }
  1449. /**
  1450. * prepare_task_switch - prepare to switch tasks
  1451. * @rq: the runqueue preparing to switch
  1452. * @next: the task we are going to switch to.
  1453. *
  1454. * This is called with the rq lock held and interrupts off. It must
  1455. * be paired with a subsequent finish_task_switch after the context
  1456. * switch.
  1457. *
  1458. * prepare_task_switch sets up locking and calls architecture specific
  1459. * hooks.
  1460. */
  1461. static inline void prepare_task_switch(struct rq *rq, struct task_struct *next)
  1462. {
  1463. prepare_lock_switch(rq, next);
  1464. prepare_arch_switch(next);
  1465. }
  1466. /**
  1467. * finish_task_switch - clean up after a task-switch
  1468. * @rq: runqueue associated with task-switch
  1469. * @prev: the thread we just switched away from.
  1470. *
  1471. * finish_task_switch must be called after the context switch, paired
  1472. * with a prepare_task_switch call before the context switch.
  1473. * finish_task_switch will reconcile locking set up by prepare_task_switch,
  1474. * and do any other architecture-specific cleanup actions.
  1475. *
  1476. * Note that we may have delayed dropping an mm in context_switch(). If
  1477. * so, we finish that here outside of the runqueue lock. (Doing it
  1478. * with the lock held can cause deadlocks; see schedule() for
  1479. * details.)
  1480. */
  1481. static inline void finish_task_switch(struct rq *rq, struct task_struct *prev)
  1482. __releases(rq->lock)
  1483. {
  1484. struct mm_struct *mm = rq->prev_mm;
  1485. long prev_state;
  1486. rq->prev_mm = NULL;
  1487. /*
  1488. * A task struct has one reference for the use as "current".
  1489. * If a task dies, then it sets TASK_DEAD in tsk->state and calls
  1490. * schedule one last time. The schedule call will never return, and
  1491. * the scheduled task must drop that reference.
  1492. * The test for TASK_DEAD must occur while the runqueue locks are
  1493. * still held, otherwise prev could be scheduled on another cpu, die
  1494. * there before we look at prev->state, and then the reference would
  1495. * be dropped twice.
  1496. * Manfred Spraul <manfred@colorfullife.com>
  1497. */
  1498. prev_state = prev->state;
  1499. finish_arch_switch(prev);
  1500. finish_lock_switch(rq, prev);
  1501. if (mm)
  1502. mmdrop(mm);
  1503. if (unlikely(prev_state == TASK_DEAD)) {
  1504. /*
  1505. * Remove function-return probe instances associated with this
  1506. * task and put them back on the free list.
  1507. */
  1508. kprobe_flush_task(prev);
  1509. put_task_struct(prev);
  1510. }
  1511. }
  1512. /**
  1513. * schedule_tail - first thing a freshly forked thread must call.
  1514. * @prev: the thread we just switched away from.
  1515. */
  1516. asmlinkage void schedule_tail(struct task_struct *prev)
  1517. __releases(rq->lock)
  1518. {
  1519. struct rq *rq = this_rq();
  1520. finish_task_switch(rq, prev);
  1521. #ifdef __ARCH_WANT_UNLOCKED_CTXSW
  1522. /* In this case, finish_task_switch does not reenable preemption */
  1523. preempt_enable();
  1524. #endif
  1525. if (current->set_child_tid)
  1526. put_user(current->pid, current->set_child_tid);
  1527. }
  1528. /*
  1529. * context_switch - switch to the new MM and the new
  1530. * thread's register state.
  1531. */
  1532. static inline void
  1533. context_switch(struct rq *rq, struct task_struct *prev,
  1534. struct task_struct *next)
  1535. {
  1536. struct mm_struct *mm, *oldmm;
  1537. prepare_task_switch(rq, next);
  1538. mm = next->mm;
  1539. oldmm = prev->active_mm;
  1540. /*
  1541. * For paravirt, this is coupled with an exit in switch_to to
  1542. * combine the page table reload and the switch backend into
  1543. * one hypercall.
  1544. */
  1545. arch_enter_lazy_cpu_mode();
  1546. if (unlikely(!mm)) {
  1547. next->active_mm = oldmm;
  1548. atomic_inc(&oldmm->mm_count);
  1549. enter_lazy_tlb(oldmm, next);
  1550. } else
  1551. switch_mm(oldmm, mm, next);
  1552. if (unlikely(!prev->mm)) {
  1553. prev->active_mm = NULL;
  1554. rq->prev_mm = oldmm;
  1555. }
  1556. /*
  1557. * Since the runqueue lock will be released by the next
  1558. * task (which is an invalid locking op but in the case
  1559. * of the scheduler it's an obvious special-case), so we
  1560. * do an early lockdep release here:
  1561. */
  1562. #ifndef __ARCH_WANT_UNLOCKED_CTXSW
  1563. spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
  1564. #endif
  1565. /* Here we just switch the register state and the stack. */
  1566. switch_to(prev, next, prev);
  1567. barrier();
  1568. /*
  1569. * this_rq must be evaluated again because prev may have moved
  1570. * CPUs since it called schedule(), thus the 'rq' on its stack
  1571. * frame will be invalid.
  1572. */
  1573. finish_task_switch(this_rq(), prev);
  1574. }
  1575. /*
  1576. * nr_running, nr_uninterruptible and nr_context_switches:
  1577. *
  1578. * externally visible scheduler statistics: current number of runnable
  1579. * threads, current number of uninterruptible-sleeping threads, total
  1580. * number of context switches performed since bootup.
  1581. */
  1582. unsigned long nr_running(void)
  1583. {
  1584. unsigned long i, sum = 0;
  1585. for_each_online_cpu(i)
  1586. sum += cpu_rq(i)->nr_running;
  1587. return sum;
  1588. }
  1589. unsigned long nr_uninterruptible(void)
  1590. {
  1591. unsigned long i, sum = 0;
  1592. for_each_possible_cpu(i)
  1593. sum += cpu_rq(i)->nr_uninterruptible;
  1594. /*
  1595. * Since we read the counters lockless, it might be slightly
  1596. * inaccurate. Do not allow it to go below zero though:
  1597. */
  1598. if (unlikely((long)sum < 0))
  1599. sum = 0;
  1600. return sum;
  1601. }
  1602. unsigned long long nr_context_switches(void)
  1603. {
  1604. int i;
  1605. unsigned long long sum = 0;
  1606. for_each_possible_cpu(i)
  1607. sum += cpu_rq(i)->nr_switches;
  1608. return sum;
  1609. }
  1610. unsigned long nr_iowait(void)
  1611. {
  1612. unsigned long i, sum = 0;
  1613. for_each_possible_cpu(i)
  1614. sum += atomic_read(&cpu_rq(i)->nr_iowait);
  1615. return sum;
  1616. }
  1617. unsigned long nr_active(void)
  1618. {
  1619. unsigned long i, running = 0, uninterruptible = 0;
  1620. for_each_online_cpu(i) {
  1621. running += cpu_rq(i)->nr_running;
  1622. uninterruptible += cpu_rq(i)->nr_uninterruptible;
  1623. }
  1624. if (unlikely((long)uninterruptible < 0))
  1625. uninterruptible = 0;
  1626. return running + uninterruptible;
  1627. }
  1628. /*
  1629. * Update rq->cpu_load[] statistics. This function is usually called every
  1630. * scheduler tick (TICK_NSEC).
  1631. */
  1632. static void update_cpu_load(struct rq *this_rq)
  1633. {
  1634. u64 fair_delta64, exec_delta64, idle_delta64, sample_interval64, tmp64;
  1635. unsigned long total_load = this_rq->ls.load.weight;
  1636. unsigned long this_load = total_load;
  1637. struct load_stat *ls = &this_rq->ls;
  1638. u64 now = __rq_clock(this_rq);
  1639. int i, scale;
  1640. this_rq->nr_load_updates++;
  1641. if (unlikely(!(sysctl_sched_features & SCHED_FEAT_PRECISE_CPU_LOAD)))
  1642. goto do_avg;
  1643. /* Update delta_fair/delta_exec fields first */
  1644. update_curr_load(this_rq, now);
  1645. fair_delta64 = ls->delta_fair + 1;
  1646. ls->delta_fair = 0;
  1647. exec_delta64 = ls->delta_exec + 1;
  1648. ls->delta_exec = 0;
  1649. sample_interval64 = now - ls->load_update_last;
  1650. ls->load_update_last = now;
  1651. if ((s64)sample_interval64 < (s64)TICK_NSEC)
  1652. sample_interval64 = TICK_NSEC;
  1653. if (exec_delta64 > sample_interval64)
  1654. exec_delta64 = sample_interval64;
  1655. idle_delta64 = sample_interval64 - exec_delta64;
  1656. tmp64 = div64_64(SCHED_LOAD_SCALE * exec_delta64, fair_delta64);
  1657. tmp64 = div64_64(tmp64 * exec_delta64, sample_interval64);
  1658. this_load = (unsigned long)tmp64;
  1659. do_avg:
  1660. /* Update our load: */
  1661. for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) {
  1662. unsigned long old_load, new_load;
  1663. /* scale is effectively 1 << i now, and >> i divides by scale */
  1664. old_load = this_rq->cpu_load[i];
  1665. new_load = this_load;
  1666. this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i;
  1667. }
  1668. }
  1669. #ifdef CONFIG_SMP
  1670. /*
  1671. * double_rq_lock - safely lock two runqueues
  1672. *
  1673. * Note this does not disable interrupts like task_rq_lock,
  1674. * you need to do so manually before calling.
  1675. */
  1676. static void double_rq_lock(struct rq *rq1, struct rq *rq2)
  1677. __acquires(rq1->lock)
  1678. __acquires(rq2->lock)
  1679. {
  1680. BUG_ON(!irqs_disabled());
  1681. if (rq1 == rq2) {
  1682. spin_lock(&rq1->lock);
  1683. __acquire(rq2->lock); /* Fake it out ;) */
  1684. } else {
  1685. if (rq1 < rq2) {
  1686. spin_lock(&rq1->lock);
  1687. spin_lock(&rq2->lock);
  1688. } else {
  1689. spin_lock(&rq2->lock);
  1690. spin_lock(&rq1->lock);
  1691. }
  1692. }
  1693. }
  1694. /*
  1695. * double_rq_unlock - safely unlock two runqueues
  1696. *
  1697. * Note this does not restore interrupts like task_rq_unlock,
  1698. * you need to do so manually after calling.
  1699. */
  1700. static void double_rq_unlock(struct rq *rq1, struct rq *rq2)
  1701. __releases(rq1->lock)
  1702. __releases(rq2->lock)
  1703. {
  1704. spin_unlock(&rq1->lock);
  1705. if (rq1 != rq2)
  1706. spin_unlock(&rq2->lock);
  1707. else
  1708. __release(rq2->lock);
  1709. }
  1710. /*
  1711. * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
  1712. */
  1713. static void double_lock_balance(struct rq *this_rq, struct rq *busiest)
  1714. __releases(this_rq->lock)
  1715. __acquires(busiest->lock)
  1716. __acquires(this_rq->lock)
  1717. {
  1718. if (unlikely(!irqs_disabled())) {
  1719. /* printk() doesn't work good under rq->lock */
  1720. spin_unlock(&this_rq->lock);
  1721. BUG_ON(1);
  1722. }
  1723. if (unlikely(!spin_trylock(&busiest->lock))) {
  1724. if (busiest < this_rq) {
  1725. spin_unlock(&this_rq->lock);
  1726. spin_lock(&busiest->lock);
  1727. spin_lock(&this_rq->lock);
  1728. } else
  1729. spin_lock(&busiest->lock);
  1730. }
  1731. }
  1732. /*
  1733. * If dest_cpu is allowed for this process, migrate the task to it.
  1734. * This is accomplished by forcing the cpu_allowed mask to only
  1735. * allow dest_cpu, which will force the cpu onto dest_cpu. Then
  1736. * the cpu_allowed mask is restored.
  1737. */
  1738. static void sched_migrate_task(struct task_struct *p, int dest_cpu)
  1739. {
  1740. struct migration_req req;
  1741. unsigned long flags;
  1742. struct rq *rq;
  1743. rq = task_rq_lock(p, &flags);
  1744. if (!cpu_isset(dest_cpu, p->cpus_allowed)
  1745. || unlikely(cpu_is_offline(dest_cpu)))
  1746. goto out;
  1747. /* force the process onto the specified CPU */
  1748. if (migrate_task(p, dest_cpu, &req)) {
  1749. /* Need to wait for migration thread (might exit: take ref). */
  1750. struct task_struct *mt = rq->migration_thread;
  1751. get_task_struct(mt);
  1752. task_rq_unlock(rq, &flags);
  1753. wake_up_process(mt);
  1754. put_task_struct(mt);
  1755. wait_for_completion(&req.done);
  1756. return;
  1757. }
  1758. out:
  1759. task_rq_unlock(rq, &flags);
  1760. }
  1761. /*
  1762. * sched_exec - execve() is a valuable balancing opportunity, because at
  1763. * this point the task has the smallest effective memory and cache footprint.
  1764. */
  1765. void sched_exec(void)
  1766. {
  1767. int new_cpu, this_cpu = get_cpu();
  1768. new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC);
  1769. put_cpu();
  1770. if (new_cpu != this_cpu)
  1771. sched_migrate_task(current, new_cpu);
  1772. }
  1773. /*
  1774. * pull_task - move a task from a remote runqueue to the local runqueue.
  1775. * Both runqueues must be locked.
  1776. */
  1777. static void pull_task(struct rq *src_rq, struct task_struct *p,
  1778. struct rq *this_rq, int this_cpu)
  1779. {
  1780. deactivate_task(src_rq, p, 0);
  1781. set_task_cpu(p, this_cpu);
  1782. activate_task(this_rq, p, 0);
  1783. /*
  1784. * Note that idle threads have a prio of MAX_PRIO, for this test
  1785. * to be always true for them.
  1786. */
  1787. check_preempt_curr(this_rq, p);
  1788. }
  1789. /*
  1790. * can_migrate_task - may task p from runqueue rq be migrated to this_cpu?
  1791. */
  1792. static
  1793. int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu,
  1794. struct sched_domain *sd, enum cpu_idle_type idle,
  1795. int *all_pinned)
  1796. {
  1797. /*
  1798. * We do not migrate tasks that are:
  1799. * 1) running (obviously), or
  1800. * 2) cannot be migrated to this CPU due to cpus_allowed, or
  1801. * 3) are cache-hot on their current CPU.
  1802. */
  1803. if (!cpu_isset(this_cpu, p->cpus_allowed))
  1804. return 0;
  1805. *all_pinned = 0;
  1806. if (task_running(rq, p))
  1807. return 0;
  1808. /*
  1809. * Aggressive migration if too many balance attempts have failed:
  1810. */
  1811. if (sd->nr_balance_failed > sd->cache_nice_tries)
  1812. return 1;
  1813. return 1;
  1814. }
  1815. static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest,
  1816. unsigned long max_nr_move, unsigned long max_load_move,
  1817. struct sched_domain *sd, enum cpu_idle_type idle,
  1818. int *all_pinned, unsigned long *load_moved,
  1819. int this_best_prio, int best_prio, int best_prio_seen,
  1820. struct rq_iterator *iterator)
  1821. {
  1822. int pulled = 0, pinned = 0, skip_for_load;
  1823. struct task_struct *p;
  1824. long rem_load_move = max_load_move;
  1825. if (max_nr_move == 0 || max_load_move == 0)
  1826. goto out;
  1827. pinned = 1;
  1828. /*
  1829. * Start the load-balancing iterator:
  1830. */
  1831. p = iterator->start(iterator->arg);
  1832. next:
  1833. if (!p)
  1834. goto out;
  1835. /*
  1836. * To help distribute high priority tasks accross CPUs we don't
  1837. * skip a task if it will be the highest priority task (i.e. smallest
  1838. * prio value) on its new queue regardless of its load weight
  1839. */
  1840. skip_for_load = (p->se.load.weight >> 1) > rem_load_move +
  1841. SCHED_LOAD_SCALE_FUZZ;
  1842. if (skip_for_load && p->prio < this_best_prio)
  1843. skip_for_load = !best_prio_seen && p->prio == best_prio;
  1844. if (skip_for_load ||
  1845. !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) {
  1846. best_prio_seen |= p->prio == best_prio;
  1847. p = iterator->next(iterator->arg);
  1848. goto next;
  1849. }
  1850. pull_task(busiest, p, this_rq, this_cpu);
  1851. pulled++;
  1852. rem_load_move -= p->se.load.weight;
  1853. /*
  1854. * We only want to steal up to the prescribed number of tasks
  1855. * and the prescribed amount of weighted load.
  1856. */
  1857. if (pulled < max_nr_move && rem_load_move > 0) {
  1858. if (p->prio < this_best_prio)
  1859. this_best_prio = p->prio;
  1860. p = iterator->next(iterator->arg);
  1861. goto next;
  1862. }
  1863. out:
  1864. /*
  1865. * Right now, this is the only place pull_task() is called,
  1866. * so we can safely collect pull_task() stats here rather than
  1867. * inside pull_task().
  1868. */
  1869. schedstat_add(sd, lb_gained[idle], pulled);
  1870. if (all_pinned)
  1871. *all_pinned = pinned;
  1872. *load_moved = max_load_move - rem_load_move;
  1873. return pulled;
  1874. }
  1875. /*
  1876. * move_tasks tries to move up to max_nr_move tasks and max_load_move weighted
  1877. * load from busiest to this_rq, as part of a balancing operation within
  1878. * "domain". Returns the number of tasks moved.
  1879. *
  1880. * Called with both runqueues locked.
  1881. */
  1882. static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest,
  1883. unsigned long max_nr_move, unsigned long max_load_move,
  1884. struct sched_domain *sd, enum cpu_idle_type idle,
  1885. int *all_pinned)
  1886. {
  1887. struct sched_class *class = sched_class_highest;
  1888. unsigned long load_moved, total_nr_moved = 0, nr_moved;
  1889. long rem_load_move = max_load_move;
  1890. do {
  1891. nr_moved = class->load_balance(this_rq, this_cpu, busiest,
  1892. max_nr_move, (unsigned long)rem_load_move,
  1893. sd, idle, all_pinned, &load_moved);
  1894. total_nr_moved += nr_moved;
  1895. max_nr_move -= nr_moved;
  1896. rem_load_move -= load_moved;
  1897. class = class->next;
  1898. } while (class && max_nr_move && rem_load_move > 0);
  1899. return total_nr_moved;
  1900. }
  1901. /*
  1902. * find_busiest_group finds and returns the busiest CPU group within the
  1903. * domain. It calculates and returns the amount of weighted load which
  1904. * should be moved to restore balance via the imbalance parameter.
  1905. */
  1906. static struct sched_group *
  1907. find_busiest_group(struct sched_domain *sd, int this_cpu,
  1908. unsigned long *imbalance, enum cpu_idle_type idle,
  1909. int *sd_idle, cpumask_t *cpus, int *balance)
  1910. {
  1911. struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups;
  1912. unsigned long max_load, avg_load, total_load, this_load, total_pwr;
  1913. unsigned long max_pull;
  1914. unsigned long busiest_load_per_task, busiest_nr_running;
  1915. unsigned long this_load_per_task, this_nr_running;
  1916. int load_idx;
  1917. #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT)
  1918. int power_savings_balance = 1;
  1919. unsigned long leader_nr_running = 0, min_load_per_task = 0;
  1920. unsigned long min_nr_running = ULONG_MAX;
  1921. struct sched_group *group_min = NULL, *group_leader = NULL;
  1922. #endif
  1923. max_load = this_load = total_load = total_pwr = 0;
  1924. busiest_load_per_task = busiest_nr_running = 0;
  1925. this_load_per_task = this_nr_running = 0;
  1926. if (idle == CPU_NOT_IDLE)
  1927. load_idx = sd->busy_idx;
  1928. else if (idle == CPU_NEWLY_IDLE)
  1929. load_idx = sd->newidle_idx;
  1930. else
  1931. load_idx = sd->idle_idx;
  1932. do {
  1933. unsigned long load, group_capacity;
  1934. int local_group;
  1935. int i;
  1936. unsigned int balance_cpu = -1, first_idle_cpu = 0;
  1937. unsigned long sum_nr_running, sum_weighted_load;
  1938. local_group = cpu_isset(this_cpu, group->cpumask);
  1939. if (local_group)
  1940. balance_cpu = first_cpu(group->cpumask);
  1941. /* Tally up the load of all CPUs in the group */
  1942. sum_weighted_load = sum_nr_running = avg_load = 0;
  1943. for_each_cpu_mask(i, group->cpumask) {
  1944. struct rq *rq;
  1945. if (!cpu_isset(i, *cpus))
  1946. continue;
  1947. rq = cpu_rq(i);
  1948. if (*sd_idle && rq->nr_running)
  1949. *sd_idle = 0;
  1950. /* Bias balancing toward cpus of our domain */
  1951. if (local_group) {
  1952. if (idle_cpu(i) && !first_idle_cpu) {
  1953. first_idle_cpu = 1;
  1954. balance_cpu = i;
  1955. }
  1956. load = target_load(i, load_idx);
  1957. } else
  1958. load = source_load(i, load_idx);
  1959. avg_load += load;
  1960. sum_nr_running += rq->nr_running;
  1961. sum_weighted_load += weighted_cpuload(i);
  1962. }
  1963. /*
  1964. * First idle cpu or the first cpu(busiest) in this sched group
  1965. * is eligible for doing load balancing at this and above
  1966. * domains. In the newly idle case, we will allow all the cpu's
  1967. * to do the newly idle load balance.
  1968. */
  1969. if (idle != CPU_NEWLY_IDLE && local_group &&
  1970. balance_cpu != this_cpu && balance) {
  1971. *balance = 0;
  1972. goto ret;
  1973. }
  1974. total_load += avg_load;
  1975. total_pwr += group->__cpu_power;
  1976. /* Adjust by relative CPU power of the group */
  1977. avg_load = sg_div_cpu_power(group,
  1978. avg_load * SCHED_LOAD_SCALE);
  1979. group_capacity = group->__cpu_power / SCHED_LOAD_SCALE;
  1980. if (local_group) {
  1981. this_load = avg_load;
  1982. this = group;
  1983. this_nr_running = sum_nr_running;
  1984. this_load_per_task = sum_weighted_load;
  1985. } else if (avg_load > max_load &&
  1986. sum_nr_running > group_capacity) {
  1987. max_load = avg_load;
  1988. busiest = group;
  1989. busiest_nr_running = sum_nr_running;
  1990. busiest_load_per_task = sum_weighted_load;
  1991. }
  1992. #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT)
  1993. /*
  1994. * Busy processors will not participate in power savings
  1995. * balance.
  1996. */
  1997. if (idle == CPU_NOT_IDLE ||
  1998. !(sd->flags & SD_POWERSAVINGS_BALANCE))
  1999. goto group_next;
  2000. /*
  2001. * If the local group is idle or completely loaded
  2002. * no need to do power savings balance at this domain
  2003. */
  2004. if (local_group && (this_nr_running >= group_capacity ||
  2005. !this_nr_running))
  2006. power_savings_balance = 0;
  2007. /*
  2008. * If a group is already running at full capacity or idle,
  2009. * don't include that group in power savings calculations
  2010. */
  2011. if (!power_savings_balance || sum_nr_running >= group_capacity
  2012. || !sum_nr_running)
  2013. goto group_next;
  2014. /*
  2015. * Calculate the group which has the least non-idle load.
  2016. * This is the group from where we need to pick up the load
  2017. * for saving power
  2018. */
  2019. if ((sum_nr_running < min_nr_running) ||
  2020. (sum_nr_running == min_nr_running &&
  2021. first_cpu(group->cpumask) <
  2022. first_cpu(group_min->cpumask))) {
  2023. group_min = group;
  2024. min_nr_running = sum_nr_running;
  2025. min_load_per_task = sum_weighted_load /
  2026. sum_nr_running;
  2027. }
  2028. /*
  2029. * Calculate the group which is almost near its
  2030. * capacity but still has some space to pick up some load
  2031. * from other group and save more power
  2032. */
  2033. if (sum_nr_running <= group_capacity - 1) {
  2034. if (sum_nr_running > leader_nr_running ||
  2035. (sum_nr_running == leader_nr_running &&
  2036. first_cpu(group->cpumask) >
  2037. first_cpu(group_leader->cpumask))) {
  2038. group_leader = group;
  2039. leader_nr_running = sum_nr_running;
  2040. }
  2041. }
  2042. group_next:
  2043. #endif
  2044. group = group->next;
  2045. } while (group != sd->groups);
  2046. if (!busiest || this_load >= max_load || busiest_nr_running == 0)
  2047. goto out_balanced;
  2048. avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr;
  2049. if (this_load >= avg_load ||
  2050. 100*max_load <= sd->imbalance_pct*this_load)
  2051. goto out_balanced;
  2052. busiest_load_per_task /= busiest_nr_running;
  2053. /*
  2054. * We're trying to get all the cpus to the average_load, so we don't
  2055. * want to push ourselves above the average load, nor do we wish to
  2056. * reduce the max loaded cpu below the average load, as either of these
  2057. * actions would just result in more rebalancing later, and ping-pong
  2058. * tasks around. Thus we look for the minimum possible imbalance.
  2059. * Negative imbalances (*we* are more loaded than anyone else) will
  2060. * be counted as no imbalance for these purposes -- we can't fix that
  2061. * by pulling tasks to us. Be careful of negative numbers as they'll
  2062. * appear as very large values with unsigned longs.
  2063. */
  2064. if (max_load <= busiest_load_per_task)
  2065. goto out_balanced;
  2066. /*
  2067. * In the presence of smp nice balancing, certain scenarios can have
  2068. * max load less than avg load(as we skip the groups at or below
  2069. * its cpu_power, while calculating max_load..)
  2070. */
  2071. if (max_load < avg_load) {
  2072. *imbalance = 0;
  2073. goto small_imbalance;
  2074. }
  2075. /* Don't want to pull so many tasks that a group would go idle */
  2076. max_pull = min(max_load - avg_load, max_load - busiest_load_per_task);
  2077. /* How much load to actually move to equalise the imbalance */
  2078. *imbalance = min(max_pull * busiest->__cpu_power,
  2079. (avg_load - this_load) * this->__cpu_power)
  2080. / SCHED_LOAD_SCALE;
  2081. /*
  2082. * if *imbalance is less than the average load per runnable task
  2083. * there is no gaurantee that any tasks will be moved so we'll have
  2084. * a think about bumping its value to force at least one task to be
  2085. * moved
  2086. */
  2087. if (*imbalance + SCHED_LOAD_SCALE_FUZZ < busiest_load_per_task/2) {
  2088. unsigned long tmp, pwr_now, pwr_move;
  2089. unsigned int imbn;
  2090. small_imbalance:
  2091. pwr_move = pwr_now = 0;
  2092. imbn = 2;
  2093. if (this_nr_running) {
  2094. this_load_per_task /= this_nr_running;
  2095. if (busiest_load_per_task > this_load_per_task)
  2096. imbn = 1;
  2097. } else
  2098. this_load_per_task = SCHED_LOAD_SCALE;
  2099. if (max_load - this_load + SCHED_LOAD_SCALE_FUZZ >=
  2100. busiest_load_per_task * imbn) {
  2101. *imbalance = busiest_load_per_task;
  2102. return busiest;
  2103. }
  2104. /*
  2105. * OK, we don't have enough imbalance to justify moving tasks,
  2106. * however we may be able to increase total CPU power used by
  2107. * moving them.
  2108. */
  2109. pwr_now += busiest->__cpu_power *
  2110. min(busiest_load_per_task, max_load);
  2111. pwr_now += this->__cpu_power *
  2112. min(this_load_per_task, this_load);
  2113. pwr_now /= SCHED_LOAD_SCALE;
  2114. /* Amount of load we'd subtract */
  2115. tmp = sg_div_cpu_power(busiest,
  2116. busiest_load_per_task * SCHED_LOAD_SCALE);
  2117. if (max_load > tmp)
  2118. pwr_move += busiest->__cpu_power *
  2119. min(busiest_load_per_task, max_load - tmp);
  2120. /* Amount of load we'd add */
  2121. if (max_load * busiest->__cpu_power <
  2122. busiest_load_per_task * SCHED_LOAD_SCALE)
  2123. tmp = sg_div_cpu_power(this,
  2124. max_load * busiest->__cpu_power);
  2125. else
  2126. tmp = sg_div_cpu_power(this,
  2127. busiest_load_per_task * SCHED_LOAD_SCALE);
  2128. pwr_move += this->__cpu_power *
  2129. min(this_load_per_task, this_load + tmp);
  2130. pwr_move /= SCHED_LOAD_SCALE;
  2131. /* Move if we gain throughput */
  2132. if (pwr_move <= pwr_now)
  2133. goto out_balanced;
  2134. *imbalance = busiest_load_per_task;
  2135. }
  2136. return busiest;
  2137. out_balanced:
  2138. #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT)
  2139. if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE))
  2140. goto ret;
  2141. if (this == group_leader && group_leader != group_min) {
  2142. *imbalance = min_load_per_task;
  2143. return group_min;
  2144. }
  2145. #endif
  2146. ret:
  2147. *imbalance = 0;
  2148. return NULL;
  2149. }
  2150. /*
  2151. * find_busiest_queue - find the busiest runqueue among the cpus in group.
  2152. */
  2153. static struct rq *
  2154. find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle,
  2155. unsigned long imbalance, cpumask_t *cpus)
  2156. {
  2157. struct rq *busiest = NULL, *rq;
  2158. unsigned long max_load = 0;
  2159. int i;
  2160. for_each_cpu_mask(i, group->cpumask) {
  2161. unsigned long wl;
  2162. if (!cpu_isset(i, *cpus))
  2163. continue;
  2164. rq = cpu_rq(i);
  2165. wl = weighted_cpuload(i);
  2166. if (rq->nr_running == 1 && wl > imbalance)
  2167. continue;
  2168. if (wl > max_load) {
  2169. max_load = wl;
  2170. busiest = rq;
  2171. }
  2172. }
  2173. return busiest;
  2174. }
  2175. /*
  2176. * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but
  2177. * so long as it is large enough.
  2178. */
  2179. #define MAX_PINNED_INTERVAL 512
  2180. static inline unsigned long minus_1_or_zero(unsigned long n)
  2181. {
  2182. return n > 0 ? n - 1 : 0;
  2183. }
  2184. /*
  2185. * Check this_cpu to ensure it is balanced within domain. Attempt to move
  2186. * tasks if there is an imbalance.
  2187. */
  2188. static int load_balance(int this_cpu, struct rq *this_rq,
  2189. struct sched_domain *sd, enum cpu_idle_type idle,
  2190. int *balance)
  2191. {
  2192. int nr_moved, all_pinned = 0, active_balance = 0, sd_idle = 0;
  2193. struct sched_group *group;
  2194. unsigned long imbalance;
  2195. struct rq *busiest;
  2196. cpumask_t cpus = CPU_MASK_ALL;
  2197. unsigned long flags;
  2198. /*
  2199. * When power savings policy is enabled for the parent domain, idle
  2200. * sibling can pick up load irrespective of busy siblings. In this case,
  2201. * let the state of idle sibling percolate up as CPU_IDLE, instead of
  2202. * portraying it as CPU_NOT_IDLE.
  2203. */
  2204. if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER &&
  2205. !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE))
  2206. sd_idle = 1;
  2207. schedstat_inc(sd, lb_cnt[idle]);
  2208. redo:
  2209. group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle,
  2210. &cpus, balance);
  2211. if (*balance == 0)
  2212. goto out_balanced;
  2213. if (!group) {
  2214. schedstat_inc(sd, lb_nobusyg[idle]);
  2215. goto out_balanced;
  2216. }
  2217. busiest = find_busiest_queue(group, idle, imbalance, &cpus);
  2218. if (!busiest) {
  2219. schedstat_inc(sd, lb_nobusyq[idle]);
  2220. goto out_balanced;
  2221. }
  2222. BUG_ON(busiest == this_rq);
  2223. schedstat_add(sd, lb_imbalance[idle], imbalance);
  2224. nr_moved = 0;
  2225. if (busiest->nr_running > 1) {
  2226. /*
  2227. * Attempt to move tasks. If find_busiest_group has found
  2228. * an imbalance but busiest->nr_running <= 1, the group is
  2229. * still unbalanced. nr_moved simply stays zero, so it is
  2230. * correctly treated as an imbalance.
  2231. */
  2232. local_irq_save(flags);
  2233. double_rq_lock(this_rq, busiest);
  2234. nr_moved = move_tasks(this_rq, this_cpu, busiest,
  2235. minus_1_or_zero(busiest->nr_running),
  2236. imbalance, sd, idle, &all_pinned);
  2237. double_rq_unlock(this_rq, busiest);
  2238. local_irq_restore(flags);
  2239. /*
  2240. * some other cpu did the load balance for us.
  2241. */
  2242. if (nr_moved && this_cpu != smp_processor_id())
  2243. resched_cpu(this_cpu);
  2244. /* All tasks on this runqueue were pinned by CPU affinity */
  2245. if (unlikely(all_pinned)) {
  2246. cpu_clear(cpu_of(busiest), cpus);
  2247. if (!cpus_empty(cpus))
  2248. goto redo;
  2249. goto out_balanced;
  2250. }
  2251. }
  2252. if (!nr_moved) {
  2253. schedstat_inc(sd, lb_failed[idle]);
  2254. sd->nr_balance_failed++;
  2255. if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) {
  2256. spin_lock_irqsave(&busiest->lock, flags);
  2257. /* don't kick the migration_thread, if the curr
  2258. * task on busiest cpu can't be moved to this_cpu
  2259. */
  2260. if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) {
  2261. spin_unlock_irqrestore(&busiest->lock, flags);
  2262. all_pinned = 1;
  2263. goto out_one_pinned;
  2264. }
  2265. if (!busiest->active_balance) {
  2266. busiest->active_balance = 1;
  2267. busiest->push_cpu = this_cpu;
  2268. active_balance = 1;
  2269. }
  2270. spin_unlock_irqrestore(&busiest->lock, flags);
  2271. if (active_balance)
  2272. wake_up_process(busiest->migration_thread);
  2273. /*
  2274. * We've kicked active balancing, reset the failure
  2275. * counter.
  2276. */
  2277. sd->nr_balance_failed = sd->cache_nice_tries+1;
  2278. }
  2279. } else
  2280. sd->nr_balance_failed = 0;
  2281. if (likely(!active_balance)) {
  2282. /* We were unbalanced, so reset the balancing interval */
  2283. sd->balance_interval = sd->min_interval;
  2284. } else {
  2285. /*
  2286. * If we've begun active balancing, start to back off. This
  2287. * case may not be covered by the all_pinned logic if there
  2288. * is only 1 task on the busy runqueue (because we don't call
  2289. * move_tasks).
  2290. */
  2291. if (sd->balance_interval < sd->max_interval)
  2292. sd->balance_interval *= 2;
  2293. }
  2294. if (!nr_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER &&
  2295. !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE))
  2296. return -1;
  2297. return nr_moved;
  2298. out_balanced:
  2299. schedstat_inc(sd, lb_balanced[idle]);
  2300. sd->nr_balance_failed = 0;
  2301. out_one_pinned:
  2302. /* tune up the balancing interval */
  2303. if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) ||
  2304. (sd->balance_interval < sd->max_interval))
  2305. sd->balance_interval *= 2;
  2306. if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER &&
  2307. !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE))
  2308. return -1;
  2309. return 0;
  2310. }
  2311. /*
  2312. * Check this_cpu to ensure it is balanced within domain. Attempt to move
  2313. * tasks if there is an imbalance.
  2314. *
  2315. * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE).
  2316. * this_rq is locked.
  2317. */
  2318. static int
  2319. load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd)
  2320. {
  2321. struct sched_group *group;
  2322. struct rq *busiest = NULL;
  2323. unsigned long imbalance;
  2324. int nr_moved = 0;
  2325. int sd_idle = 0;
  2326. int all_pinned = 0;
  2327. cpumask_t cpus = CPU_MASK_ALL;
  2328. /*
  2329. * When power savings policy is enabled for the parent domain, idle
  2330. * sibling can pick up load irrespective of busy siblings. In this case,
  2331. * let the state of idle sibling percolate up as IDLE, instead of
  2332. * portraying it as CPU_NOT_IDLE.
  2333. */
  2334. if (sd->flags & SD_SHARE_CPUPOWER &&
  2335. !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE))
  2336. sd_idle = 1;
  2337. schedstat_inc(sd, lb_cnt[CPU_NEWLY_IDLE]);
  2338. redo:
  2339. group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE,
  2340. &sd_idle, &cpus, NULL);
  2341. if (!group) {
  2342. schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]);
  2343. goto out_balanced;
  2344. }
  2345. busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance,
  2346. &cpus);
  2347. if (!busiest) {
  2348. schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]);
  2349. goto out_balanced;
  2350. }
  2351. BUG_ON(busiest == this_rq);
  2352. schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance);
  2353. nr_moved = 0;
  2354. if (busiest->nr_running > 1) {
  2355. /* Attempt to move tasks */
  2356. double_lock_balance(this_rq, busiest);
  2357. nr_moved = move_tasks(this_rq, this_cpu, busiest,
  2358. minus_1_or_zero(busiest->nr_running),
  2359. imbalance, sd, CPU_NEWLY_IDLE,
  2360. &all_pinned);
  2361. spin_unlock(&busiest->lock);
  2362. if (unlikely(all_pinned)) {
  2363. cpu_clear(cpu_of(busiest), cpus);
  2364. if (!cpus_empty(cpus))
  2365. goto redo;
  2366. }
  2367. }
  2368. if (!nr_moved) {
  2369. schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]);
  2370. if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER &&
  2371. !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE))
  2372. return -1;
  2373. } else
  2374. sd->nr_balance_failed = 0;
  2375. return nr_moved;
  2376. out_balanced:
  2377. schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]);
  2378. if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER &&
  2379. !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE))
  2380. return -1;
  2381. sd->nr_balance_failed = 0;
  2382. return 0;
  2383. }
  2384. /*
  2385. * idle_balance is called by schedule() if this_cpu is about to become
  2386. * idle. Attempts to pull tasks from other CPUs.
  2387. */
  2388. static void idle_balance(int this_cpu, struct rq *this_rq)
  2389. {
  2390. struct sched_domain *sd;
  2391. int pulled_task = -1;
  2392. unsigned long next_balance = jiffies + HZ;
  2393. for_each_domain(this_cpu, sd) {
  2394. unsigned long interval;
  2395. if (!(sd->flags & SD_LOAD_BALANCE))
  2396. continue;
  2397. if (sd->flags & SD_BALANCE_NEWIDLE)
  2398. /* If we've pulled tasks over stop searching: */
  2399. pulled_task = load_balance_newidle(this_cpu,
  2400. this_rq, sd);
  2401. interval = msecs_to_jiffies(sd->balance_interval);
  2402. if (time_after(next_balance, sd->last_balance + interval))
  2403. next_balance = sd->last_balance + interval;
  2404. if (pulled_task)
  2405. break;
  2406. }
  2407. if (pulled_task || time_after(jiffies, this_rq->next_balance)) {
  2408. /*
  2409. * We are going idle. next_balance may be set based on
  2410. * a busy processor. So reset next_balance.
  2411. */
  2412. this_rq->next_balance = next_balance;
  2413. }
  2414. }
  2415. /*
  2416. * active_load_balance is run by migration threads. It pushes running tasks
  2417. * off the busiest CPU onto idle CPUs. It requires at least 1 task to be
  2418. * running on each physical CPU where possible, and avoids physical /
  2419. * logical imbalances.
  2420. *
  2421. * Called with busiest_rq locked.
  2422. */
  2423. static void active_load_balance(struct rq *busiest_rq, int busiest_cpu)
  2424. {
  2425. int target_cpu = busiest_rq->push_cpu;
  2426. struct sched_domain *sd;
  2427. struct rq *target_rq;
  2428. /* Is there any task to move? */
  2429. if (busiest_rq->nr_running <= 1)
  2430. return;
  2431. target_rq = cpu_rq(target_cpu);
  2432. /*
  2433. * This condition is "impossible", if it occurs
  2434. * we need to fix it. Originally reported by
  2435. * Bjorn Helgaas on a 128-cpu setup.
  2436. */
  2437. BUG_ON(busiest_rq == target_rq);
  2438. /* move a task from busiest_rq to target_rq */
  2439. double_lock_balance(busiest_rq, target_rq);
  2440. /* Search for an sd spanning us and the target CPU. */
  2441. for_each_domain(target_cpu, sd) {
  2442. if ((sd->flags & SD_LOAD_BALANCE) &&
  2443. cpu_isset(busiest_cpu, sd->span))
  2444. break;
  2445. }
  2446. if (likely(sd)) {
  2447. schedstat_inc(sd, alb_cnt);
  2448. if (move_tasks(target_rq, target_cpu, busiest_rq, 1,
  2449. RTPRIO_TO_LOAD_WEIGHT(100), sd, CPU_IDLE,
  2450. NULL))
  2451. schedstat_inc(sd, alb_pushed);
  2452. else
  2453. schedstat_inc(sd, alb_failed);
  2454. }
  2455. spin_unlock(&target_rq->lock);
  2456. }
  2457. #ifdef CONFIG_NO_HZ
  2458. static struct {
  2459. atomic_t load_balancer;
  2460. cpumask_t cpu_mask;
  2461. } nohz ____cacheline_aligned = {
  2462. .load_balancer = ATOMIC_INIT(-1),
  2463. .cpu_mask = CPU_MASK_NONE,
  2464. };
  2465. /*
  2466. * This routine will try to nominate the ilb (idle load balancing)
  2467. * owner among the cpus whose ticks are stopped. ilb owner will do the idle
  2468. * load balancing on behalf of all those cpus. If all the cpus in the system
  2469. * go into this tickless mode, then there will be no ilb owner (as there is
  2470. * no need for one) and all the cpus will sleep till the next wakeup event
  2471. * arrives...
  2472. *
  2473. * For the ilb owner, tick is not stopped. And this tick will be used
  2474. * for idle load balancing. ilb owner will still be part of
  2475. * nohz.cpu_mask..
  2476. *
  2477. * While stopping the tick, this cpu will become the ilb owner if there
  2478. * is no other owner. And will be the owner till that cpu becomes busy
  2479. * or if all cpus in the system stop their ticks at which point
  2480. * there is no need for ilb owner.
  2481. *
  2482. * When the ilb owner becomes busy, it nominates another owner, during the
  2483. * next busy scheduler_tick()
  2484. */
  2485. int select_nohz_load_balancer(int stop_tick)
  2486. {
  2487. int cpu = smp_processor_id();
  2488. if (stop_tick) {
  2489. cpu_set(cpu, nohz.cpu_mask);
  2490. cpu_rq(cpu)->in_nohz_recently = 1;
  2491. /*
  2492. * If we are going offline and still the leader, give up!
  2493. */
  2494. if (cpu_is_offline(cpu) &&
  2495. atomic_read(&nohz.load_balancer) == cpu) {
  2496. if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu)
  2497. BUG();
  2498. return 0;
  2499. }
  2500. /* time for ilb owner also to sleep */
  2501. if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) {
  2502. if (atomic_read(&nohz.load_balancer) == cpu)
  2503. atomic_set(&nohz.load_balancer, -1);
  2504. return 0;
  2505. }
  2506. if (atomic_read(&nohz.load_balancer) == -1) {
  2507. /* make me the ilb owner */
  2508. if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1)
  2509. return 1;
  2510. } else if (atomic_read(&nohz.load_balancer) == cpu)
  2511. return 1;
  2512. } else {
  2513. if (!cpu_isset(cpu, nohz.cpu_mask))
  2514. return 0;
  2515. cpu_clear(cpu, nohz.cpu_mask);
  2516. if (atomic_read(&nohz.load_balancer) == cpu)
  2517. if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu)
  2518. BUG();
  2519. }
  2520. return 0;
  2521. }
  2522. #endif
  2523. static DEFINE_SPINLOCK(balancing);
  2524. /*
  2525. * It checks each scheduling domain to see if it is due to be balanced,
  2526. * and initiates a balancing operation if so.
  2527. *
  2528. * Balancing parameters are set up in arch_init_sched_domains.
  2529. */
  2530. static inline void rebalance_domains(int cpu, enum cpu_idle_type idle)
  2531. {
  2532. int balance = 1;
  2533. struct rq *rq = cpu_rq(cpu);
  2534. unsigned long interval;
  2535. struct sched_domain *sd;
  2536. /* Earliest time when we have to do rebalance again */
  2537. unsigned long next_balance = jiffies + 60*HZ;
  2538. for_each_domain(cpu, sd) {
  2539. if (!(sd->flags & SD_LOAD_BALANCE))
  2540. continue;
  2541. interval = sd->balance_interval;
  2542. if (idle != CPU_IDLE)
  2543. interval *= sd->busy_factor;
  2544. /* scale ms to jiffies */
  2545. interval = msecs_to_jiffies(interval);
  2546. if (unlikely(!interval))
  2547. interval = 1;
  2548. if (interval > HZ*NR_CPUS/10)
  2549. interval = HZ*NR_CPUS/10;
  2550. if (sd->flags & SD_SERIALIZE) {
  2551. if (!spin_trylock(&balancing))
  2552. goto out;
  2553. }
  2554. if (time_after_eq(jiffies, sd->last_balance + interval)) {
  2555. if (load_balance(cpu, rq, sd, idle, &balance)) {
  2556. /*
  2557. * We've pulled tasks over so either we're no
  2558. * longer idle, or one of our SMT siblings is
  2559. * not idle.
  2560. */
  2561. idle = CPU_NOT_IDLE;
  2562. }
  2563. sd->last_balance = jiffies;
  2564. }
  2565. if (sd->flags & SD_SERIALIZE)
  2566. spin_unlock(&balancing);
  2567. out:
  2568. if (time_after(next_balance, sd->last_balance + interval))
  2569. next_balance = sd->last_balance + interval;
  2570. /*
  2571. * Stop the load balance at this level. There is another
  2572. * CPU in our sched group which is doing load balancing more
  2573. * actively.
  2574. */
  2575. if (!balance)
  2576. break;
  2577. }
  2578. rq->next_balance = next_balance;
  2579. }
  2580. /*
  2581. * run_rebalance_domains is triggered when needed from the scheduler tick.
  2582. * In CONFIG_NO_HZ case, the idle load balance owner will do the
  2583. * rebalancing for all the cpus for whom scheduler ticks are stopped.
  2584. */
  2585. static void run_rebalance_domains(struct softirq_action *h)
  2586. {
  2587. int this_cpu = smp_processor_id();
  2588. struct rq *this_rq = cpu_rq(this_cpu);
  2589. enum cpu_idle_type idle = this_rq->idle_at_tick ?
  2590. CPU_IDLE : CPU_NOT_IDLE;
  2591. rebalance_domains(this_cpu, idle);
  2592. #ifdef CONFIG_NO_HZ
  2593. /*
  2594. * If this cpu is the owner for idle load balancing, then do the
  2595. * balancing on behalf of the other idle cpus whose ticks are
  2596. * stopped.
  2597. */
  2598. if (this_rq->idle_at_tick &&
  2599. atomic_read(&nohz.load_balancer) == this_cpu) {
  2600. cpumask_t cpus = nohz.cpu_mask;
  2601. struct rq *rq;
  2602. int balance_cpu;
  2603. cpu_clear(this_cpu, cpus);
  2604. for_each_cpu_mask(balance_cpu, cpus) {
  2605. /*
  2606. * If this cpu gets work to do, stop the load balancing
  2607. * work being done for other cpus. Next load
  2608. * balancing owner will pick it up.
  2609. */
  2610. if (need_resched())
  2611. break;
  2612. rebalance_domains(balance_cpu, SCHED_IDLE);
  2613. rq = cpu_rq(balance_cpu);
  2614. if (time_after(this_rq->next_balance, rq->next_balance))
  2615. this_rq->next_balance = rq->next_balance;
  2616. }
  2617. }
  2618. #endif
  2619. }
  2620. /*
  2621. * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing.
  2622. *
  2623. * In case of CONFIG_NO_HZ, this is the place where we nominate a new
  2624. * idle load balancing owner or decide to stop the periodic load balancing,
  2625. * if the whole system is idle.
  2626. */
  2627. static inline void trigger_load_balance(struct rq *rq, int cpu)
  2628. {
  2629. #ifdef CONFIG_NO_HZ
  2630. /*
  2631. * If we were in the nohz mode recently and busy at the current
  2632. * scheduler tick, then check if we need to nominate new idle
  2633. * load balancer.
  2634. */
  2635. if (rq->in_nohz_recently && !rq->idle_at_tick) {
  2636. rq->in_nohz_recently = 0;
  2637. if (atomic_read(&nohz.load_balancer) == cpu) {
  2638. cpu_clear(cpu, nohz.cpu_mask);
  2639. atomic_set(&nohz.load_balancer, -1);
  2640. }
  2641. if (atomic_read(&nohz.load_balancer) == -1) {
  2642. /*
  2643. * simple selection for now: Nominate the
  2644. * first cpu in the nohz list to be the next
  2645. * ilb owner.
  2646. *
  2647. * TBD: Traverse the sched domains and nominate
  2648. * the nearest cpu in the nohz.cpu_mask.
  2649. */
  2650. int ilb = first_cpu(nohz.cpu_mask);
  2651. if (ilb != NR_CPUS)
  2652. resched_cpu(ilb);
  2653. }
  2654. }
  2655. /*
  2656. * If this cpu is idle and doing idle load balancing for all the
  2657. * cpus with ticks stopped, is it time for that to stop?
  2658. */
  2659. if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu &&
  2660. cpus_weight(nohz.cpu_mask) == num_online_cpus()) {
  2661. resched_cpu(cpu);
  2662. return;
  2663. }
  2664. /*
  2665. * If this cpu is idle and the idle load balancing is done by
  2666. * someone else, then no need raise the SCHED_SOFTIRQ
  2667. */
  2668. if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu &&
  2669. cpu_isset(cpu, nohz.cpu_mask))
  2670. return;
  2671. #endif
  2672. if (time_after_eq(jiffies, rq->next_balance))
  2673. raise_softirq(SCHED_SOFTIRQ);
  2674. }
  2675. #else /* CONFIG_SMP */
  2676. /*
  2677. * on UP we do not need to balance between CPUs:
  2678. */
  2679. static inline void idle_balance(int cpu, struct rq *rq)
  2680. {
  2681. }
  2682. /* Avoid "used but not defined" warning on UP */
  2683. static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest,
  2684. unsigned long max_nr_move, unsigned long max_load_move,
  2685. struct sched_domain *sd, enum cpu_idle_type idle,
  2686. int *all_pinned, unsigned long *load_moved,
  2687. int this_best_prio, int best_prio, int best_prio_seen,
  2688. struct rq_iterator *iterator)
  2689. {
  2690. *load_moved = 0;
  2691. return 0;
  2692. }
  2693. #endif
  2694. DEFINE_PER_CPU(struct kernel_stat, kstat);
  2695. EXPORT_PER_CPU_SYMBOL(kstat);
  2696. /*
  2697. * Return p->sum_exec_runtime plus any more ns on the sched_clock
  2698. * that have not yet been banked in case the task is currently running.
  2699. */
  2700. unsigned long long task_sched_runtime(struct task_struct *p)
  2701. {
  2702. unsigned long flags;
  2703. u64 ns, delta_exec;
  2704. struct rq *rq;
  2705. rq = task_rq_lock(p, &flags);
  2706. ns = p->se.sum_exec_runtime;
  2707. if (rq->curr == p) {
  2708. delta_exec = rq_clock(rq) - p->se.exec_start;
  2709. if ((s64)delta_exec > 0)
  2710. ns += delta_exec;
  2711. }
  2712. task_rq_unlock(rq, &flags);
  2713. return ns;
  2714. }
  2715. /*
  2716. * Account user cpu time to a process.
  2717. * @p: the process that the cpu time gets accounted to
  2718. * @hardirq_offset: the offset to subtract from hardirq_count()
  2719. * @cputime: the cpu time spent in user space since the last update
  2720. */
  2721. void account_user_time(struct task_struct *p, cputime_t cputime)
  2722. {
  2723. struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
  2724. cputime64_t tmp;
  2725. p->utime = cputime_add(p->utime, cputime);
  2726. /* Add user time to cpustat. */
  2727. tmp = cputime_to_cputime64(cputime);
  2728. if (TASK_NICE(p) > 0)
  2729. cpustat->nice = cputime64_add(cpustat->nice, tmp);
  2730. else
  2731. cpustat->user = cputime64_add(cpustat->user, tmp);
  2732. }
  2733. /*
  2734. * Account system cpu time to a process.
  2735. * @p: the process that the cpu time gets accounted to
  2736. * @hardirq_offset: the offset to subtract from hardirq_count()
  2737. * @cputime: the cpu time spent in kernel space since the last update
  2738. */
  2739. void account_system_time(struct task_struct *p, int hardirq_offset,
  2740. cputime_t cputime)
  2741. {
  2742. struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
  2743. struct rq *rq = this_rq();
  2744. cputime64_t tmp;
  2745. p->stime = cputime_add(p->stime, cputime);
  2746. /* Add system time to cpustat. */
  2747. tmp = cputime_to_cputime64(cputime);
  2748. if (hardirq_count() - hardirq_offset)
  2749. cpustat->irq = cputime64_add(cpustat->irq, tmp);
  2750. else if (softirq_count())
  2751. cpustat->softirq = cputime64_add(cpustat->softirq, tmp);
  2752. else if (p != rq->idle)
  2753. cpustat->system = cputime64_add(cpustat->system, tmp);
  2754. else if (atomic_read(&rq->nr_iowait) > 0)
  2755. cpustat->iowait = cputime64_add(cpustat->iowait, tmp);
  2756. else
  2757. cpustat->idle = cputime64_add(cpustat->idle, tmp);
  2758. /* Account for system time used */
  2759. acct_update_integrals(p);
  2760. }
  2761. /*
  2762. * Account for involuntary wait time.
  2763. * @p: the process from which the cpu time has been stolen
  2764. * @steal: the cpu time spent in involuntary wait
  2765. */
  2766. void account_steal_time(struct task_struct *p, cputime_t steal)
  2767. {
  2768. struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
  2769. cputime64_t tmp = cputime_to_cputime64(steal);
  2770. struct rq *rq = this_rq();
  2771. if (p == rq->idle) {
  2772. p->stime = cputime_add(p->stime, steal);
  2773. if (atomic_read(&rq->nr_iowait) > 0)
  2774. cpustat->iowait = cputime64_add(cpustat->iowait, tmp);
  2775. else
  2776. cpustat->idle = cputime64_add(cpustat->idle, tmp);
  2777. } else
  2778. cpustat->steal = cputime64_add(cpustat->steal, tmp);
  2779. }
  2780. /*
  2781. * This function gets called by the timer code, with HZ frequency.
  2782. * We call it with interrupts disabled.
  2783. *
  2784. * It also gets called by the fork code, when changing the parent's
  2785. * timeslices.
  2786. */
  2787. void scheduler_tick(void)
  2788. {
  2789. int cpu = smp_processor_id();
  2790. struct rq *rq = cpu_rq(cpu);
  2791. struct task_struct *curr = rq->curr;
  2792. spin_lock(&rq->lock);
  2793. if (curr != rq->idle) /* FIXME: needed? */
  2794. curr->sched_class->task_tick(rq, curr);
  2795. update_cpu_load(rq);
  2796. spin_unlock(&rq->lock);
  2797. #ifdef CONFIG_SMP
  2798. rq->idle_at_tick = idle_cpu(cpu);
  2799. trigger_load_balance(rq, cpu);
  2800. #endif
  2801. }
  2802. #if defined(CONFIG_PREEMPT) && defined(CONFIG_DEBUG_PREEMPT)
  2803. void fastcall add_preempt_count(int val)
  2804. {
  2805. /*
  2806. * Underflow?
  2807. */
  2808. if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0)))
  2809. return;
  2810. preempt_count() += val;
  2811. /*
  2812. * Spinlock count overflowing soon?
  2813. */
  2814. DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >=
  2815. PREEMPT_MASK - 10);
  2816. }
  2817. EXPORT_SYMBOL(add_preempt_count);
  2818. void fastcall sub_preempt_count(int val)
  2819. {
  2820. /*
  2821. * Underflow?
  2822. */
  2823. if (DEBUG_LOCKS_WARN_ON(val > preempt_count()))
  2824. return;
  2825. /*
  2826. * Is the spinlock portion underflowing?
  2827. */
  2828. if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) &&
  2829. !(preempt_count() & PREEMPT_MASK)))
  2830. return;
  2831. preempt_count() -= val;
  2832. }
  2833. EXPORT_SYMBOL(sub_preempt_count);
  2834. #endif
  2835. /*
  2836. * Print scheduling while atomic bug:
  2837. */
  2838. static noinline void __schedule_bug(struct task_struct *prev)
  2839. {
  2840. printk(KERN_ERR "BUG: scheduling while atomic: %s/0x%08x/%d\n",
  2841. prev->comm, preempt_count(), prev->pid);
  2842. debug_show_held_locks(prev);
  2843. if (irqs_disabled())
  2844. print_irqtrace_events(prev);
  2845. dump_stack();
  2846. }
  2847. /*
  2848. * Various schedule()-time debugging checks and statistics:
  2849. */
  2850. static inline void schedule_debug(struct task_struct *prev)
  2851. {
  2852. /*
  2853. * Test if we are atomic. Since do_exit() needs to call into
  2854. * schedule() atomically, we ignore that path for now.
  2855. * Otherwise, whine if we are scheduling when we should not be.
  2856. */
  2857. if (unlikely(in_atomic_preempt_off()) && unlikely(!prev->exit_state))
  2858. __schedule_bug(prev);
  2859. profile_hit(SCHED_PROFILING, __builtin_return_address(0));
  2860. schedstat_inc(this_rq(), sched_cnt);
  2861. }
  2862. /*
  2863. * Pick up the highest-prio task:
  2864. */
  2865. static inline struct task_struct *
  2866. pick_next_task(struct rq *rq, struct task_struct *prev, u64 now)
  2867. {
  2868. struct sched_class *class;
  2869. struct task_struct *p;
  2870. /*
  2871. * Optimization: we know that if all tasks are in
  2872. * the fair class we can call that function directly:
  2873. */
  2874. if (likely(rq->nr_running == rq->cfs.nr_running)) {
  2875. p = fair_sched_class.pick_next_task(rq, now);
  2876. if (likely(p))
  2877. return p;
  2878. }
  2879. class = sched_class_highest;
  2880. for ( ; ; ) {
  2881. p = class->pick_next_task(rq, now);
  2882. if (p)
  2883. return p;
  2884. /*
  2885. * Will never be NULL as the idle class always
  2886. * returns a non-NULL p:
  2887. */
  2888. class = class->next;
  2889. }
  2890. }
  2891. /*
  2892. * schedule() is the main scheduler function.
  2893. */
  2894. asmlinkage void __sched schedule(void)
  2895. {
  2896. struct task_struct *prev, *next;
  2897. long *switch_count;
  2898. struct rq *rq;
  2899. u64 now;
  2900. int cpu;
  2901. need_resched:
  2902. preempt_disable();
  2903. cpu = smp_processor_id();
  2904. rq = cpu_rq(cpu);
  2905. rcu_qsctr_inc(cpu);
  2906. prev = rq->curr;
  2907. switch_count = &prev->nivcsw;
  2908. release_kernel_lock(prev);
  2909. need_resched_nonpreemptible:
  2910. schedule_debug(prev);
  2911. spin_lock_irq(&rq->lock);
  2912. clear_tsk_need_resched(prev);
  2913. if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) {
  2914. if (unlikely((prev->state & TASK_INTERRUPTIBLE) &&
  2915. unlikely(signal_pending(prev)))) {
  2916. prev->state = TASK_RUNNING;
  2917. } else {
  2918. deactivate_task(rq, prev, 1);
  2919. }
  2920. switch_count = &prev->nvcsw;
  2921. }
  2922. if (unlikely(!rq->nr_running))
  2923. idle_balance(cpu, rq);
  2924. now = __rq_clock(rq);
  2925. prev->sched_class->put_prev_task(rq, prev, now);
  2926. next = pick_next_task(rq, prev, now);
  2927. sched_info_switch(prev, next);
  2928. if (likely(prev != next)) {
  2929. rq->nr_switches++;
  2930. rq->curr = next;
  2931. ++*switch_count;
  2932. context_switch(rq, prev, next); /* unlocks the rq */
  2933. } else
  2934. spin_unlock_irq(&rq->lock);
  2935. if (unlikely(reacquire_kernel_lock(current) < 0)) {
  2936. cpu = smp_processor_id();
  2937. rq = cpu_rq(cpu);
  2938. goto need_resched_nonpreemptible;
  2939. }
  2940. preempt_enable_no_resched();
  2941. if (unlikely(test_thread_flag(TIF_NEED_RESCHED)))
  2942. goto need_resched;
  2943. }
  2944. EXPORT_SYMBOL(schedule);
  2945. #ifdef CONFIG_PREEMPT
  2946. /*
  2947. * this is the entry point to schedule() from in-kernel preemption
  2948. * off of preempt_enable. Kernel preemptions off return from interrupt
  2949. * occur there and call schedule directly.
  2950. */
  2951. asmlinkage void __sched preempt_schedule(void)
  2952. {
  2953. struct thread_info *ti = current_thread_info();
  2954. #ifdef CONFIG_PREEMPT_BKL
  2955. struct task_struct *task = current;
  2956. int saved_lock_depth;
  2957. #endif
  2958. /*
  2959. * If there is a non-zero preempt_count or interrupts are disabled,
  2960. * we do not want to preempt the current task. Just return..
  2961. */
  2962. if (likely(ti->preempt_count || irqs_disabled()))
  2963. return;
  2964. need_resched:
  2965. add_preempt_count(PREEMPT_ACTIVE);
  2966. /*
  2967. * We keep the big kernel semaphore locked, but we
  2968. * clear ->lock_depth so that schedule() doesnt
  2969. * auto-release the semaphore:
  2970. */
  2971. #ifdef CONFIG_PREEMPT_BKL
  2972. saved_lock_depth = task->lock_depth;
  2973. task->lock_depth = -1;
  2974. #endif
  2975. schedule();
  2976. #ifdef CONFIG_PREEMPT_BKL
  2977. task->lock_depth = saved_lock_depth;
  2978. #endif
  2979. sub_preempt_count(PREEMPT_ACTIVE);
  2980. /* we could miss a preemption opportunity between schedule and now */
  2981. barrier();
  2982. if (unlikely(test_thread_flag(TIF_NEED_RESCHED)))
  2983. goto need_resched;
  2984. }
  2985. EXPORT_SYMBOL(preempt_schedule);
  2986. /*
  2987. * this is the entry point to schedule() from kernel preemption
  2988. * off of irq context.
  2989. * Note, that this is called and return with irqs disabled. This will
  2990. * protect us against recursive calling from irq.
  2991. */
  2992. asmlinkage void __sched preempt_schedule_irq(void)
  2993. {
  2994. struct thread_info *ti = current_thread_info();
  2995. #ifdef CONFIG_PREEMPT_BKL
  2996. struct task_struct *task = current;
  2997. int saved_lock_depth;
  2998. #endif
  2999. /* Catch callers which need to be fixed */
  3000. BUG_ON(ti->preempt_count || !irqs_disabled());
  3001. need_resched:
  3002. add_preempt_count(PREEMPT_ACTIVE);
  3003. /*
  3004. * We keep the big kernel semaphore locked, but we
  3005. * clear ->lock_depth so that schedule() doesnt
  3006. * auto-release the semaphore:
  3007. */
  3008. #ifdef CONFIG_PREEMPT_BKL
  3009. saved_lock_depth = task->lock_depth;
  3010. task->lock_depth = -1;
  3011. #endif
  3012. local_irq_enable();
  3013. schedule();
  3014. local_irq_disable();
  3015. #ifdef CONFIG_PREEMPT_BKL
  3016. task->lock_depth = saved_lock_depth;
  3017. #endif
  3018. sub_preempt_count(PREEMPT_ACTIVE);
  3019. /* we could miss a preemption opportunity between schedule and now */
  3020. barrier();
  3021. if (unlikely(test_thread_flag(TIF_NEED_RESCHED)))
  3022. goto need_resched;
  3023. }
  3024. #endif /* CONFIG_PREEMPT */
  3025. int default_wake_function(wait_queue_t *curr, unsigned mode, int sync,
  3026. void *key)
  3027. {
  3028. return try_to_wake_up(curr->private, mode, sync);
  3029. }
  3030. EXPORT_SYMBOL(default_wake_function);
  3031. /*
  3032. * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just
  3033. * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve
  3034. * number) then we wake all the non-exclusive tasks and one exclusive task.
  3035. *
  3036. * There are circumstances in which we can try to wake a task which has already
  3037. * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns
  3038. * zero in this (rare) case, and we handle it by continuing to scan the queue.
  3039. */
  3040. static void __wake_up_common(wait_queue_head_t *q, unsigned int mode,
  3041. int nr_exclusive, int sync, void *key)
  3042. {
  3043. struct list_head *tmp, *next;
  3044. list_for_each_safe(tmp, next, &q->task_list) {
  3045. wait_queue_t *curr = list_entry(tmp, wait_queue_t, task_list);
  3046. unsigned flags = curr->flags;
  3047. if (curr->func(curr, mode, sync, key) &&
  3048. (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive)
  3049. break;
  3050. }
  3051. }
  3052. /**
  3053. * __wake_up - wake up threads blocked on a waitqueue.
  3054. * @q: the waitqueue
  3055. * @mode: which threads
  3056. * @nr_exclusive: how many wake-one or wake-many threads to wake up
  3057. * @key: is directly passed to the wakeup function
  3058. */
  3059. void fastcall __wake_up(wait_queue_head_t *q, unsigned int mode,
  3060. int nr_exclusive, void *key)
  3061. {
  3062. unsigned long flags;
  3063. spin_lock_irqsave(&q->lock, flags);
  3064. __wake_up_common(q, mode, nr_exclusive, 0, key);
  3065. spin_unlock_irqrestore(&q->lock, flags);
  3066. }
  3067. EXPORT_SYMBOL(__wake_up);
  3068. /*
  3069. * Same as __wake_up but called with the spinlock in wait_queue_head_t held.
  3070. */
  3071. void fastcall __wake_up_locked(wait_queue_head_t *q, unsigned int mode)
  3072. {
  3073. __wake_up_common(q, mode, 1, 0, NULL);
  3074. }
  3075. /**
  3076. * __wake_up_sync - wake up threads blocked on a waitqueue.
  3077. * @q: the waitqueue
  3078. * @mode: which threads
  3079. * @nr_exclusive: how many wake-one or wake-many threads to wake up
  3080. *
  3081. * The sync wakeup differs that the waker knows that it will schedule
  3082. * away soon, so while the target thread will be woken up, it will not
  3083. * be migrated to another CPU - ie. the two threads are 'synchronized'
  3084. * with each other. This can prevent needless bouncing between CPUs.
  3085. *
  3086. * On UP it can prevent extra preemption.
  3087. */
  3088. void fastcall
  3089. __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive)
  3090. {
  3091. unsigned long flags;
  3092. int sync = 1;
  3093. if (unlikely(!q))
  3094. return;
  3095. if (unlikely(!nr_exclusive))
  3096. sync = 0;
  3097. spin_lock_irqsave(&q->lock, flags);
  3098. __wake_up_common(q, mode, nr_exclusive, sync, NULL);
  3099. spin_unlock_irqrestore(&q->lock, flags);
  3100. }
  3101. EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */
  3102. void fastcall complete(struct completion *x)
  3103. {
  3104. unsigned long flags;
  3105. spin_lock_irqsave(&x->wait.lock, flags);
  3106. x->done++;
  3107. __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
  3108. 1, 0, NULL);
  3109. spin_unlock_irqrestore(&x->wait.lock, flags);
  3110. }
  3111. EXPORT_SYMBOL(complete);
  3112. void fastcall complete_all(struct completion *x)
  3113. {
  3114. unsigned long flags;
  3115. spin_lock_irqsave(&x->wait.lock, flags);
  3116. x->done += UINT_MAX/2;
  3117. __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
  3118. 0, 0, NULL);
  3119. spin_unlock_irqrestore(&x->wait.lock, flags);
  3120. }
  3121. EXPORT_SYMBOL(complete_all);
  3122. void fastcall __sched wait_for_completion(struct completion *x)
  3123. {
  3124. might_sleep();
  3125. spin_lock_irq(&x->wait.lock);
  3126. if (!x->done) {
  3127. DECLARE_WAITQUEUE(wait, current);
  3128. wait.flags |= WQ_FLAG_EXCLUSIVE;
  3129. __add_wait_queue_tail(&x->wait, &wait);
  3130. do {
  3131. __set_current_state(TASK_UNINTERRUPTIBLE);
  3132. spin_unlock_irq(&x->wait.lock);
  3133. schedule();
  3134. spin_lock_irq(&x->wait.lock);
  3135. } while (!x->done);
  3136. __remove_wait_queue(&x->wait, &wait);
  3137. }
  3138. x->done--;
  3139. spin_unlock_irq(&x->wait.lock);
  3140. }
  3141. EXPORT_SYMBOL(wait_for_completion);
  3142. unsigned long fastcall __sched
  3143. wait_for_completion_timeout(struct completion *x, unsigned long timeout)
  3144. {
  3145. might_sleep();
  3146. spin_lock_irq(&x->wait.lock);
  3147. if (!x->done) {
  3148. DECLARE_WAITQUEUE(wait, current);
  3149. wait.flags |= WQ_FLAG_EXCLUSIVE;
  3150. __add_wait_queue_tail(&x->wait, &wait);
  3151. do {
  3152. __set_current_state(TASK_UNINTERRUPTIBLE);
  3153. spin_unlock_irq(&x->wait.lock);
  3154. timeout = schedule_timeout(timeout);
  3155. spin_lock_irq(&x->wait.lock);
  3156. if (!timeout) {
  3157. __remove_wait_queue(&x->wait, &wait);
  3158. goto out;
  3159. }
  3160. } while (!x->done);
  3161. __remove_wait_queue(&x->wait, &wait);
  3162. }
  3163. x->done--;
  3164. out:
  3165. spin_unlock_irq(&x->wait.lock);
  3166. return timeout;
  3167. }
  3168. EXPORT_SYMBOL(wait_for_completion_timeout);
  3169. int fastcall __sched wait_for_completion_interruptible(struct completion *x)
  3170. {
  3171. int ret = 0;
  3172. might_sleep();
  3173. spin_lock_irq(&x->wait.lock);
  3174. if (!x->done) {
  3175. DECLARE_WAITQUEUE(wait, current);
  3176. wait.flags |= WQ_FLAG_EXCLUSIVE;
  3177. __add_wait_queue_tail(&x->wait, &wait);
  3178. do {
  3179. if (signal_pending(current)) {
  3180. ret = -ERESTARTSYS;
  3181. __remove_wait_queue(&x->wait, &wait);
  3182. goto out;
  3183. }
  3184. __set_current_state(TASK_INTERRUPTIBLE);
  3185. spin_unlock_irq(&x->wait.lock);
  3186. schedule();
  3187. spin_lock_irq(&x->wait.lock);
  3188. } while (!x->done);
  3189. __remove_wait_queue(&x->wait, &wait);
  3190. }
  3191. x->done--;
  3192. out:
  3193. spin_unlock_irq(&x->wait.lock);
  3194. return ret;
  3195. }
  3196. EXPORT_SYMBOL(wait_for_completion_interruptible);
  3197. unsigned long fastcall __sched
  3198. wait_for_completion_interruptible_timeout(struct completion *x,
  3199. unsigned long timeout)
  3200. {
  3201. might_sleep();
  3202. spin_lock_irq(&x->wait.lock);
  3203. if (!x->done) {
  3204. DECLARE_WAITQUEUE(wait, current);
  3205. wait.flags |= WQ_FLAG_EXCLUSIVE;
  3206. __add_wait_queue_tail(&x->wait, &wait);
  3207. do {
  3208. if (signal_pending(current)) {
  3209. timeout = -ERESTARTSYS;
  3210. __remove_wait_queue(&x->wait, &wait);
  3211. goto out;
  3212. }
  3213. __set_current_state(TASK_INTERRUPTIBLE);
  3214. spin_unlock_irq(&x->wait.lock);
  3215. timeout = schedule_timeout(timeout);
  3216. spin_lock_irq(&x->wait.lock);
  3217. if (!timeout) {
  3218. __remove_wait_queue(&x->wait, &wait);
  3219. goto out;
  3220. }
  3221. } while (!x->done);
  3222. __remove_wait_queue(&x->wait, &wait);
  3223. }
  3224. x->done--;
  3225. out:
  3226. spin_unlock_irq(&x->wait.lock);
  3227. return timeout;
  3228. }
  3229. EXPORT_SYMBOL(wait_for_completion_interruptible_timeout);
  3230. static inline void
  3231. sleep_on_head(wait_queue_head_t *q, wait_queue_t *wait, unsigned long *flags)
  3232. {
  3233. spin_lock_irqsave(&q->lock, *flags);
  3234. __add_wait_queue(q, wait);
  3235. spin_unlock(&q->lock);
  3236. }
  3237. static inline void
  3238. sleep_on_tail(wait_queue_head_t *q, wait_queue_t *wait, unsigned long *flags)
  3239. {
  3240. spin_lock_irq(&q->lock);
  3241. __remove_wait_queue(q, wait);
  3242. spin_unlock_irqrestore(&q->lock, *flags);
  3243. }
  3244. void __sched interruptible_sleep_on(wait_queue_head_t *q)
  3245. {
  3246. unsigned long flags;
  3247. wait_queue_t wait;
  3248. init_waitqueue_entry(&wait, current);
  3249. current->state = TASK_INTERRUPTIBLE;
  3250. sleep_on_head(q, &wait, &flags);
  3251. schedule();
  3252. sleep_on_tail(q, &wait, &flags);
  3253. }
  3254. EXPORT_SYMBOL(interruptible_sleep_on);
  3255. long __sched
  3256. interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout)
  3257. {
  3258. unsigned long flags;
  3259. wait_queue_t wait;
  3260. init_waitqueue_entry(&wait, current);
  3261. current->state = TASK_INTERRUPTIBLE;
  3262. sleep_on_head(q, &wait, &flags);
  3263. timeout = schedule_timeout(timeout);
  3264. sleep_on_tail(q, &wait, &flags);
  3265. return timeout;
  3266. }
  3267. EXPORT_SYMBOL(interruptible_sleep_on_timeout);
  3268. void __sched sleep_on(wait_queue_head_t *q)
  3269. {
  3270. unsigned long flags;
  3271. wait_queue_t wait;
  3272. init_waitqueue_entry(&wait, current);
  3273. current->state = TASK_UNINTERRUPTIBLE;
  3274. sleep_on_head(q, &wait, &flags);
  3275. schedule();
  3276. sleep_on_tail(q, &wait, &flags);
  3277. }
  3278. EXPORT_SYMBOL(sleep_on);
  3279. long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout)
  3280. {
  3281. unsigned long flags;
  3282. wait_queue_t wait;
  3283. init_waitqueue_entry(&wait, current);
  3284. current->state = TASK_UNINTERRUPTIBLE;
  3285. sleep_on_head(q, &wait, &flags);
  3286. timeout = schedule_timeout(timeout);
  3287. sleep_on_tail(q, &wait, &flags);
  3288. return timeout;
  3289. }
  3290. EXPORT_SYMBOL(sleep_on_timeout);
  3291. #ifdef CONFIG_RT_MUTEXES
  3292. /*
  3293. * rt_mutex_setprio - set the current priority of a task
  3294. * @p: task
  3295. * @prio: prio value (kernel-internal form)
  3296. *
  3297. * This function changes the 'effective' priority of a task. It does
  3298. * not touch ->normal_prio like __setscheduler().
  3299. *
  3300. * Used by the rt_mutex code to implement priority inheritance logic.
  3301. */
  3302. void rt_mutex_setprio(struct task_struct *p, int prio)
  3303. {
  3304. unsigned long flags;
  3305. int oldprio, on_rq;
  3306. struct rq *rq;
  3307. u64 now;
  3308. BUG_ON(prio < 0 || prio > MAX_PRIO);
  3309. rq = task_rq_lock(p, &flags);
  3310. now = rq_clock(rq);
  3311. oldprio = p->prio;
  3312. on_rq = p->se.on_rq;
  3313. if (on_rq)
  3314. dequeue_task(rq, p, 0, now);
  3315. if (rt_prio(prio))
  3316. p->sched_class = &rt_sched_class;
  3317. else
  3318. p->sched_class = &fair_sched_class;
  3319. p->prio = prio;
  3320. if (on_rq) {
  3321. enqueue_task(rq, p, 0, now);
  3322. /*
  3323. * Reschedule if we are currently running on this runqueue and
  3324. * our priority decreased, or if we are not currently running on
  3325. * this runqueue and our priority is higher than the current's
  3326. */
  3327. if (task_running(rq, p)) {
  3328. if (p->prio > oldprio)
  3329. resched_task(rq->curr);
  3330. } else {
  3331. check_preempt_curr(rq, p);
  3332. }
  3333. }
  3334. task_rq_unlock(rq, &flags);
  3335. }
  3336. #endif
  3337. void set_user_nice(struct task_struct *p, long nice)
  3338. {
  3339. int old_prio, delta, on_rq;
  3340. unsigned long flags;
  3341. struct rq *rq;
  3342. u64 now;
  3343. if (TASK_NICE(p) == nice || nice < -20 || nice > 19)
  3344. return;
  3345. /*
  3346. * We have to be careful, if called from sys_setpriority(),
  3347. * the task might be in the middle of scheduling on another CPU.
  3348. */
  3349. rq = task_rq_lock(p, &flags);
  3350. now = rq_clock(rq);
  3351. /*
  3352. * The RT priorities are set via sched_setscheduler(), but we still
  3353. * allow the 'normal' nice value to be set - but as expected
  3354. * it wont have any effect on scheduling until the task is
  3355. * SCHED_FIFO/SCHED_RR:
  3356. */
  3357. if (task_has_rt_policy(p)) {
  3358. p->static_prio = NICE_TO_PRIO(nice);
  3359. goto out_unlock;
  3360. }
  3361. on_rq = p->se.on_rq;
  3362. if (on_rq) {
  3363. dequeue_task(rq, p, 0, now);
  3364. dec_load(rq, p, now);
  3365. }
  3366. p->static_prio = NICE_TO_PRIO(nice);
  3367. set_load_weight(p);
  3368. old_prio = p->prio;
  3369. p->prio = effective_prio(p);
  3370. delta = p->prio - old_prio;
  3371. if (on_rq) {
  3372. enqueue_task(rq, p, 0, now);
  3373. inc_load(rq, p, now);
  3374. /*
  3375. * If the task increased its priority or is running and
  3376. * lowered its priority, then reschedule its CPU:
  3377. */
  3378. if (delta < 0 || (delta > 0 && task_running(rq, p)))
  3379. resched_task(rq->curr);
  3380. }
  3381. out_unlock:
  3382. task_rq_unlock(rq, &flags);
  3383. }
  3384. EXPORT_SYMBOL(set_user_nice);
  3385. /*
  3386. * can_nice - check if a task can reduce its nice value
  3387. * @p: task
  3388. * @nice: nice value
  3389. */
  3390. int can_nice(const struct task_struct *p, const int nice)
  3391. {
  3392. /* convert nice value [19,-20] to rlimit style value [1,40] */
  3393. int nice_rlim = 20 - nice;
  3394. return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur ||
  3395. capable(CAP_SYS_NICE));
  3396. }
  3397. #ifdef __ARCH_WANT_SYS_NICE
  3398. /*
  3399. * sys_nice - change the priority of the current process.
  3400. * @increment: priority increment
  3401. *
  3402. * sys_setpriority is a more generic, but much slower function that
  3403. * does similar things.
  3404. */
  3405. asmlinkage long sys_nice(int increment)
  3406. {
  3407. long nice, retval;
  3408. /*
  3409. * Setpriority might change our priority at the same moment.
  3410. * We don't have to worry. Conceptually one call occurs first
  3411. * and we have a single winner.
  3412. */
  3413. if (increment < -40)
  3414. increment = -40;
  3415. if (increment > 40)
  3416. increment = 40;
  3417. nice = PRIO_TO_NICE(current->static_prio) + increment;
  3418. if (nice < -20)
  3419. nice = -20;
  3420. if (nice > 19)
  3421. nice = 19;
  3422. if (increment < 0 && !can_nice(current, nice))
  3423. return -EPERM;
  3424. retval = security_task_setnice(current, nice);
  3425. if (retval)
  3426. return retval;
  3427. set_user_nice(current, nice);
  3428. return 0;
  3429. }
  3430. #endif
  3431. /**
  3432. * task_prio - return the priority value of a given task.
  3433. * @p: the task in question.
  3434. *
  3435. * This is the priority value as seen by users in /proc.
  3436. * RT tasks are offset by -200. Normal tasks are centered
  3437. * around 0, value goes from -16 to +15.
  3438. */
  3439. int task_prio(const struct task_struct *p)
  3440. {
  3441. return p->prio - MAX_RT_PRIO;
  3442. }
  3443. /**
  3444. * task_nice - return the nice value of a given task.
  3445. * @p: the task in question.
  3446. */
  3447. int task_nice(const struct task_struct *p)
  3448. {
  3449. return TASK_NICE(p);
  3450. }
  3451. EXPORT_SYMBOL_GPL(task_nice);
  3452. /**
  3453. * idle_cpu - is a given cpu idle currently?
  3454. * @cpu: the processor in question.
  3455. */
  3456. int idle_cpu(int cpu)
  3457. {
  3458. return cpu_curr(cpu) == cpu_rq(cpu)->idle;
  3459. }
  3460. /**
  3461. * idle_task - return the idle task for a given cpu.
  3462. * @cpu: the processor in question.
  3463. */
  3464. struct task_struct *idle_task(int cpu)
  3465. {
  3466. return cpu_rq(cpu)->idle;
  3467. }
  3468. /**
  3469. * find_process_by_pid - find a process with a matching PID value.
  3470. * @pid: the pid in question.
  3471. */
  3472. static inline struct task_struct *find_process_by_pid(pid_t pid)
  3473. {
  3474. return pid ? find_task_by_pid(pid) : current;
  3475. }
  3476. /* Actually do priority change: must hold rq lock. */
  3477. static void
  3478. __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio)
  3479. {
  3480. BUG_ON(p->se.on_rq);
  3481. p->policy = policy;
  3482. switch (p->policy) {
  3483. case SCHED_NORMAL:
  3484. case SCHED_BATCH:
  3485. case SCHED_IDLE:
  3486. p->sched_class = &fair_sched_class;
  3487. break;
  3488. case SCHED_FIFO:
  3489. case SCHED_RR:
  3490. p->sched_class = &rt_sched_class;
  3491. break;
  3492. }
  3493. p->rt_priority = prio;
  3494. p->normal_prio = normal_prio(p);
  3495. /* we are holding p->pi_lock already */
  3496. p->prio = rt_mutex_getprio(p);
  3497. set_load_weight(p);
  3498. }
  3499. /**
  3500. * sched_setscheduler - change the scheduling policy and/or RT priority of a thread.
  3501. * @p: the task in question.
  3502. * @policy: new policy.
  3503. * @param: structure containing the new RT priority.
  3504. *
  3505. * NOTE that the task may be already dead.
  3506. */
  3507. int sched_setscheduler(struct task_struct *p, int policy,
  3508. struct sched_param *param)
  3509. {
  3510. int retval, oldprio, oldpolicy = -1, on_rq;
  3511. unsigned long flags;
  3512. struct rq *rq;
  3513. /* may grab non-irq protected spin_locks */
  3514. BUG_ON(in_interrupt());
  3515. recheck:
  3516. /* double check policy once rq lock held */
  3517. if (policy < 0)
  3518. policy = oldpolicy = p->policy;
  3519. else if (policy != SCHED_FIFO && policy != SCHED_RR &&
  3520. policy != SCHED_NORMAL && policy != SCHED_BATCH &&
  3521. policy != SCHED_IDLE)
  3522. return -EINVAL;
  3523. /*
  3524. * Valid priorities for SCHED_FIFO and SCHED_RR are
  3525. * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL,
  3526. * SCHED_BATCH and SCHED_IDLE is 0.
  3527. */
  3528. if (param->sched_priority < 0 ||
  3529. (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) ||
  3530. (!p->mm && param->sched_priority > MAX_RT_PRIO-1))
  3531. return -EINVAL;
  3532. if (rt_policy(policy) != (param->sched_priority != 0))
  3533. return -EINVAL;
  3534. /*
  3535. * Allow unprivileged RT tasks to decrease priority:
  3536. */
  3537. if (!capable(CAP_SYS_NICE)) {
  3538. if (rt_policy(policy)) {
  3539. unsigned long rlim_rtprio;
  3540. if (!lock_task_sighand(p, &flags))
  3541. return -ESRCH;
  3542. rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur;
  3543. unlock_task_sighand(p, &flags);
  3544. /* can't set/change the rt policy */
  3545. if (policy != p->policy && !rlim_rtprio)
  3546. return -EPERM;
  3547. /* can't increase priority */
  3548. if (param->sched_priority > p->rt_priority &&
  3549. param->sched_priority > rlim_rtprio)
  3550. return -EPERM;
  3551. }
  3552. /*
  3553. * Like positive nice levels, dont allow tasks to
  3554. * move out of SCHED_IDLE either:
  3555. */
  3556. if (p->policy == SCHED_IDLE && policy != SCHED_IDLE)
  3557. return -EPERM;
  3558. /* can't change other user's priorities */
  3559. if ((current->euid != p->euid) &&
  3560. (current->euid != p->uid))
  3561. return -EPERM;
  3562. }
  3563. retval = security_task_setscheduler(p, policy, param);
  3564. if (retval)
  3565. return retval;
  3566. /*
  3567. * make sure no PI-waiters arrive (or leave) while we are
  3568. * changing the priority of the task:
  3569. */
  3570. spin_lock_irqsave(&p->pi_lock, flags);
  3571. /*
  3572. * To be able to change p->policy safely, the apropriate
  3573. * runqueue lock must be held.
  3574. */
  3575. rq = __task_rq_lock(p);
  3576. /* recheck policy now with rq lock held */
  3577. if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) {
  3578. policy = oldpolicy = -1;
  3579. __task_rq_unlock(rq);
  3580. spin_unlock_irqrestore(&p->pi_lock, flags);
  3581. goto recheck;
  3582. }
  3583. on_rq = p->se.on_rq;
  3584. if (on_rq)
  3585. deactivate_task(rq, p, 0);
  3586. oldprio = p->prio;
  3587. __setscheduler(rq, p, policy, param->sched_priority);
  3588. if (on_rq) {
  3589. activate_task(rq, p, 0);
  3590. /*
  3591. * Reschedule if we are currently running on this runqueue and
  3592. * our priority decreased, or if we are not currently running on
  3593. * this runqueue and our priority is higher than the current's
  3594. */
  3595. if (task_running(rq, p)) {
  3596. if (p->prio > oldprio)
  3597. resched_task(rq->curr);
  3598. } else {
  3599. check_preempt_curr(rq, p);
  3600. }
  3601. }
  3602. __task_rq_unlock(rq);
  3603. spin_unlock_irqrestore(&p->pi_lock, flags);
  3604. rt_mutex_adjust_pi(p);
  3605. return 0;
  3606. }
  3607. EXPORT_SYMBOL_GPL(sched_setscheduler);
  3608. static int
  3609. do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param)
  3610. {
  3611. struct sched_param lparam;
  3612. struct task_struct *p;
  3613. int retval;
  3614. if (!param || pid < 0)
  3615. return -EINVAL;
  3616. if (copy_from_user(&lparam, param, sizeof(struct sched_param)))
  3617. return -EFAULT;
  3618. rcu_read_lock();
  3619. retval = -ESRCH;
  3620. p = find_process_by_pid(pid);
  3621. if (p != NULL)
  3622. retval = sched_setscheduler(p, policy, &lparam);
  3623. rcu_read_unlock();
  3624. return retval;
  3625. }
  3626. /**
  3627. * sys_sched_setscheduler - set/change the scheduler policy and RT priority
  3628. * @pid: the pid in question.
  3629. * @policy: new policy.
  3630. * @param: structure containing the new RT priority.
  3631. */
  3632. asmlinkage long sys_sched_setscheduler(pid_t pid, int policy,
  3633. struct sched_param __user *param)
  3634. {
  3635. /* negative values for policy are not valid */
  3636. if (policy < 0)
  3637. return -EINVAL;
  3638. return do_sched_setscheduler(pid, policy, param);
  3639. }
  3640. /**
  3641. * sys_sched_setparam - set/change the RT priority of a thread
  3642. * @pid: the pid in question.
  3643. * @param: structure containing the new RT priority.
  3644. */
  3645. asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param)
  3646. {
  3647. return do_sched_setscheduler(pid, -1, param);
  3648. }
  3649. /**
  3650. * sys_sched_getscheduler - get the policy (scheduling class) of a thread
  3651. * @pid: the pid in question.
  3652. */
  3653. asmlinkage long sys_sched_getscheduler(pid_t pid)
  3654. {
  3655. struct task_struct *p;
  3656. int retval = -EINVAL;
  3657. if (pid < 0)
  3658. goto out_nounlock;
  3659. retval = -ESRCH;
  3660. read_lock(&tasklist_lock);
  3661. p = find_process_by_pid(pid);
  3662. if (p) {
  3663. retval = security_task_getscheduler(p);
  3664. if (!retval)
  3665. retval = p->policy;
  3666. }
  3667. read_unlock(&tasklist_lock);
  3668. out_nounlock:
  3669. return retval;
  3670. }
  3671. /**
  3672. * sys_sched_getscheduler - get the RT priority of a thread
  3673. * @pid: the pid in question.
  3674. * @param: structure containing the RT priority.
  3675. */
  3676. asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param)
  3677. {
  3678. struct sched_param lp;
  3679. struct task_struct *p;
  3680. int retval = -EINVAL;
  3681. if (!param || pid < 0)
  3682. goto out_nounlock;
  3683. read_lock(&tasklist_lock);
  3684. p = find_process_by_pid(pid);
  3685. retval = -ESRCH;
  3686. if (!p)
  3687. goto out_unlock;
  3688. retval = security_task_getscheduler(p);
  3689. if (retval)
  3690. goto out_unlock;
  3691. lp.sched_priority = p->rt_priority;
  3692. read_unlock(&tasklist_lock);
  3693. /*
  3694. * This one might sleep, we cannot do it with a spinlock held ...
  3695. */
  3696. retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0;
  3697. out_nounlock:
  3698. return retval;
  3699. out_unlock:
  3700. read_unlock(&tasklist_lock);
  3701. return retval;
  3702. }
  3703. long sched_setaffinity(pid_t pid, cpumask_t new_mask)
  3704. {
  3705. cpumask_t cpus_allowed;
  3706. struct task_struct *p;
  3707. int retval;
  3708. mutex_lock(&sched_hotcpu_mutex);
  3709. read_lock(&tasklist_lock);
  3710. p = find_process_by_pid(pid);
  3711. if (!p) {
  3712. read_unlock(&tasklist_lock);
  3713. mutex_unlock(&sched_hotcpu_mutex);
  3714. return -ESRCH;
  3715. }
  3716. /*
  3717. * It is not safe to call set_cpus_allowed with the
  3718. * tasklist_lock held. We will bump the task_struct's
  3719. * usage count and then drop tasklist_lock.
  3720. */
  3721. get_task_struct(p);
  3722. read_unlock(&tasklist_lock);
  3723. retval = -EPERM;
  3724. if ((current->euid != p->euid) && (current->euid != p->uid) &&
  3725. !capable(CAP_SYS_NICE))
  3726. goto out_unlock;
  3727. retval = security_task_setscheduler(p, 0, NULL);
  3728. if (retval)
  3729. goto out_unlock;
  3730. cpus_allowed = cpuset_cpus_allowed(p);
  3731. cpus_and(new_mask, new_mask, cpus_allowed);
  3732. retval = set_cpus_allowed(p, new_mask);
  3733. out_unlock:
  3734. put_task_struct(p);
  3735. mutex_unlock(&sched_hotcpu_mutex);
  3736. return retval;
  3737. }
  3738. static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len,
  3739. cpumask_t *new_mask)
  3740. {
  3741. if (len < sizeof(cpumask_t)) {
  3742. memset(new_mask, 0, sizeof(cpumask_t));
  3743. } else if (len > sizeof(cpumask_t)) {
  3744. len = sizeof(cpumask_t);
  3745. }
  3746. return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0;
  3747. }
  3748. /**
  3749. * sys_sched_setaffinity - set the cpu affinity of a process
  3750. * @pid: pid of the process
  3751. * @len: length in bytes of the bitmask pointed to by user_mask_ptr
  3752. * @user_mask_ptr: user-space pointer to the new cpu mask
  3753. */
  3754. asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len,
  3755. unsigned long __user *user_mask_ptr)
  3756. {
  3757. cpumask_t new_mask;
  3758. int retval;
  3759. retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask);
  3760. if (retval)
  3761. return retval;
  3762. return sched_setaffinity(pid, new_mask);
  3763. }
  3764. /*
  3765. * Represents all cpu's present in the system
  3766. * In systems capable of hotplug, this map could dynamically grow
  3767. * as new cpu's are detected in the system via any platform specific
  3768. * method, such as ACPI for e.g.
  3769. */
  3770. cpumask_t cpu_present_map __read_mostly;
  3771. EXPORT_SYMBOL(cpu_present_map);
  3772. #ifndef CONFIG_SMP
  3773. cpumask_t cpu_online_map __read_mostly = CPU_MASK_ALL;
  3774. EXPORT_SYMBOL(cpu_online_map);
  3775. cpumask_t cpu_possible_map __read_mostly = CPU_MASK_ALL;
  3776. EXPORT_SYMBOL(cpu_possible_map);
  3777. #endif
  3778. long sched_getaffinity(pid_t pid, cpumask_t *mask)
  3779. {
  3780. struct task_struct *p;
  3781. int retval;
  3782. mutex_lock(&sched_hotcpu_mutex);
  3783. read_lock(&tasklist_lock);
  3784. retval = -ESRCH;
  3785. p = find_process_by_pid(pid);
  3786. if (!p)
  3787. goto out_unlock;
  3788. retval = security_task_getscheduler(p);
  3789. if (retval)
  3790. goto out_unlock;
  3791. cpus_and(*mask, p->cpus_allowed, cpu_online_map);
  3792. out_unlock:
  3793. read_unlock(&tasklist_lock);
  3794. mutex_unlock(&sched_hotcpu_mutex);
  3795. if (retval)
  3796. return retval;
  3797. return 0;
  3798. }
  3799. /**
  3800. * sys_sched_getaffinity - get the cpu affinity of a process
  3801. * @pid: pid of the process
  3802. * @len: length in bytes of the bitmask pointed to by user_mask_ptr
  3803. * @user_mask_ptr: user-space pointer to hold the current cpu mask
  3804. */
  3805. asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len,
  3806. unsigned long __user *user_mask_ptr)
  3807. {
  3808. int ret;
  3809. cpumask_t mask;
  3810. if (len < sizeof(cpumask_t))
  3811. return -EINVAL;
  3812. ret = sched_getaffinity(pid, &mask);
  3813. if (ret < 0)
  3814. return ret;
  3815. if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t)))
  3816. return -EFAULT;
  3817. return sizeof(cpumask_t);
  3818. }
  3819. /**
  3820. * sys_sched_yield - yield the current processor to other threads.
  3821. *
  3822. * This function yields the current CPU to other tasks. If there are no
  3823. * other threads running on this CPU then this function will return.
  3824. */
  3825. asmlinkage long sys_sched_yield(void)
  3826. {
  3827. struct rq *rq = this_rq_lock();
  3828. schedstat_inc(rq, yld_cnt);
  3829. if (unlikely(rq->nr_running == 1))
  3830. schedstat_inc(rq, yld_act_empty);
  3831. else
  3832. current->sched_class->yield_task(rq, current);
  3833. /*
  3834. * Since we are going to call schedule() anyway, there's
  3835. * no need to preempt or enable interrupts:
  3836. */
  3837. __release(rq->lock);
  3838. spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
  3839. _raw_spin_unlock(&rq->lock);
  3840. preempt_enable_no_resched();
  3841. schedule();
  3842. return 0;
  3843. }
  3844. static void __cond_resched(void)
  3845. {
  3846. #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP
  3847. __might_sleep(__FILE__, __LINE__);
  3848. #endif
  3849. /*
  3850. * The BKS might be reacquired before we have dropped
  3851. * PREEMPT_ACTIVE, which could trigger a second
  3852. * cond_resched() call.
  3853. */
  3854. do {
  3855. add_preempt_count(PREEMPT_ACTIVE);
  3856. schedule();
  3857. sub_preempt_count(PREEMPT_ACTIVE);
  3858. } while (need_resched());
  3859. }
  3860. int __sched cond_resched(void)
  3861. {
  3862. if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) &&
  3863. system_state == SYSTEM_RUNNING) {
  3864. __cond_resched();
  3865. return 1;
  3866. }
  3867. return 0;
  3868. }
  3869. EXPORT_SYMBOL(cond_resched);
  3870. /*
  3871. * cond_resched_lock() - if a reschedule is pending, drop the given lock,
  3872. * call schedule, and on return reacquire the lock.
  3873. *
  3874. * This works OK both with and without CONFIG_PREEMPT. We do strange low-level
  3875. * operations here to prevent schedule() from being called twice (once via
  3876. * spin_unlock(), once by hand).
  3877. */
  3878. int cond_resched_lock(spinlock_t *lock)
  3879. {
  3880. int ret = 0;
  3881. if (need_lockbreak(lock)) {
  3882. spin_unlock(lock);
  3883. cpu_relax();
  3884. ret = 1;
  3885. spin_lock(lock);
  3886. }
  3887. if (need_resched() && system_state == SYSTEM_RUNNING) {
  3888. spin_release(&lock->dep_map, 1, _THIS_IP_);
  3889. _raw_spin_unlock(lock);
  3890. preempt_enable_no_resched();
  3891. __cond_resched();
  3892. ret = 1;
  3893. spin_lock(lock);
  3894. }
  3895. return ret;
  3896. }
  3897. EXPORT_SYMBOL(cond_resched_lock);
  3898. int __sched cond_resched_softirq(void)
  3899. {
  3900. BUG_ON(!in_softirq());
  3901. if (need_resched() && system_state == SYSTEM_RUNNING) {
  3902. local_bh_enable();
  3903. __cond_resched();
  3904. local_bh_disable();
  3905. return 1;
  3906. }
  3907. return 0;
  3908. }
  3909. EXPORT_SYMBOL(cond_resched_softirq);
  3910. /**
  3911. * yield - yield the current processor to other threads.
  3912. *
  3913. * This is a shortcut for kernel-space yielding - it marks the
  3914. * thread runnable and calls sys_sched_yield().
  3915. */
  3916. void __sched yield(void)
  3917. {
  3918. set_current_state(TASK_RUNNING);
  3919. sys_sched_yield();
  3920. }
  3921. EXPORT_SYMBOL(yield);
  3922. /*
  3923. * This task is about to go to sleep on IO. Increment rq->nr_iowait so
  3924. * that process accounting knows that this is a task in IO wait state.
  3925. *
  3926. * But don't do that if it is a deliberate, throttling IO wait (this task
  3927. * has set its backing_dev_info: the queue against which it should throttle)
  3928. */
  3929. void __sched io_schedule(void)
  3930. {
  3931. struct rq *rq = &__raw_get_cpu_var(runqueues);
  3932. delayacct_blkio_start();
  3933. atomic_inc(&rq->nr_iowait);
  3934. schedule();
  3935. atomic_dec(&rq->nr_iowait);
  3936. delayacct_blkio_end();
  3937. }
  3938. EXPORT_SYMBOL(io_schedule);
  3939. long __sched io_schedule_timeout(long timeout)
  3940. {
  3941. struct rq *rq = &__raw_get_cpu_var(runqueues);
  3942. long ret;
  3943. delayacct_blkio_start();
  3944. atomic_inc(&rq->nr_iowait);
  3945. ret = schedule_timeout(timeout);
  3946. atomic_dec(&rq->nr_iowait);
  3947. delayacct_blkio_end();
  3948. return ret;
  3949. }
  3950. /**
  3951. * sys_sched_get_priority_max - return maximum RT priority.
  3952. * @policy: scheduling class.
  3953. *
  3954. * this syscall returns the maximum rt_priority that can be used
  3955. * by a given scheduling class.
  3956. */
  3957. asmlinkage long sys_sched_get_priority_max(int policy)
  3958. {
  3959. int ret = -EINVAL;
  3960. switch (policy) {
  3961. case SCHED_FIFO:
  3962. case SCHED_RR:
  3963. ret = MAX_USER_RT_PRIO-1;
  3964. break;
  3965. case SCHED_NORMAL:
  3966. case SCHED_BATCH:
  3967. case SCHED_IDLE:
  3968. ret = 0;
  3969. break;
  3970. }
  3971. return ret;
  3972. }
  3973. /**
  3974. * sys_sched_get_priority_min - return minimum RT priority.
  3975. * @policy: scheduling class.
  3976. *
  3977. * this syscall returns the minimum rt_priority that can be used
  3978. * by a given scheduling class.
  3979. */
  3980. asmlinkage long sys_sched_get_priority_min(int policy)
  3981. {
  3982. int ret = -EINVAL;
  3983. switch (policy) {
  3984. case SCHED_FIFO:
  3985. case SCHED_RR:
  3986. ret = 1;
  3987. break;
  3988. case SCHED_NORMAL:
  3989. case SCHED_BATCH:
  3990. case SCHED_IDLE:
  3991. ret = 0;
  3992. }
  3993. return ret;
  3994. }
  3995. /**
  3996. * sys_sched_rr_get_interval - return the default timeslice of a process.
  3997. * @pid: pid of the process.
  3998. * @interval: userspace pointer to the timeslice value.
  3999. *
  4000. * this syscall writes the default timeslice value of a given process
  4001. * into the user-space timespec buffer. A value of '0' means infinity.
  4002. */
  4003. asmlinkage
  4004. long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval)
  4005. {
  4006. struct task_struct *p;
  4007. int retval = -EINVAL;
  4008. struct timespec t;
  4009. if (pid < 0)
  4010. goto out_nounlock;
  4011. retval = -ESRCH;
  4012. read_lock(&tasklist_lock);
  4013. p = find_process_by_pid(pid);
  4014. if (!p)
  4015. goto out_unlock;
  4016. retval = security_task_getscheduler(p);
  4017. if (retval)
  4018. goto out_unlock;
  4019. jiffies_to_timespec(p->policy == SCHED_FIFO ?
  4020. 0 : static_prio_timeslice(p->static_prio), &t);
  4021. read_unlock(&tasklist_lock);
  4022. retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0;
  4023. out_nounlock:
  4024. return retval;
  4025. out_unlock:
  4026. read_unlock(&tasklist_lock);
  4027. return retval;
  4028. }
  4029. static const char stat_nam[] = "RSDTtZX";
  4030. static void show_task(struct task_struct *p)
  4031. {
  4032. unsigned long free = 0;
  4033. unsigned state;
  4034. state = p->state ? __ffs(p->state) + 1 : 0;
  4035. printk("%-13.13s %c", p->comm,
  4036. state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?');
  4037. #if BITS_PER_LONG == 32
  4038. if (state == TASK_RUNNING)
  4039. printk(" running ");
  4040. else
  4041. printk(" %08lx ", thread_saved_pc(p));
  4042. #else
  4043. if (state == TASK_RUNNING)
  4044. printk(" running task ");
  4045. else
  4046. printk(" %016lx ", thread_saved_pc(p));
  4047. #endif
  4048. #ifdef CONFIG_DEBUG_STACK_USAGE
  4049. {
  4050. unsigned long *n = end_of_stack(p);
  4051. while (!*n)
  4052. n++;
  4053. free = (unsigned long)n - (unsigned long)end_of_stack(p);
  4054. }
  4055. #endif
  4056. printk("%5lu %5d %6d\n", free, p->pid, p->parent->pid);
  4057. if (state != TASK_RUNNING)
  4058. show_stack(p, NULL);
  4059. }
  4060. void show_state_filter(unsigned long state_filter)
  4061. {
  4062. struct task_struct *g, *p;
  4063. #if BITS_PER_LONG == 32
  4064. printk(KERN_INFO
  4065. " task PC stack pid father\n");
  4066. #else
  4067. printk(KERN_INFO
  4068. " task PC stack pid father\n");
  4069. #endif
  4070. read_lock(&tasklist_lock);
  4071. do_each_thread(g, p) {
  4072. /*
  4073. * reset the NMI-timeout, listing all files on a slow
  4074. * console might take alot of time:
  4075. */
  4076. touch_nmi_watchdog();
  4077. if (!state_filter || (p->state & state_filter))
  4078. show_task(p);
  4079. } while_each_thread(g, p);
  4080. touch_all_softlockup_watchdogs();
  4081. #ifdef CONFIG_SCHED_DEBUG
  4082. sysrq_sched_debug_show();
  4083. #endif
  4084. read_unlock(&tasklist_lock);
  4085. /*
  4086. * Only show locks if all tasks are dumped:
  4087. */
  4088. if (state_filter == -1)
  4089. debug_show_all_locks();
  4090. }
  4091. void __cpuinit init_idle_bootup_task(struct task_struct *idle)
  4092. {
  4093. idle->sched_class = &idle_sched_class;
  4094. }
  4095. /**
  4096. * init_idle - set up an idle thread for a given CPU
  4097. * @idle: task in question
  4098. * @cpu: cpu the idle task belongs to
  4099. *
  4100. * NOTE: this function does not set the idle thread's NEED_RESCHED
  4101. * flag, to make booting more robust.
  4102. */
  4103. void __cpuinit init_idle(struct task_struct *idle, int cpu)
  4104. {
  4105. struct rq *rq = cpu_rq(cpu);
  4106. unsigned long flags;
  4107. __sched_fork(idle);
  4108. idle->se.exec_start = sched_clock();
  4109. idle->prio = idle->normal_prio = MAX_PRIO;
  4110. idle->cpus_allowed = cpumask_of_cpu(cpu);
  4111. __set_task_cpu(idle, cpu);
  4112. spin_lock_irqsave(&rq->lock, flags);
  4113. rq->curr = rq->idle = idle;
  4114. #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW)
  4115. idle->oncpu = 1;
  4116. #endif
  4117. spin_unlock_irqrestore(&rq->lock, flags);
  4118. /* Set the preempt count _outside_ the spinlocks! */
  4119. #if defined(CONFIG_PREEMPT) && !defined(CONFIG_PREEMPT_BKL)
  4120. task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0);
  4121. #else
  4122. task_thread_info(idle)->preempt_count = 0;
  4123. #endif
  4124. /*
  4125. * The idle tasks have their own, simple scheduling class:
  4126. */
  4127. idle->sched_class = &idle_sched_class;
  4128. }
  4129. /*
  4130. * In a system that switches off the HZ timer nohz_cpu_mask
  4131. * indicates which cpus entered this state. This is used
  4132. * in the rcu update to wait only for active cpus. For system
  4133. * which do not switch off the HZ timer nohz_cpu_mask should
  4134. * always be CPU_MASK_NONE.
  4135. */
  4136. cpumask_t nohz_cpu_mask = CPU_MASK_NONE;
  4137. /*
  4138. * Increase the granularity value when there are more CPUs,
  4139. * because with more CPUs the 'effective latency' as visible
  4140. * to users decreases. But the relationship is not linear,
  4141. * so pick a second-best guess by going with the log2 of the
  4142. * number of CPUs.
  4143. *
  4144. * This idea comes from the SD scheduler of Con Kolivas:
  4145. */
  4146. static inline void sched_init_granularity(void)
  4147. {
  4148. unsigned int factor = 1 + ilog2(num_online_cpus());
  4149. const unsigned long gran_limit = 100000000;
  4150. sysctl_sched_granularity *= factor;
  4151. if (sysctl_sched_granularity > gran_limit)
  4152. sysctl_sched_granularity = gran_limit;
  4153. sysctl_sched_runtime_limit = sysctl_sched_granularity * 4;
  4154. sysctl_sched_wakeup_granularity = sysctl_sched_granularity / 2;
  4155. }
  4156. #ifdef CONFIG_SMP
  4157. /*
  4158. * This is how migration works:
  4159. *
  4160. * 1) we queue a struct migration_req structure in the source CPU's
  4161. * runqueue and wake up that CPU's migration thread.
  4162. * 2) we down() the locked semaphore => thread blocks.
  4163. * 3) migration thread wakes up (implicitly it forces the migrated
  4164. * thread off the CPU)
  4165. * 4) it gets the migration request and checks whether the migrated
  4166. * task is still in the wrong runqueue.
  4167. * 5) if it's in the wrong runqueue then the migration thread removes
  4168. * it and puts it into the right queue.
  4169. * 6) migration thread up()s the semaphore.
  4170. * 7) we wake up and the migration is done.
  4171. */
  4172. /*
  4173. * Change a given task's CPU affinity. Migrate the thread to a
  4174. * proper CPU and schedule it away if the CPU it's executing on
  4175. * is removed from the allowed bitmask.
  4176. *
  4177. * NOTE: the caller must have a valid reference to the task, the
  4178. * task must not exit() & deallocate itself prematurely. The
  4179. * call is not atomic; no spinlocks may be held.
  4180. */
  4181. int set_cpus_allowed(struct task_struct *p, cpumask_t new_mask)
  4182. {
  4183. struct migration_req req;
  4184. unsigned long flags;
  4185. struct rq *rq;
  4186. int ret = 0;
  4187. rq = task_rq_lock(p, &flags);
  4188. if (!cpus_intersects(new_mask, cpu_online_map)) {
  4189. ret = -EINVAL;
  4190. goto out;
  4191. }
  4192. p->cpus_allowed = new_mask;
  4193. /* Can the task run on the task's current CPU? If so, we're done */
  4194. if (cpu_isset(task_cpu(p), new_mask))
  4195. goto out;
  4196. if (migrate_task(p, any_online_cpu(new_mask), &req)) {
  4197. /* Need help from migration thread: drop lock and wait. */
  4198. task_rq_unlock(rq, &flags);
  4199. wake_up_process(rq->migration_thread);
  4200. wait_for_completion(&req.done);
  4201. tlb_migrate_finish(p->mm);
  4202. return 0;
  4203. }
  4204. out:
  4205. task_rq_unlock(rq, &flags);
  4206. return ret;
  4207. }
  4208. EXPORT_SYMBOL_GPL(set_cpus_allowed);
  4209. /*
  4210. * Move (not current) task off this cpu, onto dest cpu. We're doing
  4211. * this because either it can't run here any more (set_cpus_allowed()
  4212. * away from this CPU, or CPU going down), or because we're
  4213. * attempting to rebalance this task on exec (sched_exec).
  4214. *
  4215. * So we race with normal scheduler movements, but that's OK, as long
  4216. * as the task is no longer on this CPU.
  4217. *
  4218. * Returns non-zero if task was successfully migrated.
  4219. */
  4220. static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu)
  4221. {
  4222. struct rq *rq_dest, *rq_src;
  4223. int ret = 0, on_rq;
  4224. if (unlikely(cpu_is_offline(dest_cpu)))
  4225. return ret;
  4226. rq_src = cpu_rq(src_cpu);
  4227. rq_dest = cpu_rq(dest_cpu);
  4228. double_rq_lock(rq_src, rq_dest);
  4229. /* Already moved. */
  4230. if (task_cpu(p) != src_cpu)
  4231. goto out;
  4232. /* Affinity changed (again). */
  4233. if (!cpu_isset(dest_cpu, p->cpus_allowed))
  4234. goto out;
  4235. on_rq = p->se.on_rq;
  4236. if (on_rq)
  4237. deactivate_task(rq_src, p, 0);
  4238. set_task_cpu(p, dest_cpu);
  4239. if (on_rq) {
  4240. activate_task(rq_dest, p, 0);
  4241. check_preempt_curr(rq_dest, p);
  4242. }
  4243. ret = 1;
  4244. out:
  4245. double_rq_unlock(rq_src, rq_dest);
  4246. return ret;
  4247. }
  4248. /*
  4249. * migration_thread - this is a highprio system thread that performs
  4250. * thread migration by bumping thread off CPU then 'pushing' onto
  4251. * another runqueue.
  4252. */
  4253. static int migration_thread(void *data)
  4254. {
  4255. int cpu = (long)data;
  4256. struct rq *rq;
  4257. rq = cpu_rq(cpu);
  4258. BUG_ON(rq->migration_thread != current);
  4259. set_current_state(TASK_INTERRUPTIBLE);
  4260. while (!kthread_should_stop()) {
  4261. struct migration_req *req;
  4262. struct list_head *head;
  4263. spin_lock_irq(&rq->lock);
  4264. if (cpu_is_offline(cpu)) {
  4265. spin_unlock_irq(&rq->lock);
  4266. goto wait_to_die;
  4267. }
  4268. if (rq->active_balance) {
  4269. active_load_balance(rq, cpu);
  4270. rq->active_balance = 0;
  4271. }
  4272. head = &rq->migration_queue;
  4273. if (list_empty(head)) {
  4274. spin_unlock_irq(&rq->lock);
  4275. schedule();
  4276. set_current_state(TASK_INTERRUPTIBLE);
  4277. continue;
  4278. }
  4279. req = list_entry(head->next, struct migration_req, list);
  4280. list_del_init(head->next);
  4281. spin_unlock(&rq->lock);
  4282. __migrate_task(req->task, cpu, req->dest_cpu);
  4283. local_irq_enable();
  4284. complete(&req->done);
  4285. }
  4286. __set_current_state(TASK_RUNNING);
  4287. return 0;
  4288. wait_to_die:
  4289. /* Wait for kthread_stop */
  4290. set_current_state(TASK_INTERRUPTIBLE);
  4291. while (!kthread_should_stop()) {
  4292. schedule();
  4293. set_current_state(TASK_INTERRUPTIBLE);
  4294. }
  4295. __set_current_state(TASK_RUNNING);
  4296. return 0;
  4297. }
  4298. #ifdef CONFIG_HOTPLUG_CPU
  4299. /*
  4300. * Figure out where task on dead CPU should go, use force if neccessary.
  4301. * NOTE: interrupts should be disabled by the caller
  4302. */
  4303. static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p)
  4304. {
  4305. unsigned long flags;
  4306. cpumask_t mask;
  4307. struct rq *rq;
  4308. int dest_cpu;
  4309. restart:
  4310. /* On same node? */
  4311. mask = node_to_cpumask(cpu_to_node(dead_cpu));
  4312. cpus_and(mask, mask, p->cpus_allowed);
  4313. dest_cpu = any_online_cpu(mask);
  4314. /* On any allowed CPU? */
  4315. if (dest_cpu == NR_CPUS)
  4316. dest_cpu = any_online_cpu(p->cpus_allowed);
  4317. /* No more Mr. Nice Guy. */
  4318. if (dest_cpu == NR_CPUS) {
  4319. rq = task_rq_lock(p, &flags);
  4320. cpus_setall(p->cpus_allowed);
  4321. dest_cpu = any_online_cpu(p->cpus_allowed);
  4322. task_rq_unlock(rq, &flags);
  4323. /*
  4324. * Don't tell them about moving exiting tasks or
  4325. * kernel threads (both mm NULL), since they never
  4326. * leave kernel.
  4327. */
  4328. if (p->mm && printk_ratelimit())
  4329. printk(KERN_INFO "process %d (%s) no "
  4330. "longer affine to cpu%d\n",
  4331. p->pid, p->comm, dead_cpu);
  4332. }
  4333. if (!__migrate_task(p, dead_cpu, dest_cpu))
  4334. goto restart;
  4335. }
  4336. /*
  4337. * While a dead CPU has no uninterruptible tasks queued at this point,
  4338. * it might still have a nonzero ->nr_uninterruptible counter, because
  4339. * for performance reasons the counter is not stricly tracking tasks to
  4340. * their home CPUs. So we just add the counter to another CPU's counter,
  4341. * to keep the global sum constant after CPU-down:
  4342. */
  4343. static void migrate_nr_uninterruptible(struct rq *rq_src)
  4344. {
  4345. struct rq *rq_dest = cpu_rq(any_online_cpu(CPU_MASK_ALL));
  4346. unsigned long flags;
  4347. local_irq_save(flags);
  4348. double_rq_lock(rq_src, rq_dest);
  4349. rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible;
  4350. rq_src->nr_uninterruptible = 0;
  4351. double_rq_unlock(rq_src, rq_dest);
  4352. local_irq_restore(flags);
  4353. }
  4354. /* Run through task list and migrate tasks from the dead cpu. */
  4355. static void migrate_live_tasks(int src_cpu)
  4356. {
  4357. struct task_struct *p, *t;
  4358. write_lock_irq(&tasklist_lock);
  4359. do_each_thread(t, p) {
  4360. if (p == current)
  4361. continue;
  4362. if (task_cpu(p) == src_cpu)
  4363. move_task_off_dead_cpu(src_cpu, p);
  4364. } while_each_thread(t, p);
  4365. write_unlock_irq(&tasklist_lock);
  4366. }
  4367. /*
  4368. * Schedules idle task to be the next runnable task on current CPU.
  4369. * It does so by boosting its priority to highest possible and adding it to
  4370. * the _front_ of the runqueue. Used by CPU offline code.
  4371. */
  4372. void sched_idle_next(void)
  4373. {
  4374. int this_cpu = smp_processor_id();
  4375. struct rq *rq = cpu_rq(this_cpu);
  4376. struct task_struct *p = rq->idle;
  4377. unsigned long flags;
  4378. /* cpu has to be offline */
  4379. BUG_ON(cpu_online(this_cpu));
  4380. /*
  4381. * Strictly not necessary since rest of the CPUs are stopped by now
  4382. * and interrupts disabled on the current cpu.
  4383. */
  4384. spin_lock_irqsave(&rq->lock, flags);
  4385. __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1);
  4386. /* Add idle task to the _front_ of its priority queue: */
  4387. activate_idle_task(p, rq);
  4388. spin_unlock_irqrestore(&rq->lock, flags);
  4389. }
  4390. /*
  4391. * Ensures that the idle task is using init_mm right before its cpu goes
  4392. * offline.
  4393. */
  4394. void idle_task_exit(void)
  4395. {
  4396. struct mm_struct *mm = current->active_mm;
  4397. BUG_ON(cpu_online(smp_processor_id()));
  4398. if (mm != &init_mm)
  4399. switch_mm(mm, &init_mm, current);
  4400. mmdrop(mm);
  4401. }
  4402. /* called under rq->lock with disabled interrupts */
  4403. static void migrate_dead(unsigned int dead_cpu, struct task_struct *p)
  4404. {
  4405. struct rq *rq = cpu_rq(dead_cpu);
  4406. /* Must be exiting, otherwise would be on tasklist. */
  4407. BUG_ON(p->exit_state != EXIT_ZOMBIE && p->exit_state != EXIT_DEAD);
  4408. /* Cannot have done final schedule yet: would have vanished. */
  4409. BUG_ON(p->state == TASK_DEAD);
  4410. get_task_struct(p);
  4411. /*
  4412. * Drop lock around migration; if someone else moves it,
  4413. * that's OK. No task can be added to this CPU, so iteration is
  4414. * fine.
  4415. * NOTE: interrupts should be left disabled --dev@
  4416. */
  4417. spin_unlock(&rq->lock);
  4418. move_task_off_dead_cpu(dead_cpu, p);
  4419. spin_lock(&rq->lock);
  4420. put_task_struct(p);
  4421. }
  4422. /* release_task() removes task from tasklist, so we won't find dead tasks. */
  4423. static void migrate_dead_tasks(unsigned int dead_cpu)
  4424. {
  4425. struct rq *rq = cpu_rq(dead_cpu);
  4426. struct task_struct *next;
  4427. for ( ; ; ) {
  4428. if (!rq->nr_running)
  4429. break;
  4430. next = pick_next_task(rq, rq->curr, rq_clock(rq));
  4431. if (!next)
  4432. break;
  4433. migrate_dead(dead_cpu, next);
  4434. }
  4435. }
  4436. #endif /* CONFIG_HOTPLUG_CPU */
  4437. /*
  4438. * migration_call - callback that gets triggered when a CPU is added.
  4439. * Here we can start up the necessary migration thread for the new CPU.
  4440. */
  4441. static int __cpuinit
  4442. migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu)
  4443. {
  4444. struct task_struct *p;
  4445. int cpu = (long)hcpu;
  4446. unsigned long flags;
  4447. struct rq *rq;
  4448. switch (action) {
  4449. case CPU_LOCK_ACQUIRE:
  4450. mutex_lock(&sched_hotcpu_mutex);
  4451. break;
  4452. case CPU_UP_PREPARE:
  4453. case CPU_UP_PREPARE_FROZEN:
  4454. p = kthread_create(migration_thread, hcpu, "migration/%d", cpu);
  4455. if (IS_ERR(p))
  4456. return NOTIFY_BAD;
  4457. kthread_bind(p, cpu);
  4458. /* Must be high prio: stop_machine expects to yield to it. */
  4459. rq = task_rq_lock(p, &flags);
  4460. __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1);
  4461. task_rq_unlock(rq, &flags);
  4462. cpu_rq(cpu)->migration_thread = p;
  4463. break;
  4464. case CPU_ONLINE:
  4465. case CPU_ONLINE_FROZEN:
  4466. /* Strictly unneccessary, as first user will wake it. */
  4467. wake_up_process(cpu_rq(cpu)->migration_thread);
  4468. break;
  4469. #ifdef CONFIG_HOTPLUG_CPU
  4470. case CPU_UP_CANCELED:
  4471. case CPU_UP_CANCELED_FROZEN:
  4472. if (!cpu_rq(cpu)->migration_thread)
  4473. break;
  4474. /* Unbind it from offline cpu so it can run. Fall thru. */
  4475. kthread_bind(cpu_rq(cpu)->migration_thread,
  4476. any_online_cpu(cpu_online_map));
  4477. kthread_stop(cpu_rq(cpu)->migration_thread);
  4478. cpu_rq(cpu)->migration_thread = NULL;
  4479. break;
  4480. case CPU_DEAD:
  4481. case CPU_DEAD_FROZEN:
  4482. migrate_live_tasks(cpu);
  4483. rq = cpu_rq(cpu);
  4484. kthread_stop(rq->migration_thread);
  4485. rq->migration_thread = NULL;
  4486. /* Idle task back to normal (off runqueue, low prio) */
  4487. rq = task_rq_lock(rq->idle, &flags);
  4488. deactivate_task(rq, rq->idle, 0);
  4489. rq->idle->static_prio = MAX_PRIO;
  4490. __setscheduler(rq, rq->idle, SCHED_NORMAL, 0);
  4491. rq->idle->sched_class = &idle_sched_class;
  4492. migrate_dead_tasks(cpu);
  4493. task_rq_unlock(rq, &flags);
  4494. migrate_nr_uninterruptible(rq);
  4495. BUG_ON(rq->nr_running != 0);
  4496. /* No need to migrate the tasks: it was best-effort if
  4497. * they didn't take sched_hotcpu_mutex. Just wake up
  4498. * the requestors. */
  4499. spin_lock_irq(&rq->lock);
  4500. while (!list_empty(&rq->migration_queue)) {
  4501. struct migration_req *req;
  4502. req = list_entry(rq->migration_queue.next,
  4503. struct migration_req, list);
  4504. list_del_init(&req->list);
  4505. complete(&req->done);
  4506. }
  4507. spin_unlock_irq(&rq->lock);
  4508. break;
  4509. #endif
  4510. case CPU_LOCK_RELEASE:
  4511. mutex_unlock(&sched_hotcpu_mutex);
  4512. break;
  4513. }
  4514. return NOTIFY_OK;
  4515. }
  4516. /* Register at highest priority so that task migration (migrate_all_tasks)
  4517. * happens before everything else.
  4518. */
  4519. static struct notifier_block __cpuinitdata migration_notifier = {
  4520. .notifier_call = migration_call,
  4521. .priority = 10
  4522. };
  4523. int __init migration_init(void)
  4524. {
  4525. void *cpu = (void *)(long)smp_processor_id();
  4526. int err;
  4527. /* Start one for the boot CPU: */
  4528. err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu);
  4529. BUG_ON(err == NOTIFY_BAD);
  4530. migration_call(&migration_notifier, CPU_ONLINE, cpu);
  4531. register_cpu_notifier(&migration_notifier);
  4532. return 0;
  4533. }
  4534. #endif
  4535. #ifdef CONFIG_SMP
  4536. /* Number of possible processor ids */
  4537. int nr_cpu_ids __read_mostly = NR_CPUS;
  4538. EXPORT_SYMBOL(nr_cpu_ids);
  4539. #undef SCHED_DOMAIN_DEBUG
  4540. #ifdef SCHED_DOMAIN_DEBUG
  4541. static void sched_domain_debug(struct sched_domain *sd, int cpu)
  4542. {
  4543. int level = 0;
  4544. if (!sd) {
  4545. printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu);
  4546. return;
  4547. }
  4548. printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu);
  4549. do {
  4550. int i;
  4551. char str[NR_CPUS];
  4552. struct sched_group *group = sd->groups;
  4553. cpumask_t groupmask;
  4554. cpumask_scnprintf(str, NR_CPUS, sd->span);
  4555. cpus_clear(groupmask);
  4556. printk(KERN_DEBUG);
  4557. for (i = 0; i < level + 1; i++)
  4558. printk(" ");
  4559. printk("domain %d: ", level);
  4560. if (!(sd->flags & SD_LOAD_BALANCE)) {
  4561. printk("does not load-balance\n");
  4562. if (sd->parent)
  4563. printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain"
  4564. " has parent");
  4565. break;
  4566. }
  4567. printk("span %s\n", str);
  4568. if (!cpu_isset(cpu, sd->span))
  4569. printk(KERN_ERR "ERROR: domain->span does not contain "
  4570. "CPU%d\n", cpu);
  4571. if (!cpu_isset(cpu, group->cpumask))
  4572. printk(KERN_ERR "ERROR: domain->groups does not contain"
  4573. " CPU%d\n", cpu);
  4574. printk(KERN_DEBUG);
  4575. for (i = 0; i < level + 2; i++)
  4576. printk(" ");
  4577. printk("groups:");
  4578. do {
  4579. if (!group) {
  4580. printk("\n");
  4581. printk(KERN_ERR "ERROR: group is NULL\n");
  4582. break;
  4583. }
  4584. if (!group->__cpu_power) {
  4585. printk("\n");
  4586. printk(KERN_ERR "ERROR: domain->cpu_power not "
  4587. "set\n");
  4588. }
  4589. if (!cpus_weight(group->cpumask)) {
  4590. printk("\n");
  4591. printk(KERN_ERR "ERROR: empty group\n");
  4592. }
  4593. if (cpus_intersects(groupmask, group->cpumask)) {
  4594. printk("\n");
  4595. printk(KERN_ERR "ERROR: repeated CPUs\n");
  4596. }
  4597. cpus_or(groupmask, groupmask, group->cpumask);
  4598. cpumask_scnprintf(str, NR_CPUS, group->cpumask);
  4599. printk(" %s", str);
  4600. group = group->next;
  4601. } while (group != sd->groups);
  4602. printk("\n");
  4603. if (!cpus_equal(sd->span, groupmask))
  4604. printk(KERN_ERR "ERROR: groups don't span "
  4605. "domain->span\n");
  4606. level++;
  4607. sd = sd->parent;
  4608. if (!sd)
  4609. continue;
  4610. if (!cpus_subset(groupmask, sd->span))
  4611. printk(KERN_ERR "ERROR: parent span is not a superset "
  4612. "of domain->span\n");
  4613. } while (sd);
  4614. }
  4615. #else
  4616. # define sched_domain_debug(sd, cpu) do { } while (0)
  4617. #endif
  4618. static int sd_degenerate(struct sched_domain *sd)
  4619. {
  4620. if (cpus_weight(sd->span) == 1)
  4621. return 1;
  4622. /* Following flags need at least 2 groups */
  4623. if (sd->flags & (SD_LOAD_BALANCE |
  4624. SD_BALANCE_NEWIDLE |
  4625. SD_BALANCE_FORK |
  4626. SD_BALANCE_EXEC |
  4627. SD_SHARE_CPUPOWER |
  4628. SD_SHARE_PKG_RESOURCES)) {
  4629. if (sd->groups != sd->groups->next)
  4630. return 0;
  4631. }
  4632. /* Following flags don't use groups */
  4633. if (sd->flags & (SD_WAKE_IDLE |
  4634. SD_WAKE_AFFINE |
  4635. SD_WAKE_BALANCE))
  4636. return 0;
  4637. return 1;
  4638. }
  4639. static int
  4640. sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent)
  4641. {
  4642. unsigned long cflags = sd->flags, pflags = parent->flags;
  4643. if (sd_degenerate(parent))
  4644. return 1;
  4645. if (!cpus_equal(sd->span, parent->span))
  4646. return 0;
  4647. /* Does parent contain flags not in child? */
  4648. /* WAKE_BALANCE is a subset of WAKE_AFFINE */
  4649. if (cflags & SD_WAKE_AFFINE)
  4650. pflags &= ~SD_WAKE_BALANCE;
  4651. /* Flags needing groups don't count if only 1 group in parent */
  4652. if (parent->groups == parent->groups->next) {
  4653. pflags &= ~(SD_LOAD_BALANCE |
  4654. SD_BALANCE_NEWIDLE |
  4655. SD_BALANCE_FORK |
  4656. SD_BALANCE_EXEC |
  4657. SD_SHARE_CPUPOWER |
  4658. SD_SHARE_PKG_RESOURCES);
  4659. }
  4660. if (~cflags & pflags)
  4661. return 0;
  4662. return 1;
  4663. }
  4664. /*
  4665. * Attach the domain 'sd' to 'cpu' as its base domain. Callers must
  4666. * hold the hotplug lock.
  4667. */
  4668. static void cpu_attach_domain(struct sched_domain *sd, int cpu)
  4669. {
  4670. struct rq *rq = cpu_rq(cpu);
  4671. struct sched_domain *tmp;
  4672. /* Remove the sched domains which do not contribute to scheduling. */
  4673. for (tmp = sd; tmp; tmp = tmp->parent) {
  4674. struct sched_domain *parent = tmp->parent;
  4675. if (!parent)
  4676. break;
  4677. if (sd_parent_degenerate(tmp, parent)) {
  4678. tmp->parent = parent->parent;
  4679. if (parent->parent)
  4680. parent->parent->child = tmp;
  4681. }
  4682. }
  4683. if (sd && sd_degenerate(sd)) {
  4684. sd = sd->parent;
  4685. if (sd)
  4686. sd->child = NULL;
  4687. }
  4688. sched_domain_debug(sd, cpu);
  4689. rcu_assign_pointer(rq->sd, sd);
  4690. }
  4691. /* cpus with isolated domains */
  4692. static cpumask_t cpu_isolated_map = CPU_MASK_NONE;
  4693. /* Setup the mask of cpus configured for isolated domains */
  4694. static int __init isolated_cpu_setup(char *str)
  4695. {
  4696. int ints[NR_CPUS], i;
  4697. str = get_options(str, ARRAY_SIZE(ints), ints);
  4698. cpus_clear(cpu_isolated_map);
  4699. for (i = 1; i <= ints[0]; i++)
  4700. if (ints[i] < NR_CPUS)
  4701. cpu_set(ints[i], cpu_isolated_map);
  4702. return 1;
  4703. }
  4704. __setup ("isolcpus=", isolated_cpu_setup);
  4705. /*
  4706. * init_sched_build_groups takes the cpumask we wish to span, and a pointer
  4707. * to a function which identifies what group(along with sched group) a CPU
  4708. * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS
  4709. * (due to the fact that we keep track of groups covered with a cpumask_t).
  4710. *
  4711. * init_sched_build_groups will build a circular linked list of the groups
  4712. * covered by the given span, and will set each group's ->cpumask correctly,
  4713. * and ->cpu_power to 0.
  4714. */
  4715. static void
  4716. init_sched_build_groups(cpumask_t span, const cpumask_t *cpu_map,
  4717. int (*group_fn)(int cpu, const cpumask_t *cpu_map,
  4718. struct sched_group **sg))
  4719. {
  4720. struct sched_group *first = NULL, *last = NULL;
  4721. cpumask_t covered = CPU_MASK_NONE;
  4722. int i;
  4723. for_each_cpu_mask(i, span) {
  4724. struct sched_group *sg;
  4725. int group = group_fn(i, cpu_map, &sg);
  4726. int j;
  4727. if (cpu_isset(i, covered))
  4728. continue;
  4729. sg->cpumask = CPU_MASK_NONE;
  4730. sg->__cpu_power = 0;
  4731. for_each_cpu_mask(j, span) {
  4732. if (group_fn(j, cpu_map, NULL) != group)
  4733. continue;
  4734. cpu_set(j, covered);
  4735. cpu_set(j, sg->cpumask);
  4736. }
  4737. if (!first)
  4738. first = sg;
  4739. if (last)
  4740. last->next = sg;
  4741. last = sg;
  4742. }
  4743. last->next = first;
  4744. }
  4745. #define SD_NODES_PER_DOMAIN 16
  4746. #ifdef CONFIG_NUMA
  4747. /**
  4748. * find_next_best_node - find the next node to include in a sched_domain
  4749. * @node: node whose sched_domain we're building
  4750. * @used_nodes: nodes already in the sched_domain
  4751. *
  4752. * Find the next node to include in a given scheduling domain. Simply
  4753. * finds the closest node not already in the @used_nodes map.
  4754. *
  4755. * Should use nodemask_t.
  4756. */
  4757. static int find_next_best_node(int node, unsigned long *used_nodes)
  4758. {
  4759. int i, n, val, min_val, best_node = 0;
  4760. min_val = INT_MAX;
  4761. for (i = 0; i < MAX_NUMNODES; i++) {
  4762. /* Start at @node */
  4763. n = (node + i) % MAX_NUMNODES;
  4764. if (!nr_cpus_node(n))
  4765. continue;
  4766. /* Skip already used nodes */
  4767. if (test_bit(n, used_nodes))
  4768. continue;
  4769. /* Simple min distance search */
  4770. val = node_distance(node, n);
  4771. if (val < min_val) {
  4772. min_val = val;
  4773. best_node = n;
  4774. }
  4775. }
  4776. set_bit(best_node, used_nodes);
  4777. return best_node;
  4778. }
  4779. /**
  4780. * sched_domain_node_span - get a cpumask for a node's sched_domain
  4781. * @node: node whose cpumask we're constructing
  4782. * @size: number of nodes to include in this span
  4783. *
  4784. * Given a node, construct a good cpumask for its sched_domain to span. It
  4785. * should be one that prevents unnecessary balancing, but also spreads tasks
  4786. * out optimally.
  4787. */
  4788. static cpumask_t sched_domain_node_span(int node)
  4789. {
  4790. DECLARE_BITMAP(used_nodes, MAX_NUMNODES);
  4791. cpumask_t span, nodemask;
  4792. int i;
  4793. cpus_clear(span);
  4794. bitmap_zero(used_nodes, MAX_NUMNODES);
  4795. nodemask = node_to_cpumask(node);
  4796. cpus_or(span, span, nodemask);
  4797. set_bit(node, used_nodes);
  4798. for (i = 1; i < SD_NODES_PER_DOMAIN; i++) {
  4799. int next_node = find_next_best_node(node, used_nodes);
  4800. nodemask = node_to_cpumask(next_node);
  4801. cpus_or(span, span, nodemask);
  4802. }
  4803. return span;
  4804. }
  4805. #endif
  4806. int sched_smt_power_savings = 0, sched_mc_power_savings = 0;
  4807. /*
  4808. * SMT sched-domains:
  4809. */
  4810. #ifdef CONFIG_SCHED_SMT
  4811. static DEFINE_PER_CPU(struct sched_domain, cpu_domains);
  4812. static DEFINE_PER_CPU(struct sched_group, sched_group_cpus);
  4813. static int cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map,
  4814. struct sched_group **sg)
  4815. {
  4816. if (sg)
  4817. *sg = &per_cpu(sched_group_cpus, cpu);
  4818. return cpu;
  4819. }
  4820. #endif
  4821. /*
  4822. * multi-core sched-domains:
  4823. */
  4824. #ifdef CONFIG_SCHED_MC
  4825. static DEFINE_PER_CPU(struct sched_domain, core_domains);
  4826. static DEFINE_PER_CPU(struct sched_group, sched_group_core);
  4827. #endif
  4828. #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT)
  4829. static int cpu_to_core_group(int cpu, const cpumask_t *cpu_map,
  4830. struct sched_group **sg)
  4831. {
  4832. int group;
  4833. cpumask_t mask = cpu_sibling_map[cpu];
  4834. cpus_and(mask, mask, *cpu_map);
  4835. group = first_cpu(mask);
  4836. if (sg)
  4837. *sg = &per_cpu(sched_group_core, group);
  4838. return group;
  4839. }
  4840. #elif defined(CONFIG_SCHED_MC)
  4841. static int cpu_to_core_group(int cpu, const cpumask_t *cpu_map,
  4842. struct sched_group **sg)
  4843. {
  4844. if (sg)
  4845. *sg = &per_cpu(sched_group_core, cpu);
  4846. return cpu;
  4847. }
  4848. #endif
  4849. static DEFINE_PER_CPU(struct sched_domain, phys_domains);
  4850. static DEFINE_PER_CPU(struct sched_group, sched_group_phys);
  4851. static int cpu_to_phys_group(int cpu, const cpumask_t *cpu_map,
  4852. struct sched_group **sg)
  4853. {
  4854. int group;
  4855. #ifdef CONFIG_SCHED_MC
  4856. cpumask_t mask = cpu_coregroup_map(cpu);
  4857. cpus_and(mask, mask, *cpu_map);
  4858. group = first_cpu(mask);
  4859. #elif defined(CONFIG_SCHED_SMT)
  4860. cpumask_t mask = cpu_sibling_map[cpu];
  4861. cpus_and(mask, mask, *cpu_map);
  4862. group = first_cpu(mask);
  4863. #else
  4864. group = cpu;
  4865. #endif
  4866. if (sg)
  4867. *sg = &per_cpu(sched_group_phys, group);
  4868. return group;
  4869. }
  4870. #ifdef CONFIG_NUMA
  4871. /*
  4872. * The init_sched_build_groups can't handle what we want to do with node
  4873. * groups, so roll our own. Now each node has its own list of groups which
  4874. * gets dynamically allocated.
  4875. */
  4876. static DEFINE_PER_CPU(struct sched_domain, node_domains);
  4877. static struct sched_group **sched_group_nodes_bycpu[NR_CPUS];
  4878. static DEFINE_PER_CPU(struct sched_domain, allnodes_domains);
  4879. static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes);
  4880. static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map,
  4881. struct sched_group **sg)
  4882. {
  4883. cpumask_t nodemask = node_to_cpumask(cpu_to_node(cpu));
  4884. int group;
  4885. cpus_and(nodemask, nodemask, *cpu_map);
  4886. group = first_cpu(nodemask);
  4887. if (sg)
  4888. *sg = &per_cpu(sched_group_allnodes, group);
  4889. return group;
  4890. }
  4891. static void init_numa_sched_groups_power(struct sched_group *group_head)
  4892. {
  4893. struct sched_group *sg = group_head;
  4894. int j;
  4895. if (!sg)
  4896. return;
  4897. next_sg:
  4898. for_each_cpu_mask(j, sg->cpumask) {
  4899. struct sched_domain *sd;
  4900. sd = &per_cpu(phys_domains, j);
  4901. if (j != first_cpu(sd->groups->cpumask)) {
  4902. /*
  4903. * Only add "power" once for each
  4904. * physical package.
  4905. */
  4906. continue;
  4907. }
  4908. sg_inc_cpu_power(sg, sd->groups->__cpu_power);
  4909. }
  4910. sg = sg->next;
  4911. if (sg != group_head)
  4912. goto next_sg;
  4913. }
  4914. #endif
  4915. #ifdef CONFIG_NUMA
  4916. /* Free memory allocated for various sched_group structures */
  4917. static void free_sched_groups(const cpumask_t *cpu_map)
  4918. {
  4919. int cpu, i;
  4920. for_each_cpu_mask(cpu, *cpu_map) {
  4921. struct sched_group **sched_group_nodes
  4922. = sched_group_nodes_bycpu[cpu];
  4923. if (!sched_group_nodes)
  4924. continue;
  4925. for (i = 0; i < MAX_NUMNODES; i++) {
  4926. cpumask_t nodemask = node_to_cpumask(i);
  4927. struct sched_group *oldsg, *sg = sched_group_nodes[i];
  4928. cpus_and(nodemask, nodemask, *cpu_map);
  4929. if (cpus_empty(nodemask))
  4930. continue;
  4931. if (sg == NULL)
  4932. continue;
  4933. sg = sg->next;
  4934. next_sg:
  4935. oldsg = sg;
  4936. sg = sg->next;
  4937. kfree(oldsg);
  4938. if (oldsg != sched_group_nodes[i])
  4939. goto next_sg;
  4940. }
  4941. kfree(sched_group_nodes);
  4942. sched_group_nodes_bycpu[cpu] = NULL;
  4943. }
  4944. }
  4945. #else
  4946. static void free_sched_groups(const cpumask_t *cpu_map)
  4947. {
  4948. }
  4949. #endif
  4950. /*
  4951. * Initialize sched groups cpu_power.
  4952. *
  4953. * cpu_power indicates the capacity of sched group, which is used while
  4954. * distributing the load between different sched groups in a sched domain.
  4955. * Typically cpu_power for all the groups in a sched domain will be same unless
  4956. * there are asymmetries in the topology. If there are asymmetries, group
  4957. * having more cpu_power will pickup more load compared to the group having
  4958. * less cpu_power.
  4959. *
  4960. * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents
  4961. * the maximum number of tasks a group can handle in the presence of other idle
  4962. * or lightly loaded groups in the same sched domain.
  4963. */
  4964. static void init_sched_groups_power(int cpu, struct sched_domain *sd)
  4965. {
  4966. struct sched_domain *child;
  4967. struct sched_group *group;
  4968. WARN_ON(!sd || !sd->groups);
  4969. if (cpu != first_cpu(sd->groups->cpumask))
  4970. return;
  4971. child = sd->child;
  4972. sd->groups->__cpu_power = 0;
  4973. /*
  4974. * For perf policy, if the groups in child domain share resources
  4975. * (for example cores sharing some portions of the cache hierarchy
  4976. * or SMT), then set this domain groups cpu_power such that each group
  4977. * can handle only one task, when there are other idle groups in the
  4978. * same sched domain.
  4979. */
  4980. if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) &&
  4981. (child->flags &
  4982. (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) {
  4983. sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE);
  4984. return;
  4985. }
  4986. /*
  4987. * add cpu_power of each child group to this groups cpu_power
  4988. */
  4989. group = child->groups;
  4990. do {
  4991. sg_inc_cpu_power(sd->groups, group->__cpu_power);
  4992. group = group->next;
  4993. } while (group != child->groups);
  4994. }
  4995. /*
  4996. * Build sched domains for a given set of cpus and attach the sched domains
  4997. * to the individual cpus
  4998. */
  4999. static int build_sched_domains(const cpumask_t *cpu_map)
  5000. {
  5001. int i;
  5002. #ifdef CONFIG_NUMA
  5003. struct sched_group **sched_group_nodes = NULL;
  5004. int sd_allnodes = 0;
  5005. /*
  5006. * Allocate the per-node list of sched groups
  5007. */
  5008. sched_group_nodes = kzalloc(sizeof(struct sched_group *)*MAX_NUMNODES,
  5009. GFP_KERNEL);
  5010. if (!sched_group_nodes) {
  5011. printk(KERN_WARNING "Can not alloc sched group node list\n");
  5012. return -ENOMEM;
  5013. }
  5014. sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes;
  5015. #endif
  5016. /*
  5017. * Set up domains for cpus specified by the cpu_map.
  5018. */
  5019. for_each_cpu_mask(i, *cpu_map) {
  5020. struct sched_domain *sd = NULL, *p;
  5021. cpumask_t nodemask = node_to_cpumask(cpu_to_node(i));
  5022. cpus_and(nodemask, nodemask, *cpu_map);
  5023. #ifdef CONFIG_NUMA
  5024. if (cpus_weight(*cpu_map) >
  5025. SD_NODES_PER_DOMAIN*cpus_weight(nodemask)) {
  5026. sd = &per_cpu(allnodes_domains, i);
  5027. *sd = SD_ALLNODES_INIT;
  5028. sd->span = *cpu_map;
  5029. cpu_to_allnodes_group(i, cpu_map, &sd->groups);
  5030. p = sd;
  5031. sd_allnodes = 1;
  5032. } else
  5033. p = NULL;
  5034. sd = &per_cpu(node_domains, i);
  5035. *sd = SD_NODE_INIT;
  5036. sd->span = sched_domain_node_span(cpu_to_node(i));
  5037. sd->parent = p;
  5038. if (p)
  5039. p->child = sd;
  5040. cpus_and(sd->span, sd->span, *cpu_map);
  5041. #endif
  5042. p = sd;
  5043. sd = &per_cpu(phys_domains, i);
  5044. *sd = SD_CPU_INIT;
  5045. sd->span = nodemask;
  5046. sd->parent = p;
  5047. if (p)
  5048. p->child = sd;
  5049. cpu_to_phys_group(i, cpu_map, &sd->groups);
  5050. #ifdef CONFIG_SCHED_MC
  5051. p = sd;
  5052. sd = &per_cpu(core_domains, i);
  5053. *sd = SD_MC_INIT;
  5054. sd->span = cpu_coregroup_map(i);
  5055. cpus_and(sd->span, sd->span, *cpu_map);
  5056. sd->parent = p;
  5057. p->child = sd;
  5058. cpu_to_core_group(i, cpu_map, &sd->groups);
  5059. #endif
  5060. #ifdef CONFIG_SCHED_SMT
  5061. p = sd;
  5062. sd = &per_cpu(cpu_domains, i);
  5063. *sd = SD_SIBLING_INIT;
  5064. sd->span = cpu_sibling_map[i];
  5065. cpus_and(sd->span, sd->span, *cpu_map);
  5066. sd->parent = p;
  5067. p->child = sd;
  5068. cpu_to_cpu_group(i, cpu_map, &sd->groups);
  5069. #endif
  5070. }
  5071. #ifdef CONFIG_SCHED_SMT
  5072. /* Set up CPU (sibling) groups */
  5073. for_each_cpu_mask(i, *cpu_map) {
  5074. cpumask_t this_sibling_map = cpu_sibling_map[i];
  5075. cpus_and(this_sibling_map, this_sibling_map, *cpu_map);
  5076. if (i != first_cpu(this_sibling_map))
  5077. continue;
  5078. init_sched_build_groups(this_sibling_map, cpu_map,
  5079. &cpu_to_cpu_group);
  5080. }
  5081. #endif
  5082. #ifdef CONFIG_SCHED_MC
  5083. /* Set up multi-core groups */
  5084. for_each_cpu_mask(i, *cpu_map) {
  5085. cpumask_t this_core_map = cpu_coregroup_map(i);
  5086. cpus_and(this_core_map, this_core_map, *cpu_map);
  5087. if (i != first_cpu(this_core_map))
  5088. continue;
  5089. init_sched_build_groups(this_core_map, cpu_map,
  5090. &cpu_to_core_group);
  5091. }
  5092. #endif
  5093. /* Set up physical groups */
  5094. for (i = 0; i < MAX_NUMNODES; i++) {
  5095. cpumask_t nodemask = node_to_cpumask(i);
  5096. cpus_and(nodemask, nodemask, *cpu_map);
  5097. if (cpus_empty(nodemask))
  5098. continue;
  5099. init_sched_build_groups(nodemask, cpu_map, &cpu_to_phys_group);
  5100. }
  5101. #ifdef CONFIG_NUMA
  5102. /* Set up node groups */
  5103. if (sd_allnodes)
  5104. init_sched_build_groups(*cpu_map, cpu_map,
  5105. &cpu_to_allnodes_group);
  5106. for (i = 0; i < MAX_NUMNODES; i++) {
  5107. /* Set up node groups */
  5108. struct sched_group *sg, *prev;
  5109. cpumask_t nodemask = node_to_cpumask(i);
  5110. cpumask_t domainspan;
  5111. cpumask_t covered = CPU_MASK_NONE;
  5112. int j;
  5113. cpus_and(nodemask, nodemask, *cpu_map);
  5114. if (cpus_empty(nodemask)) {
  5115. sched_group_nodes[i] = NULL;
  5116. continue;
  5117. }
  5118. domainspan = sched_domain_node_span(i);
  5119. cpus_and(domainspan, domainspan, *cpu_map);
  5120. sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i);
  5121. if (!sg) {
  5122. printk(KERN_WARNING "Can not alloc domain group for "
  5123. "node %d\n", i);
  5124. goto error;
  5125. }
  5126. sched_group_nodes[i] = sg;
  5127. for_each_cpu_mask(j, nodemask) {
  5128. struct sched_domain *sd;
  5129. sd = &per_cpu(node_domains, j);
  5130. sd->groups = sg;
  5131. }
  5132. sg->__cpu_power = 0;
  5133. sg->cpumask = nodemask;
  5134. sg->next = sg;
  5135. cpus_or(covered, covered, nodemask);
  5136. prev = sg;
  5137. for (j = 0; j < MAX_NUMNODES; j++) {
  5138. cpumask_t tmp, notcovered;
  5139. int n = (i + j) % MAX_NUMNODES;
  5140. cpus_complement(notcovered, covered);
  5141. cpus_and(tmp, notcovered, *cpu_map);
  5142. cpus_and(tmp, tmp, domainspan);
  5143. if (cpus_empty(tmp))
  5144. break;
  5145. nodemask = node_to_cpumask(n);
  5146. cpus_and(tmp, tmp, nodemask);
  5147. if (cpus_empty(tmp))
  5148. continue;
  5149. sg = kmalloc_node(sizeof(struct sched_group),
  5150. GFP_KERNEL, i);
  5151. if (!sg) {
  5152. printk(KERN_WARNING
  5153. "Can not alloc domain group for node %d\n", j);
  5154. goto error;
  5155. }
  5156. sg->__cpu_power = 0;
  5157. sg->cpumask = tmp;
  5158. sg->next = prev->next;
  5159. cpus_or(covered, covered, tmp);
  5160. prev->next = sg;
  5161. prev = sg;
  5162. }
  5163. }
  5164. #endif
  5165. /* Calculate CPU power for physical packages and nodes */
  5166. #ifdef CONFIG_SCHED_SMT
  5167. for_each_cpu_mask(i, *cpu_map) {
  5168. struct sched_domain *sd = &per_cpu(cpu_domains, i);
  5169. init_sched_groups_power(i, sd);
  5170. }
  5171. #endif
  5172. #ifdef CONFIG_SCHED_MC
  5173. for_each_cpu_mask(i, *cpu_map) {
  5174. struct sched_domain *sd = &per_cpu(core_domains, i);
  5175. init_sched_groups_power(i, sd);
  5176. }
  5177. #endif
  5178. for_each_cpu_mask(i, *cpu_map) {
  5179. struct sched_domain *sd = &per_cpu(phys_domains, i);
  5180. init_sched_groups_power(i, sd);
  5181. }
  5182. #ifdef CONFIG_NUMA
  5183. for (i = 0; i < MAX_NUMNODES; i++)
  5184. init_numa_sched_groups_power(sched_group_nodes[i]);
  5185. if (sd_allnodes) {
  5186. struct sched_group *sg;
  5187. cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg);
  5188. init_numa_sched_groups_power(sg);
  5189. }
  5190. #endif
  5191. /* Attach the domains */
  5192. for_each_cpu_mask(i, *cpu_map) {
  5193. struct sched_domain *sd;
  5194. #ifdef CONFIG_SCHED_SMT
  5195. sd = &per_cpu(cpu_domains, i);
  5196. #elif defined(CONFIG_SCHED_MC)
  5197. sd = &per_cpu(core_domains, i);
  5198. #else
  5199. sd = &per_cpu(phys_domains, i);
  5200. #endif
  5201. cpu_attach_domain(sd, i);
  5202. }
  5203. return 0;
  5204. #ifdef CONFIG_NUMA
  5205. error:
  5206. free_sched_groups(cpu_map);
  5207. return -ENOMEM;
  5208. #endif
  5209. }
  5210. /*
  5211. * Set up scheduler domains and groups. Callers must hold the hotplug lock.
  5212. */
  5213. static int arch_init_sched_domains(const cpumask_t *cpu_map)
  5214. {
  5215. cpumask_t cpu_default_map;
  5216. int err;
  5217. /*
  5218. * Setup mask for cpus without special case scheduling requirements.
  5219. * For now this just excludes isolated cpus, but could be used to
  5220. * exclude other special cases in the future.
  5221. */
  5222. cpus_andnot(cpu_default_map, *cpu_map, cpu_isolated_map);
  5223. err = build_sched_domains(&cpu_default_map);
  5224. return err;
  5225. }
  5226. static void arch_destroy_sched_domains(const cpumask_t *cpu_map)
  5227. {
  5228. free_sched_groups(cpu_map);
  5229. }
  5230. /*
  5231. * Detach sched domains from a group of cpus specified in cpu_map
  5232. * These cpus will now be attached to the NULL domain
  5233. */
  5234. static void detach_destroy_domains(const cpumask_t *cpu_map)
  5235. {
  5236. int i;
  5237. for_each_cpu_mask(i, *cpu_map)
  5238. cpu_attach_domain(NULL, i);
  5239. synchronize_sched();
  5240. arch_destroy_sched_domains(cpu_map);
  5241. }
  5242. /*
  5243. * Partition sched domains as specified by the cpumasks below.
  5244. * This attaches all cpus from the cpumasks to the NULL domain,
  5245. * waits for a RCU quiescent period, recalculates sched
  5246. * domain information and then attaches them back to the
  5247. * correct sched domains
  5248. * Call with hotplug lock held
  5249. */
  5250. int partition_sched_domains(cpumask_t *partition1, cpumask_t *partition2)
  5251. {
  5252. cpumask_t change_map;
  5253. int err = 0;
  5254. cpus_and(*partition1, *partition1, cpu_online_map);
  5255. cpus_and(*partition2, *partition2, cpu_online_map);
  5256. cpus_or(change_map, *partition1, *partition2);
  5257. /* Detach sched domains from all of the affected cpus */
  5258. detach_destroy_domains(&change_map);
  5259. if (!cpus_empty(*partition1))
  5260. err = build_sched_domains(partition1);
  5261. if (!err && !cpus_empty(*partition2))
  5262. err = build_sched_domains(partition2);
  5263. return err;
  5264. }
  5265. #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT)
  5266. int arch_reinit_sched_domains(void)
  5267. {
  5268. int err;
  5269. mutex_lock(&sched_hotcpu_mutex);
  5270. detach_destroy_domains(&cpu_online_map);
  5271. err = arch_init_sched_domains(&cpu_online_map);
  5272. mutex_unlock(&sched_hotcpu_mutex);
  5273. return err;
  5274. }
  5275. static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt)
  5276. {
  5277. int ret;
  5278. if (buf[0] != '0' && buf[0] != '1')
  5279. return -EINVAL;
  5280. if (smt)
  5281. sched_smt_power_savings = (buf[0] == '1');
  5282. else
  5283. sched_mc_power_savings = (buf[0] == '1');
  5284. ret = arch_reinit_sched_domains();
  5285. return ret ? ret : count;
  5286. }
  5287. int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls)
  5288. {
  5289. int err = 0;
  5290. #ifdef CONFIG_SCHED_SMT
  5291. if (smt_capable())
  5292. err = sysfs_create_file(&cls->kset.kobj,
  5293. &attr_sched_smt_power_savings.attr);
  5294. #endif
  5295. #ifdef CONFIG_SCHED_MC
  5296. if (!err && mc_capable())
  5297. err = sysfs_create_file(&cls->kset.kobj,
  5298. &attr_sched_mc_power_savings.attr);
  5299. #endif
  5300. return err;
  5301. }
  5302. #endif
  5303. #ifdef CONFIG_SCHED_MC
  5304. static ssize_t sched_mc_power_savings_show(struct sys_device *dev, char *page)
  5305. {
  5306. return sprintf(page, "%u\n", sched_mc_power_savings);
  5307. }
  5308. static ssize_t sched_mc_power_savings_store(struct sys_device *dev,
  5309. const char *buf, size_t count)
  5310. {
  5311. return sched_power_savings_store(buf, count, 0);
  5312. }
  5313. SYSDEV_ATTR(sched_mc_power_savings, 0644, sched_mc_power_savings_show,
  5314. sched_mc_power_savings_store);
  5315. #endif
  5316. #ifdef CONFIG_SCHED_SMT
  5317. static ssize_t sched_smt_power_savings_show(struct sys_device *dev, char *page)
  5318. {
  5319. return sprintf(page, "%u\n", sched_smt_power_savings);
  5320. }
  5321. static ssize_t sched_smt_power_savings_store(struct sys_device *dev,
  5322. const char *buf, size_t count)
  5323. {
  5324. return sched_power_savings_store(buf, count, 1);
  5325. }
  5326. SYSDEV_ATTR(sched_smt_power_savings, 0644, sched_smt_power_savings_show,
  5327. sched_smt_power_savings_store);
  5328. #endif
  5329. /*
  5330. * Force a reinitialization of the sched domains hierarchy. The domains
  5331. * and groups cannot be updated in place without racing with the balancing
  5332. * code, so we temporarily attach all running cpus to the NULL domain
  5333. * which will prevent rebalancing while the sched domains are recalculated.
  5334. */
  5335. static int update_sched_domains(struct notifier_block *nfb,
  5336. unsigned long action, void *hcpu)
  5337. {
  5338. switch (action) {
  5339. case CPU_UP_PREPARE:
  5340. case CPU_UP_PREPARE_FROZEN:
  5341. case CPU_DOWN_PREPARE:
  5342. case CPU_DOWN_PREPARE_FROZEN:
  5343. detach_destroy_domains(&cpu_online_map);
  5344. return NOTIFY_OK;
  5345. case CPU_UP_CANCELED:
  5346. case CPU_UP_CANCELED_FROZEN:
  5347. case CPU_DOWN_FAILED:
  5348. case CPU_DOWN_FAILED_FROZEN:
  5349. case CPU_ONLINE:
  5350. case CPU_ONLINE_FROZEN:
  5351. case CPU_DEAD:
  5352. case CPU_DEAD_FROZEN:
  5353. /*
  5354. * Fall through and re-initialise the domains.
  5355. */
  5356. break;
  5357. default:
  5358. return NOTIFY_DONE;
  5359. }
  5360. /* The hotplug lock is already held by cpu_up/cpu_down */
  5361. arch_init_sched_domains(&cpu_online_map);
  5362. return NOTIFY_OK;
  5363. }
  5364. void __init sched_init_smp(void)
  5365. {
  5366. cpumask_t non_isolated_cpus;
  5367. mutex_lock(&sched_hotcpu_mutex);
  5368. arch_init_sched_domains(&cpu_online_map);
  5369. cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map);
  5370. if (cpus_empty(non_isolated_cpus))
  5371. cpu_set(smp_processor_id(), non_isolated_cpus);
  5372. mutex_unlock(&sched_hotcpu_mutex);
  5373. /* XXX: Theoretical race here - CPU may be hotplugged now */
  5374. hotcpu_notifier(update_sched_domains, 0);
  5375. /* Move init over to a non-isolated CPU */
  5376. if (set_cpus_allowed(current, non_isolated_cpus) < 0)
  5377. BUG();
  5378. sched_init_granularity();
  5379. }
  5380. #else
  5381. void __init sched_init_smp(void)
  5382. {
  5383. sched_init_granularity();
  5384. }
  5385. #endif /* CONFIG_SMP */
  5386. int in_sched_functions(unsigned long addr)
  5387. {
  5388. /* Linker adds these: start and end of __sched functions */
  5389. extern char __sched_text_start[], __sched_text_end[];
  5390. return in_lock_functions(addr) ||
  5391. (addr >= (unsigned long)__sched_text_start
  5392. && addr < (unsigned long)__sched_text_end);
  5393. }
  5394. static inline void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq)
  5395. {
  5396. cfs_rq->tasks_timeline = RB_ROOT;
  5397. cfs_rq->fair_clock = 1;
  5398. #ifdef CONFIG_FAIR_GROUP_SCHED
  5399. cfs_rq->rq = rq;
  5400. #endif
  5401. }
  5402. void __init sched_init(void)
  5403. {
  5404. u64 now = sched_clock();
  5405. int highest_cpu = 0;
  5406. int i, j;
  5407. /*
  5408. * Link up the scheduling class hierarchy:
  5409. */
  5410. rt_sched_class.next = &fair_sched_class;
  5411. fair_sched_class.next = &idle_sched_class;
  5412. idle_sched_class.next = NULL;
  5413. for_each_possible_cpu(i) {
  5414. struct rt_prio_array *array;
  5415. struct rq *rq;
  5416. rq = cpu_rq(i);
  5417. spin_lock_init(&rq->lock);
  5418. lockdep_set_class(&rq->lock, &rq->rq_lock_key);
  5419. rq->nr_running = 0;
  5420. rq->clock = 1;
  5421. init_cfs_rq(&rq->cfs, rq);
  5422. #ifdef CONFIG_FAIR_GROUP_SCHED
  5423. INIT_LIST_HEAD(&rq->leaf_cfs_rq_list);
  5424. list_add(&rq->cfs.leaf_cfs_rq_list, &rq->leaf_cfs_rq_list);
  5425. #endif
  5426. rq->ls.load_update_last = now;
  5427. rq->ls.load_update_start = now;
  5428. for (j = 0; j < CPU_LOAD_IDX_MAX; j++)
  5429. rq->cpu_load[j] = 0;
  5430. #ifdef CONFIG_SMP
  5431. rq->sd = NULL;
  5432. rq->active_balance = 0;
  5433. rq->next_balance = jiffies;
  5434. rq->push_cpu = 0;
  5435. rq->cpu = i;
  5436. rq->migration_thread = NULL;
  5437. INIT_LIST_HEAD(&rq->migration_queue);
  5438. #endif
  5439. atomic_set(&rq->nr_iowait, 0);
  5440. array = &rq->rt.active;
  5441. for (j = 0; j < MAX_RT_PRIO; j++) {
  5442. INIT_LIST_HEAD(array->queue + j);
  5443. __clear_bit(j, array->bitmap);
  5444. }
  5445. highest_cpu = i;
  5446. /* delimiter for bitsearch: */
  5447. __set_bit(MAX_RT_PRIO, array->bitmap);
  5448. }
  5449. set_load_weight(&init_task);
  5450. #ifdef CONFIG_SMP
  5451. nr_cpu_ids = highest_cpu + 1;
  5452. open_softirq(SCHED_SOFTIRQ, run_rebalance_domains, NULL);
  5453. #endif
  5454. #ifdef CONFIG_RT_MUTEXES
  5455. plist_head_init(&init_task.pi_waiters, &init_task.pi_lock);
  5456. #endif
  5457. /*
  5458. * The boot idle thread does lazy MMU switching as well:
  5459. */
  5460. atomic_inc(&init_mm.mm_count);
  5461. enter_lazy_tlb(&init_mm, current);
  5462. /*
  5463. * Make us the idle thread. Technically, schedule() should not be
  5464. * called from this thread, however somewhere below it might be,
  5465. * but because we are the idle thread, we just pick up running again
  5466. * when this runqueue becomes "idle".
  5467. */
  5468. init_idle(current, smp_processor_id());
  5469. /*
  5470. * During early bootup we pretend to be a normal task:
  5471. */
  5472. current->sched_class = &fair_sched_class;
  5473. }
  5474. #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP
  5475. void __might_sleep(char *file, int line)
  5476. {
  5477. #ifdef in_atomic
  5478. static unsigned long prev_jiffy; /* ratelimiting */
  5479. if ((in_atomic() || irqs_disabled()) &&
  5480. system_state == SYSTEM_RUNNING && !oops_in_progress) {
  5481. if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy)
  5482. return;
  5483. prev_jiffy = jiffies;
  5484. printk(KERN_ERR "BUG: sleeping function called from invalid"
  5485. " context at %s:%d\n", file, line);
  5486. printk("in_atomic():%d, irqs_disabled():%d\n",
  5487. in_atomic(), irqs_disabled());
  5488. debug_show_held_locks(current);
  5489. if (irqs_disabled())
  5490. print_irqtrace_events(current);
  5491. dump_stack();
  5492. }
  5493. #endif
  5494. }
  5495. EXPORT_SYMBOL(__might_sleep);
  5496. #endif
  5497. #ifdef CONFIG_MAGIC_SYSRQ
  5498. void normalize_rt_tasks(void)
  5499. {
  5500. struct task_struct *g, *p;
  5501. unsigned long flags;
  5502. struct rq *rq;
  5503. int on_rq;
  5504. read_lock_irq(&tasklist_lock);
  5505. do_each_thread(g, p) {
  5506. p->se.fair_key = 0;
  5507. p->se.wait_runtime = 0;
  5508. p->se.wait_start_fair = 0;
  5509. p->se.wait_start = 0;
  5510. p->se.exec_start = 0;
  5511. p->se.sleep_start = 0;
  5512. p->se.sleep_start_fair = 0;
  5513. p->se.block_start = 0;
  5514. task_rq(p)->cfs.fair_clock = 0;
  5515. task_rq(p)->clock = 0;
  5516. if (!rt_task(p)) {
  5517. /*
  5518. * Renice negative nice level userspace
  5519. * tasks back to 0:
  5520. */
  5521. if (TASK_NICE(p) < 0 && p->mm)
  5522. set_user_nice(p, 0);
  5523. continue;
  5524. }
  5525. spin_lock_irqsave(&p->pi_lock, flags);
  5526. rq = __task_rq_lock(p);
  5527. #ifdef CONFIG_SMP
  5528. /*
  5529. * Do not touch the migration thread:
  5530. */
  5531. if (p == rq->migration_thread)
  5532. goto out_unlock;
  5533. #endif
  5534. on_rq = p->se.on_rq;
  5535. if (on_rq)
  5536. deactivate_task(task_rq(p), p, 0);
  5537. __setscheduler(rq, p, SCHED_NORMAL, 0);
  5538. if (on_rq) {
  5539. activate_task(task_rq(p), p, 0);
  5540. resched_task(rq->curr);
  5541. }
  5542. #ifdef CONFIG_SMP
  5543. out_unlock:
  5544. #endif
  5545. __task_rq_unlock(rq);
  5546. spin_unlock_irqrestore(&p->pi_lock, flags);
  5547. } while_each_thread(g, p);
  5548. read_unlock_irq(&tasklist_lock);
  5549. }
  5550. #endif /* CONFIG_MAGIC_SYSRQ */
  5551. #ifdef CONFIG_IA64
  5552. /*
  5553. * These functions are only useful for the IA64 MCA handling.
  5554. *
  5555. * They can only be called when the whole system has been
  5556. * stopped - every CPU needs to be quiescent, and no scheduling
  5557. * activity can take place. Using them for anything else would
  5558. * be a serious bug, and as a result, they aren't even visible
  5559. * under any other configuration.
  5560. */
  5561. /**
  5562. * curr_task - return the current task for a given cpu.
  5563. * @cpu: the processor in question.
  5564. *
  5565. * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED!
  5566. */
  5567. struct task_struct *curr_task(int cpu)
  5568. {
  5569. return cpu_curr(cpu);
  5570. }
  5571. /**
  5572. * set_curr_task - set the current task for a given cpu.
  5573. * @cpu: the processor in question.
  5574. * @p: the task pointer to set.
  5575. *
  5576. * Description: This function must only be used when non-maskable interrupts
  5577. * are serviced on a separate stack. It allows the architecture to switch the
  5578. * notion of the current task on a cpu in a non-blocking manner. This function
  5579. * must be called with all CPU's synchronized, and interrupts disabled, the
  5580. * and caller must save the original value of the current task (see
  5581. * curr_task() above) and restore that value before reenabling interrupts and
  5582. * re-starting the system.
  5583. *
  5584. * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED!
  5585. */
  5586. void set_curr_task(int cpu, struct task_struct *p)
  5587. {
  5588. cpu_curr(cpu) = p;
  5589. }
  5590. #endif