slub.c 130 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489149014911492149314941495149614971498149915001501150215031504150515061507150815091510151115121513151415151516151715181519152015211522152315241525152615271528152915301531153215331534153515361537153815391540154115421543154415451546154715481549155015511552155315541555155615571558155915601561156215631564156515661567156815691570157115721573157415751576157715781579158015811582158315841585158615871588158915901591159215931594159515961597159815991600160116021603160416051606160716081609161016111612161316141615161616171618161916201621162216231624162516261627162816291630163116321633163416351636163716381639164016411642164316441645164616471648164916501651165216531654165516561657165816591660166116621663166416651666166716681669167016711672167316741675167616771678167916801681168216831684168516861687168816891690169116921693169416951696169716981699170017011702170317041705170617071708170917101711171217131714171517161717171817191720172117221723172417251726172717281729173017311732173317341735173617371738173917401741174217431744174517461747174817491750175117521753175417551756175717581759176017611762176317641765176617671768176917701771177217731774177517761777177817791780178117821783178417851786178717881789179017911792179317941795179617971798179918001801180218031804180518061807180818091810181118121813181418151816181718181819182018211822182318241825182618271828182918301831183218331834183518361837183818391840184118421843184418451846184718481849185018511852185318541855185618571858185918601861186218631864186518661867186818691870187118721873187418751876187718781879188018811882188318841885188618871888188918901891189218931894189518961897189818991900190119021903190419051906190719081909191019111912191319141915191619171918191919201921192219231924192519261927192819291930193119321933193419351936193719381939194019411942194319441945194619471948194919501951195219531954195519561957195819591960196119621963196419651966196719681969197019711972197319741975197619771978197919801981198219831984198519861987198819891990199119921993199419951996199719981999200020012002200320042005200620072008200920102011201220132014201520162017201820192020202120222023202420252026202720282029203020312032203320342035203620372038203920402041204220432044204520462047204820492050205120522053205420552056205720582059206020612062206320642065206620672068206920702071207220732074207520762077207820792080208120822083208420852086208720882089209020912092209320942095209620972098209921002101210221032104210521062107210821092110211121122113211421152116211721182119212021212122212321242125212621272128212921302131213221332134213521362137213821392140214121422143214421452146214721482149215021512152215321542155215621572158215921602161216221632164216521662167216821692170217121722173217421752176217721782179218021812182218321842185218621872188218921902191219221932194219521962197219821992200220122022203220422052206220722082209221022112212221322142215221622172218221922202221222222232224222522262227222822292230223122322233223422352236223722382239224022412242224322442245224622472248224922502251225222532254225522562257225822592260226122622263226422652266226722682269227022712272227322742275227622772278227922802281228222832284228522862287228822892290229122922293229422952296229722982299230023012302230323042305230623072308230923102311231223132314231523162317231823192320232123222323232423252326232723282329233023312332233323342335233623372338233923402341234223432344234523462347234823492350235123522353235423552356235723582359236023612362236323642365236623672368236923702371237223732374237523762377237823792380238123822383238423852386238723882389239023912392239323942395239623972398239924002401240224032404240524062407240824092410241124122413241424152416241724182419242024212422242324242425242624272428242924302431243224332434243524362437243824392440244124422443244424452446244724482449245024512452245324542455245624572458245924602461246224632464246524662467246824692470247124722473247424752476247724782479248024812482248324842485248624872488248924902491249224932494249524962497249824992500250125022503250425052506250725082509251025112512251325142515251625172518251925202521252225232524252525262527252825292530253125322533253425352536253725382539254025412542254325442545254625472548254925502551255225532554255525562557255825592560256125622563256425652566256725682569257025712572257325742575257625772578257925802581258225832584258525862587258825892590259125922593259425952596259725982599260026012602260326042605260626072608260926102611261226132614261526162617261826192620262126222623262426252626262726282629263026312632263326342635263626372638263926402641264226432644264526462647264826492650265126522653265426552656265726582659266026612662266326642665266626672668266926702671267226732674267526762677267826792680268126822683268426852686268726882689269026912692269326942695269626972698269927002701270227032704270527062707270827092710271127122713271427152716271727182719272027212722272327242725272627272728272927302731273227332734273527362737273827392740274127422743274427452746274727482749275027512752275327542755275627572758275927602761276227632764276527662767276827692770277127722773277427752776277727782779278027812782278327842785278627872788278927902791279227932794279527962797279827992800280128022803280428052806280728082809281028112812281328142815281628172818281928202821282228232824282528262827282828292830283128322833283428352836283728382839284028412842284328442845284628472848284928502851285228532854285528562857285828592860286128622863286428652866286728682869287028712872287328742875287628772878287928802881288228832884288528862887288828892890289128922893289428952896289728982899290029012902290329042905290629072908290929102911291229132914291529162917291829192920292129222923292429252926292729282929293029312932293329342935293629372938293929402941294229432944294529462947294829492950295129522953295429552956295729582959296029612962296329642965296629672968296929702971297229732974297529762977297829792980298129822983298429852986298729882989299029912992299329942995299629972998299930003001300230033004300530063007300830093010301130123013301430153016301730183019302030213022302330243025302630273028302930303031303230333034303530363037303830393040304130423043304430453046304730483049305030513052305330543055305630573058305930603061306230633064306530663067306830693070307130723073307430753076307730783079308030813082308330843085308630873088308930903091309230933094309530963097309830993100310131023103310431053106310731083109311031113112311331143115311631173118311931203121312231233124312531263127312831293130313131323133313431353136313731383139314031413142314331443145314631473148314931503151315231533154315531563157315831593160316131623163316431653166316731683169317031713172317331743175317631773178317931803181318231833184318531863187318831893190319131923193319431953196319731983199320032013202320332043205320632073208320932103211321232133214321532163217321832193220322132223223322432253226322732283229323032313232323332343235323632373238323932403241324232433244324532463247324832493250325132523253325432553256325732583259326032613262326332643265326632673268326932703271327232733274327532763277327832793280328132823283328432853286328732883289329032913292329332943295329632973298329933003301330233033304330533063307330833093310331133123313331433153316331733183319332033213322332333243325332633273328332933303331333233333334333533363337333833393340334133423343334433453346334733483349335033513352335333543355335633573358335933603361336233633364336533663367336833693370337133723373337433753376337733783379338033813382338333843385338633873388338933903391339233933394339533963397339833993400340134023403340434053406340734083409341034113412341334143415341634173418341934203421342234233424342534263427342834293430343134323433343434353436343734383439344034413442344334443445344634473448344934503451345234533454345534563457345834593460346134623463346434653466346734683469347034713472347334743475347634773478347934803481348234833484348534863487348834893490349134923493349434953496349734983499350035013502350335043505350635073508350935103511351235133514351535163517351835193520352135223523352435253526352735283529353035313532353335343535353635373538353935403541354235433544354535463547354835493550355135523553355435553556355735583559356035613562356335643565356635673568356935703571357235733574357535763577357835793580358135823583358435853586358735883589359035913592359335943595359635973598359936003601360236033604360536063607360836093610361136123613361436153616361736183619362036213622362336243625362636273628362936303631363236333634363536363637363836393640364136423643364436453646364736483649365036513652365336543655365636573658365936603661366236633664366536663667366836693670367136723673367436753676367736783679368036813682368336843685368636873688368936903691369236933694369536963697369836993700370137023703370437053706370737083709371037113712371337143715371637173718371937203721372237233724372537263727372837293730373137323733373437353736373737383739374037413742374337443745374637473748374937503751375237533754375537563757375837593760376137623763376437653766376737683769377037713772377337743775377637773778377937803781378237833784378537863787378837893790379137923793379437953796379737983799380038013802380338043805380638073808380938103811381238133814381538163817381838193820382138223823382438253826382738283829383038313832383338343835383638373838383938403841384238433844384538463847384838493850385138523853385438553856385738583859386038613862386338643865386638673868386938703871387238733874387538763877387838793880388138823883388438853886388738883889389038913892389338943895389638973898389939003901390239033904390539063907390839093910391139123913391439153916391739183919392039213922392339243925392639273928392939303931393239333934393539363937393839393940394139423943394439453946394739483949395039513952395339543955395639573958395939603961396239633964396539663967396839693970397139723973397439753976397739783979398039813982398339843985398639873988398939903991399239933994399539963997399839994000400140024003400440054006400740084009401040114012401340144015401640174018401940204021402240234024402540264027402840294030403140324033403440354036403740384039404040414042404340444045404640474048404940504051405240534054405540564057405840594060406140624063406440654066406740684069407040714072407340744075407640774078407940804081408240834084408540864087408840894090409140924093409440954096409740984099410041014102410341044105410641074108410941104111411241134114411541164117411841194120412141224123412441254126412741284129413041314132413341344135413641374138413941404141414241434144414541464147414841494150415141524153415441554156415741584159416041614162416341644165416641674168416941704171417241734174417541764177417841794180418141824183418441854186418741884189419041914192419341944195419641974198419942004201420242034204420542064207420842094210421142124213421442154216421742184219422042214222422342244225422642274228422942304231423242334234423542364237423842394240424142424243424442454246424742484249425042514252425342544255425642574258425942604261426242634264426542664267426842694270427142724273427442754276427742784279428042814282428342844285428642874288428942904291429242934294429542964297429842994300430143024303430443054306430743084309431043114312431343144315431643174318431943204321432243234324432543264327432843294330433143324333433443354336433743384339434043414342434343444345434643474348434943504351435243534354435543564357435843594360436143624363436443654366436743684369437043714372437343744375437643774378437943804381438243834384438543864387438843894390439143924393439443954396439743984399440044014402440344044405440644074408440944104411441244134414441544164417441844194420442144224423442444254426442744284429443044314432443344344435443644374438443944404441444244434444444544464447444844494450445144524453445444554456445744584459446044614462446344644465446644674468446944704471447244734474447544764477447844794480448144824483448444854486448744884489449044914492449344944495449644974498449945004501450245034504450545064507450845094510451145124513451445154516451745184519452045214522452345244525452645274528452945304531453245334534453545364537453845394540454145424543454445454546454745484549455045514552455345544555455645574558455945604561456245634564456545664567456845694570457145724573457445754576457745784579458045814582458345844585458645874588458945904591459245934594459545964597459845994600460146024603460446054606460746084609461046114612461346144615461646174618461946204621462246234624462546264627462846294630463146324633463446354636463746384639464046414642464346444645464646474648464946504651465246534654465546564657465846594660466146624663466446654666466746684669467046714672467346744675467646774678467946804681468246834684468546864687468846894690469146924693469446954696469746984699470047014702470347044705470647074708470947104711471247134714471547164717471847194720472147224723472447254726472747284729473047314732473347344735473647374738473947404741474247434744474547464747474847494750475147524753475447554756475747584759476047614762476347644765476647674768476947704771477247734774477547764777477847794780478147824783478447854786478747884789479047914792479347944795479647974798479948004801480248034804480548064807480848094810481148124813481448154816481748184819482048214822482348244825482648274828482948304831483248334834483548364837483848394840484148424843484448454846484748484849485048514852485348544855485648574858485948604861486248634864486548664867486848694870487148724873487448754876487748784879488048814882488348844885488648874888488948904891489248934894489548964897489848994900490149024903490449054906490749084909491049114912491349144915491649174918491949204921492249234924492549264927492849294930493149324933493449354936493749384939494049414942494349444945494649474948494949504951495249534954495549564957495849594960496149624963496449654966496749684969497049714972497349744975497649774978497949804981498249834984498549864987498849894990499149924993499449954996499749984999500050015002500350045005500650075008500950105011501250135014501550165017501850195020502150225023502450255026502750285029503050315032503350345035503650375038503950405041504250435044504550465047504850495050505150525053505450555056505750585059506050615062506350645065506650675068506950705071507250735074507550765077507850795080508150825083508450855086508750885089509050915092509350945095509650975098509951005101510251035104510551065107510851095110511151125113511451155116511751185119512051215122512351245125512651275128512951305131513251335134513551365137513851395140514151425143514451455146514751485149515051515152515351545155515651575158515951605161516251635164516551665167516851695170517151725173517451755176517751785179518051815182518351845185518651875188518951905191519251935194519551965197519851995200520152025203520452055206520752085209521052115212521352145215521652175218521952205221522252235224522552265227522852295230523152325233523452355236523752385239524052415242524352445245524652475248524952505251525252535254525552565257525852595260526152625263526452655266526752685269527052715272527352745275527652775278527952805281528252835284528552865287528852895290529152925293529452955296529752985299530053015302530353045305530653075308530953105311531253135314531553165317531853195320532153225323532453255326532753285329533053315332533353345335533653375338533953405341534253435344534553465347534853495350535153525353535453555356535753585359536053615362536353645365536653675368536953705371537253735374537553765377537853795380538153825383538453855386538753885389539053915392539353945395539653975398539954005401540254035404540554065407540854095410541154125413541454155416541754185419542054215422542354245425542654275428542954305431543254335434543554365437543854395440544154425443544454455446544754485449545054515452545354545455545654575458545954605461546254635464546554665467546854695470547154725473547454755476547754785479548054815482548354845485548654875488548954905491549254935494549554965497549854995500550155025503550455055506550755085509551055115512551355145515551655175518551955205521552255235524552555265527552855295530553155325533553455355536
  1. /*
  2. * SLUB: A slab allocator that limits cache line use instead of queuing
  3. * objects in per cpu and per node lists.
  4. *
  5. * The allocator synchronizes using per slab locks or atomic operatios
  6. * and only uses a centralized lock to manage a pool of partial slabs.
  7. *
  8. * (C) 2007 SGI, Christoph Lameter
  9. * (C) 2011 Linux Foundation, Christoph Lameter
  10. */
  11. #include <linux/mm.h>
  12. #include <linux/swap.h> /* struct reclaim_state */
  13. #include <linux/module.h>
  14. #include <linux/bit_spinlock.h>
  15. #include <linux/interrupt.h>
  16. #include <linux/bitops.h>
  17. #include <linux/slab.h>
  18. #include <linux/proc_fs.h>
  19. #include <linux/seq_file.h>
  20. #include <linux/kmemcheck.h>
  21. #include <linux/cpu.h>
  22. #include <linux/cpuset.h>
  23. #include <linux/mempolicy.h>
  24. #include <linux/ctype.h>
  25. #include <linux/debugobjects.h>
  26. #include <linux/kallsyms.h>
  27. #include <linux/memory.h>
  28. #include <linux/math64.h>
  29. #include <linux/fault-inject.h>
  30. #include <linux/stacktrace.h>
  31. #include <linux/prefetch.h>
  32. #include <trace/events/kmem.h>
  33. /*
  34. * Lock order:
  35. * 1. slub_lock (Global Semaphore)
  36. * 2. node->list_lock
  37. * 3. slab_lock(page) (Only on some arches and for debugging)
  38. *
  39. * slub_lock
  40. *
  41. * The role of the slub_lock is to protect the list of all the slabs
  42. * and to synchronize major metadata changes to slab cache structures.
  43. *
  44. * The slab_lock is only used for debugging and on arches that do not
  45. * have the ability to do a cmpxchg_double. It only protects the second
  46. * double word in the page struct. Meaning
  47. * A. page->freelist -> List of object free in a page
  48. * B. page->counters -> Counters of objects
  49. * C. page->frozen -> frozen state
  50. *
  51. * If a slab is frozen then it is exempt from list management. It is not
  52. * on any list. The processor that froze the slab is the one who can
  53. * perform list operations on the page. Other processors may put objects
  54. * onto the freelist but the processor that froze the slab is the only
  55. * one that can retrieve the objects from the page's freelist.
  56. *
  57. * The list_lock protects the partial and full list on each node and
  58. * the partial slab counter. If taken then no new slabs may be added or
  59. * removed from the lists nor make the number of partial slabs be modified.
  60. * (Note that the total number of slabs is an atomic value that may be
  61. * modified without taking the list lock).
  62. *
  63. * The list_lock is a centralized lock and thus we avoid taking it as
  64. * much as possible. As long as SLUB does not have to handle partial
  65. * slabs, operations can continue without any centralized lock. F.e.
  66. * allocating a long series of objects that fill up slabs does not require
  67. * the list lock.
  68. * Interrupts are disabled during allocation and deallocation in order to
  69. * make the slab allocator safe to use in the context of an irq. In addition
  70. * interrupts are disabled to ensure that the processor does not change
  71. * while handling per_cpu slabs, due to kernel preemption.
  72. *
  73. * SLUB assigns one slab for allocation to each processor.
  74. * Allocations only occur from these slabs called cpu slabs.
  75. *
  76. * Slabs with free elements are kept on a partial list and during regular
  77. * operations no list for full slabs is used. If an object in a full slab is
  78. * freed then the slab will show up again on the partial lists.
  79. * We track full slabs for debugging purposes though because otherwise we
  80. * cannot scan all objects.
  81. *
  82. * Slabs are freed when they become empty. Teardown and setup is
  83. * minimal so we rely on the page allocators per cpu caches for
  84. * fast frees and allocs.
  85. *
  86. * Overloading of page flags that are otherwise used for LRU management.
  87. *
  88. * PageActive The slab is frozen and exempt from list processing.
  89. * This means that the slab is dedicated to a purpose
  90. * such as satisfying allocations for a specific
  91. * processor. Objects may be freed in the slab while
  92. * it is frozen but slab_free will then skip the usual
  93. * list operations. It is up to the processor holding
  94. * the slab to integrate the slab into the slab lists
  95. * when the slab is no longer needed.
  96. *
  97. * One use of this flag is to mark slabs that are
  98. * used for allocations. Then such a slab becomes a cpu
  99. * slab. The cpu slab may be equipped with an additional
  100. * freelist that allows lockless access to
  101. * free objects in addition to the regular freelist
  102. * that requires the slab lock.
  103. *
  104. * PageError Slab requires special handling due to debug
  105. * options set. This moves slab handling out of
  106. * the fast path and disables lockless freelists.
  107. */
  108. #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
  109. SLAB_TRACE | SLAB_DEBUG_FREE)
  110. static inline int kmem_cache_debug(struct kmem_cache *s)
  111. {
  112. #ifdef CONFIG_SLUB_DEBUG
  113. return unlikely(s->flags & SLAB_DEBUG_FLAGS);
  114. #else
  115. return 0;
  116. #endif
  117. }
  118. /*
  119. * Issues still to be resolved:
  120. *
  121. * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
  122. *
  123. * - Variable sizing of the per node arrays
  124. */
  125. /* Enable to test recovery from slab corruption on boot */
  126. #undef SLUB_RESILIENCY_TEST
  127. /* Enable to log cmpxchg failures */
  128. #undef SLUB_DEBUG_CMPXCHG
  129. /*
  130. * Mininum number of partial slabs. These will be left on the partial
  131. * lists even if they are empty. kmem_cache_shrink may reclaim them.
  132. */
  133. #define MIN_PARTIAL 5
  134. /*
  135. * Maximum number of desirable partial slabs.
  136. * The existence of more partial slabs makes kmem_cache_shrink
  137. * sort the partial list by the number of objects in the.
  138. */
  139. #define MAX_PARTIAL 10
  140. #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
  141. SLAB_POISON | SLAB_STORE_USER)
  142. /*
  143. * Debugging flags that require metadata to be stored in the slab. These get
  144. * disabled when slub_debug=O is used and a cache's min order increases with
  145. * metadata.
  146. */
  147. #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
  148. /*
  149. * Set of flags that will prevent slab merging
  150. */
  151. #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
  152. SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \
  153. SLAB_FAILSLAB)
  154. #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
  155. SLAB_CACHE_DMA | SLAB_NOTRACK)
  156. #define OO_SHIFT 16
  157. #define OO_MASK ((1 << OO_SHIFT) - 1)
  158. #define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */
  159. /* Internal SLUB flags */
  160. #define __OBJECT_POISON 0x80000000UL /* Poison object */
  161. #define __CMPXCHG_DOUBLE 0x40000000UL /* Use cmpxchg_double */
  162. static int kmem_size = sizeof(struct kmem_cache);
  163. #ifdef CONFIG_SMP
  164. static struct notifier_block slab_notifier;
  165. #endif
  166. static enum {
  167. DOWN, /* No slab functionality available */
  168. PARTIAL, /* Kmem_cache_node works */
  169. UP, /* Everything works but does not show up in sysfs */
  170. SYSFS /* Sysfs up */
  171. } slab_state = DOWN;
  172. /* A list of all slab caches on the system */
  173. static DECLARE_RWSEM(slub_lock);
  174. static LIST_HEAD(slab_caches);
  175. /*
  176. * Tracking user of a slab.
  177. */
  178. #define TRACK_ADDRS_COUNT 16
  179. struct track {
  180. unsigned long addr; /* Called from address */
  181. #ifdef CONFIG_STACKTRACE
  182. unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */
  183. #endif
  184. int cpu; /* Was running on cpu */
  185. int pid; /* Pid context */
  186. unsigned long when; /* When did the operation occur */
  187. };
  188. enum track_item { TRACK_ALLOC, TRACK_FREE };
  189. #ifdef CONFIG_SYSFS
  190. static int sysfs_slab_add(struct kmem_cache *);
  191. static int sysfs_slab_alias(struct kmem_cache *, const char *);
  192. static void sysfs_slab_remove(struct kmem_cache *);
  193. #else
  194. static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; }
  195. static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
  196. { return 0; }
  197. static inline void sysfs_slab_remove(struct kmem_cache *s)
  198. {
  199. kfree(s->name);
  200. kfree(s);
  201. }
  202. #endif
  203. static inline void stat(const struct kmem_cache *s, enum stat_item si)
  204. {
  205. #ifdef CONFIG_SLUB_STATS
  206. __this_cpu_inc(s->cpu_slab->stat[si]);
  207. #endif
  208. }
  209. /********************************************************************
  210. * Core slab cache functions
  211. *******************************************************************/
  212. int slab_is_available(void)
  213. {
  214. return slab_state >= UP;
  215. }
  216. static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
  217. {
  218. return s->node[node];
  219. }
  220. /* Verify that a pointer has an address that is valid within a slab page */
  221. static inline int check_valid_pointer(struct kmem_cache *s,
  222. struct page *page, const void *object)
  223. {
  224. void *base;
  225. if (!object)
  226. return 1;
  227. base = page_address(page);
  228. if (object < base || object >= base + page->objects * s->size ||
  229. (object - base) % s->size) {
  230. return 0;
  231. }
  232. return 1;
  233. }
  234. static inline void *get_freepointer(struct kmem_cache *s, void *object)
  235. {
  236. return *(void **)(object + s->offset);
  237. }
  238. static void prefetch_freepointer(const struct kmem_cache *s, void *object)
  239. {
  240. prefetch(object + s->offset);
  241. }
  242. static inline void *get_freepointer_safe(struct kmem_cache *s, void *object)
  243. {
  244. void *p;
  245. #ifdef CONFIG_DEBUG_PAGEALLOC
  246. probe_kernel_read(&p, (void **)(object + s->offset), sizeof(p));
  247. #else
  248. p = get_freepointer(s, object);
  249. #endif
  250. return p;
  251. }
  252. static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
  253. {
  254. *(void **)(object + s->offset) = fp;
  255. }
  256. /* Loop over all objects in a slab */
  257. #define for_each_object(__p, __s, __addr, __objects) \
  258. for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
  259. __p += (__s)->size)
  260. /* Determine object index from a given position */
  261. static inline int slab_index(void *p, struct kmem_cache *s, void *addr)
  262. {
  263. return (p - addr) / s->size;
  264. }
  265. static inline size_t slab_ksize(const struct kmem_cache *s)
  266. {
  267. #ifdef CONFIG_SLUB_DEBUG
  268. /*
  269. * Debugging requires use of the padding between object
  270. * and whatever may come after it.
  271. */
  272. if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
  273. return s->object_size;
  274. #endif
  275. /*
  276. * If we have the need to store the freelist pointer
  277. * back there or track user information then we can
  278. * only use the space before that information.
  279. */
  280. if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER))
  281. return s->inuse;
  282. /*
  283. * Else we can use all the padding etc for the allocation
  284. */
  285. return s->size;
  286. }
  287. static inline int order_objects(int order, unsigned long size, int reserved)
  288. {
  289. return ((PAGE_SIZE << order) - reserved) / size;
  290. }
  291. static inline struct kmem_cache_order_objects oo_make(int order,
  292. unsigned long size, int reserved)
  293. {
  294. struct kmem_cache_order_objects x = {
  295. (order << OO_SHIFT) + order_objects(order, size, reserved)
  296. };
  297. return x;
  298. }
  299. static inline int oo_order(struct kmem_cache_order_objects x)
  300. {
  301. return x.x >> OO_SHIFT;
  302. }
  303. static inline int oo_objects(struct kmem_cache_order_objects x)
  304. {
  305. return x.x & OO_MASK;
  306. }
  307. /*
  308. * Per slab locking using the pagelock
  309. */
  310. static __always_inline void slab_lock(struct page *page)
  311. {
  312. bit_spin_lock(PG_locked, &page->flags);
  313. }
  314. static __always_inline void slab_unlock(struct page *page)
  315. {
  316. __bit_spin_unlock(PG_locked, &page->flags);
  317. }
  318. /* Interrupts must be disabled (for the fallback code to work right) */
  319. static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page,
  320. void *freelist_old, unsigned long counters_old,
  321. void *freelist_new, unsigned long counters_new,
  322. const char *n)
  323. {
  324. VM_BUG_ON(!irqs_disabled());
  325. #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
  326. defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
  327. if (s->flags & __CMPXCHG_DOUBLE) {
  328. if (cmpxchg_double(&page->freelist, &page->counters,
  329. freelist_old, counters_old,
  330. freelist_new, counters_new))
  331. return 1;
  332. } else
  333. #endif
  334. {
  335. slab_lock(page);
  336. if (page->freelist == freelist_old && page->counters == counters_old) {
  337. page->freelist = freelist_new;
  338. page->counters = counters_new;
  339. slab_unlock(page);
  340. return 1;
  341. }
  342. slab_unlock(page);
  343. }
  344. cpu_relax();
  345. stat(s, CMPXCHG_DOUBLE_FAIL);
  346. #ifdef SLUB_DEBUG_CMPXCHG
  347. printk(KERN_INFO "%s %s: cmpxchg double redo ", n, s->name);
  348. #endif
  349. return 0;
  350. }
  351. static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page,
  352. void *freelist_old, unsigned long counters_old,
  353. void *freelist_new, unsigned long counters_new,
  354. const char *n)
  355. {
  356. #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
  357. defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
  358. if (s->flags & __CMPXCHG_DOUBLE) {
  359. if (cmpxchg_double(&page->freelist, &page->counters,
  360. freelist_old, counters_old,
  361. freelist_new, counters_new))
  362. return 1;
  363. } else
  364. #endif
  365. {
  366. unsigned long flags;
  367. local_irq_save(flags);
  368. slab_lock(page);
  369. if (page->freelist == freelist_old && page->counters == counters_old) {
  370. page->freelist = freelist_new;
  371. page->counters = counters_new;
  372. slab_unlock(page);
  373. local_irq_restore(flags);
  374. return 1;
  375. }
  376. slab_unlock(page);
  377. local_irq_restore(flags);
  378. }
  379. cpu_relax();
  380. stat(s, CMPXCHG_DOUBLE_FAIL);
  381. #ifdef SLUB_DEBUG_CMPXCHG
  382. printk(KERN_INFO "%s %s: cmpxchg double redo ", n, s->name);
  383. #endif
  384. return 0;
  385. }
  386. #ifdef CONFIG_SLUB_DEBUG
  387. /*
  388. * Determine a map of object in use on a page.
  389. *
  390. * Node listlock must be held to guarantee that the page does
  391. * not vanish from under us.
  392. */
  393. static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map)
  394. {
  395. void *p;
  396. void *addr = page_address(page);
  397. for (p = page->freelist; p; p = get_freepointer(s, p))
  398. set_bit(slab_index(p, s, addr), map);
  399. }
  400. /*
  401. * Debug settings:
  402. */
  403. #ifdef CONFIG_SLUB_DEBUG_ON
  404. static int slub_debug = DEBUG_DEFAULT_FLAGS;
  405. #else
  406. static int slub_debug;
  407. #endif
  408. static char *slub_debug_slabs;
  409. static int disable_higher_order_debug;
  410. /*
  411. * Object debugging
  412. */
  413. static void print_section(char *text, u8 *addr, unsigned int length)
  414. {
  415. print_hex_dump(KERN_ERR, text, DUMP_PREFIX_ADDRESS, 16, 1, addr,
  416. length, 1);
  417. }
  418. static struct track *get_track(struct kmem_cache *s, void *object,
  419. enum track_item alloc)
  420. {
  421. struct track *p;
  422. if (s->offset)
  423. p = object + s->offset + sizeof(void *);
  424. else
  425. p = object + s->inuse;
  426. return p + alloc;
  427. }
  428. static void set_track(struct kmem_cache *s, void *object,
  429. enum track_item alloc, unsigned long addr)
  430. {
  431. struct track *p = get_track(s, object, alloc);
  432. if (addr) {
  433. #ifdef CONFIG_STACKTRACE
  434. struct stack_trace trace;
  435. int i;
  436. trace.nr_entries = 0;
  437. trace.max_entries = TRACK_ADDRS_COUNT;
  438. trace.entries = p->addrs;
  439. trace.skip = 3;
  440. save_stack_trace(&trace);
  441. /* See rant in lockdep.c */
  442. if (trace.nr_entries != 0 &&
  443. trace.entries[trace.nr_entries - 1] == ULONG_MAX)
  444. trace.nr_entries--;
  445. for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++)
  446. p->addrs[i] = 0;
  447. #endif
  448. p->addr = addr;
  449. p->cpu = smp_processor_id();
  450. p->pid = current->pid;
  451. p->when = jiffies;
  452. } else
  453. memset(p, 0, sizeof(struct track));
  454. }
  455. static void init_tracking(struct kmem_cache *s, void *object)
  456. {
  457. if (!(s->flags & SLAB_STORE_USER))
  458. return;
  459. set_track(s, object, TRACK_FREE, 0UL);
  460. set_track(s, object, TRACK_ALLOC, 0UL);
  461. }
  462. static void print_track(const char *s, struct track *t)
  463. {
  464. if (!t->addr)
  465. return;
  466. printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
  467. s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid);
  468. #ifdef CONFIG_STACKTRACE
  469. {
  470. int i;
  471. for (i = 0; i < TRACK_ADDRS_COUNT; i++)
  472. if (t->addrs[i])
  473. printk(KERN_ERR "\t%pS\n", (void *)t->addrs[i]);
  474. else
  475. break;
  476. }
  477. #endif
  478. }
  479. static void print_tracking(struct kmem_cache *s, void *object)
  480. {
  481. if (!(s->flags & SLAB_STORE_USER))
  482. return;
  483. print_track("Allocated", get_track(s, object, TRACK_ALLOC));
  484. print_track("Freed", get_track(s, object, TRACK_FREE));
  485. }
  486. static void print_page_info(struct page *page)
  487. {
  488. printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
  489. page, page->objects, page->inuse, page->freelist, page->flags);
  490. }
  491. static void slab_bug(struct kmem_cache *s, char *fmt, ...)
  492. {
  493. va_list args;
  494. char buf[100];
  495. va_start(args, fmt);
  496. vsnprintf(buf, sizeof(buf), fmt, args);
  497. va_end(args);
  498. printk(KERN_ERR "========================================"
  499. "=====================================\n");
  500. printk(KERN_ERR "BUG %s (%s): %s\n", s->name, print_tainted(), buf);
  501. printk(KERN_ERR "----------------------------------------"
  502. "-------------------------------------\n\n");
  503. }
  504. static void slab_fix(struct kmem_cache *s, char *fmt, ...)
  505. {
  506. va_list args;
  507. char buf[100];
  508. va_start(args, fmt);
  509. vsnprintf(buf, sizeof(buf), fmt, args);
  510. va_end(args);
  511. printk(KERN_ERR "FIX %s: %s\n", s->name, buf);
  512. }
  513. static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p)
  514. {
  515. unsigned int off; /* Offset of last byte */
  516. u8 *addr = page_address(page);
  517. print_tracking(s, p);
  518. print_page_info(page);
  519. printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
  520. p, p - addr, get_freepointer(s, p));
  521. if (p > addr + 16)
  522. print_section("Bytes b4 ", p - 16, 16);
  523. print_section("Object ", p, min_t(unsigned long, s->object_size,
  524. PAGE_SIZE));
  525. if (s->flags & SLAB_RED_ZONE)
  526. print_section("Redzone ", p + s->object_size,
  527. s->inuse - s->object_size);
  528. if (s->offset)
  529. off = s->offset + sizeof(void *);
  530. else
  531. off = s->inuse;
  532. if (s->flags & SLAB_STORE_USER)
  533. off += 2 * sizeof(struct track);
  534. if (off != s->size)
  535. /* Beginning of the filler is the free pointer */
  536. print_section("Padding ", p + off, s->size - off);
  537. dump_stack();
  538. }
  539. static void object_err(struct kmem_cache *s, struct page *page,
  540. u8 *object, char *reason)
  541. {
  542. slab_bug(s, "%s", reason);
  543. print_trailer(s, page, object);
  544. }
  545. static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...)
  546. {
  547. va_list args;
  548. char buf[100];
  549. va_start(args, fmt);
  550. vsnprintf(buf, sizeof(buf), fmt, args);
  551. va_end(args);
  552. slab_bug(s, "%s", buf);
  553. print_page_info(page);
  554. dump_stack();
  555. }
  556. static void init_object(struct kmem_cache *s, void *object, u8 val)
  557. {
  558. u8 *p = object;
  559. if (s->flags & __OBJECT_POISON) {
  560. memset(p, POISON_FREE, s->object_size - 1);
  561. p[s->object_size - 1] = POISON_END;
  562. }
  563. if (s->flags & SLAB_RED_ZONE)
  564. memset(p + s->object_size, val, s->inuse - s->object_size);
  565. }
  566. static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
  567. void *from, void *to)
  568. {
  569. slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data);
  570. memset(from, data, to - from);
  571. }
  572. static int check_bytes_and_report(struct kmem_cache *s, struct page *page,
  573. u8 *object, char *what,
  574. u8 *start, unsigned int value, unsigned int bytes)
  575. {
  576. u8 *fault;
  577. u8 *end;
  578. fault = memchr_inv(start, value, bytes);
  579. if (!fault)
  580. return 1;
  581. end = start + bytes;
  582. while (end > fault && end[-1] == value)
  583. end--;
  584. slab_bug(s, "%s overwritten", what);
  585. printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
  586. fault, end - 1, fault[0], value);
  587. print_trailer(s, page, object);
  588. restore_bytes(s, what, value, fault, end);
  589. return 0;
  590. }
  591. /*
  592. * Object layout:
  593. *
  594. * object address
  595. * Bytes of the object to be managed.
  596. * If the freepointer may overlay the object then the free
  597. * pointer is the first word of the object.
  598. *
  599. * Poisoning uses 0x6b (POISON_FREE) and the last byte is
  600. * 0xa5 (POISON_END)
  601. *
  602. * object + s->object_size
  603. * Padding to reach word boundary. This is also used for Redzoning.
  604. * Padding is extended by another word if Redzoning is enabled and
  605. * object_size == inuse.
  606. *
  607. * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
  608. * 0xcc (RED_ACTIVE) for objects in use.
  609. *
  610. * object + s->inuse
  611. * Meta data starts here.
  612. *
  613. * A. Free pointer (if we cannot overwrite object on free)
  614. * B. Tracking data for SLAB_STORE_USER
  615. * C. Padding to reach required alignment boundary or at mininum
  616. * one word if debugging is on to be able to detect writes
  617. * before the word boundary.
  618. *
  619. * Padding is done using 0x5a (POISON_INUSE)
  620. *
  621. * object + s->size
  622. * Nothing is used beyond s->size.
  623. *
  624. * If slabcaches are merged then the object_size and inuse boundaries are mostly
  625. * ignored. And therefore no slab options that rely on these boundaries
  626. * may be used with merged slabcaches.
  627. */
  628. static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p)
  629. {
  630. unsigned long off = s->inuse; /* The end of info */
  631. if (s->offset)
  632. /* Freepointer is placed after the object. */
  633. off += sizeof(void *);
  634. if (s->flags & SLAB_STORE_USER)
  635. /* We also have user information there */
  636. off += 2 * sizeof(struct track);
  637. if (s->size == off)
  638. return 1;
  639. return check_bytes_and_report(s, page, p, "Object padding",
  640. p + off, POISON_INUSE, s->size - off);
  641. }
  642. /* Check the pad bytes at the end of a slab page */
  643. static int slab_pad_check(struct kmem_cache *s, struct page *page)
  644. {
  645. u8 *start;
  646. u8 *fault;
  647. u8 *end;
  648. int length;
  649. int remainder;
  650. if (!(s->flags & SLAB_POISON))
  651. return 1;
  652. start = page_address(page);
  653. length = (PAGE_SIZE << compound_order(page)) - s->reserved;
  654. end = start + length;
  655. remainder = length % s->size;
  656. if (!remainder)
  657. return 1;
  658. fault = memchr_inv(end - remainder, POISON_INUSE, remainder);
  659. if (!fault)
  660. return 1;
  661. while (end > fault && end[-1] == POISON_INUSE)
  662. end--;
  663. slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1);
  664. print_section("Padding ", end - remainder, remainder);
  665. restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end);
  666. return 0;
  667. }
  668. static int check_object(struct kmem_cache *s, struct page *page,
  669. void *object, u8 val)
  670. {
  671. u8 *p = object;
  672. u8 *endobject = object + s->object_size;
  673. if (s->flags & SLAB_RED_ZONE) {
  674. if (!check_bytes_and_report(s, page, object, "Redzone",
  675. endobject, val, s->inuse - s->object_size))
  676. return 0;
  677. } else {
  678. if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) {
  679. check_bytes_and_report(s, page, p, "Alignment padding",
  680. endobject, POISON_INUSE, s->inuse - s->object_size);
  681. }
  682. }
  683. if (s->flags & SLAB_POISON) {
  684. if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) &&
  685. (!check_bytes_and_report(s, page, p, "Poison", p,
  686. POISON_FREE, s->object_size - 1) ||
  687. !check_bytes_and_report(s, page, p, "Poison",
  688. p + s->object_size - 1, POISON_END, 1)))
  689. return 0;
  690. /*
  691. * check_pad_bytes cleans up on its own.
  692. */
  693. check_pad_bytes(s, page, p);
  694. }
  695. if (!s->offset && val == SLUB_RED_ACTIVE)
  696. /*
  697. * Object and freepointer overlap. Cannot check
  698. * freepointer while object is allocated.
  699. */
  700. return 1;
  701. /* Check free pointer validity */
  702. if (!check_valid_pointer(s, page, get_freepointer(s, p))) {
  703. object_err(s, page, p, "Freepointer corrupt");
  704. /*
  705. * No choice but to zap it and thus lose the remainder
  706. * of the free objects in this slab. May cause
  707. * another error because the object count is now wrong.
  708. */
  709. set_freepointer(s, p, NULL);
  710. return 0;
  711. }
  712. return 1;
  713. }
  714. static int check_slab(struct kmem_cache *s, struct page *page)
  715. {
  716. int maxobj;
  717. VM_BUG_ON(!irqs_disabled());
  718. if (!PageSlab(page)) {
  719. slab_err(s, page, "Not a valid slab page");
  720. return 0;
  721. }
  722. maxobj = order_objects(compound_order(page), s->size, s->reserved);
  723. if (page->objects > maxobj) {
  724. slab_err(s, page, "objects %u > max %u",
  725. s->name, page->objects, maxobj);
  726. return 0;
  727. }
  728. if (page->inuse > page->objects) {
  729. slab_err(s, page, "inuse %u > max %u",
  730. s->name, page->inuse, page->objects);
  731. return 0;
  732. }
  733. /* Slab_pad_check fixes things up after itself */
  734. slab_pad_check(s, page);
  735. return 1;
  736. }
  737. /*
  738. * Determine if a certain object on a page is on the freelist. Must hold the
  739. * slab lock to guarantee that the chains are in a consistent state.
  740. */
  741. static int on_freelist(struct kmem_cache *s, struct page *page, void *search)
  742. {
  743. int nr = 0;
  744. void *fp;
  745. void *object = NULL;
  746. unsigned long max_objects;
  747. fp = page->freelist;
  748. while (fp && nr <= page->objects) {
  749. if (fp == search)
  750. return 1;
  751. if (!check_valid_pointer(s, page, fp)) {
  752. if (object) {
  753. object_err(s, page, object,
  754. "Freechain corrupt");
  755. set_freepointer(s, object, NULL);
  756. break;
  757. } else {
  758. slab_err(s, page, "Freepointer corrupt");
  759. page->freelist = NULL;
  760. page->inuse = page->objects;
  761. slab_fix(s, "Freelist cleared");
  762. return 0;
  763. }
  764. break;
  765. }
  766. object = fp;
  767. fp = get_freepointer(s, object);
  768. nr++;
  769. }
  770. max_objects = order_objects(compound_order(page), s->size, s->reserved);
  771. if (max_objects > MAX_OBJS_PER_PAGE)
  772. max_objects = MAX_OBJS_PER_PAGE;
  773. if (page->objects != max_objects) {
  774. slab_err(s, page, "Wrong number of objects. Found %d but "
  775. "should be %d", page->objects, max_objects);
  776. page->objects = max_objects;
  777. slab_fix(s, "Number of objects adjusted.");
  778. }
  779. if (page->inuse != page->objects - nr) {
  780. slab_err(s, page, "Wrong object count. Counter is %d but "
  781. "counted were %d", page->inuse, page->objects - nr);
  782. page->inuse = page->objects - nr;
  783. slab_fix(s, "Object count adjusted.");
  784. }
  785. return search == NULL;
  786. }
  787. static void trace(struct kmem_cache *s, struct page *page, void *object,
  788. int alloc)
  789. {
  790. if (s->flags & SLAB_TRACE) {
  791. printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
  792. s->name,
  793. alloc ? "alloc" : "free",
  794. object, page->inuse,
  795. page->freelist);
  796. if (!alloc)
  797. print_section("Object ", (void *)object, s->object_size);
  798. dump_stack();
  799. }
  800. }
  801. /*
  802. * Hooks for other subsystems that check memory allocations. In a typical
  803. * production configuration these hooks all should produce no code at all.
  804. */
  805. static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags)
  806. {
  807. flags &= gfp_allowed_mask;
  808. lockdep_trace_alloc(flags);
  809. might_sleep_if(flags & __GFP_WAIT);
  810. return should_failslab(s->object_size, flags, s->flags);
  811. }
  812. static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, void *object)
  813. {
  814. flags &= gfp_allowed_mask;
  815. kmemcheck_slab_alloc(s, flags, object, slab_ksize(s));
  816. kmemleak_alloc_recursive(object, s->object_size, 1, s->flags, flags);
  817. }
  818. static inline void slab_free_hook(struct kmem_cache *s, void *x)
  819. {
  820. kmemleak_free_recursive(x, s->flags);
  821. /*
  822. * Trouble is that we may no longer disable interupts in the fast path
  823. * So in order to make the debug calls that expect irqs to be
  824. * disabled we need to disable interrupts temporarily.
  825. */
  826. #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP)
  827. {
  828. unsigned long flags;
  829. local_irq_save(flags);
  830. kmemcheck_slab_free(s, x, s->object_size);
  831. debug_check_no_locks_freed(x, s->object_size);
  832. local_irq_restore(flags);
  833. }
  834. #endif
  835. if (!(s->flags & SLAB_DEBUG_OBJECTS))
  836. debug_check_no_obj_freed(x, s->object_size);
  837. }
  838. /*
  839. * Tracking of fully allocated slabs for debugging purposes.
  840. *
  841. * list_lock must be held.
  842. */
  843. static void add_full(struct kmem_cache *s,
  844. struct kmem_cache_node *n, struct page *page)
  845. {
  846. if (!(s->flags & SLAB_STORE_USER))
  847. return;
  848. list_add(&page->lru, &n->full);
  849. }
  850. /*
  851. * list_lock must be held.
  852. */
  853. static void remove_full(struct kmem_cache *s, struct page *page)
  854. {
  855. if (!(s->flags & SLAB_STORE_USER))
  856. return;
  857. list_del(&page->lru);
  858. }
  859. /* Tracking of the number of slabs for debugging purposes */
  860. static inline unsigned long slabs_node(struct kmem_cache *s, int node)
  861. {
  862. struct kmem_cache_node *n = get_node(s, node);
  863. return atomic_long_read(&n->nr_slabs);
  864. }
  865. static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
  866. {
  867. return atomic_long_read(&n->nr_slabs);
  868. }
  869. static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
  870. {
  871. struct kmem_cache_node *n = get_node(s, node);
  872. /*
  873. * May be called early in order to allocate a slab for the
  874. * kmem_cache_node structure. Solve the chicken-egg
  875. * dilemma by deferring the increment of the count during
  876. * bootstrap (see early_kmem_cache_node_alloc).
  877. */
  878. if (n) {
  879. atomic_long_inc(&n->nr_slabs);
  880. atomic_long_add(objects, &n->total_objects);
  881. }
  882. }
  883. static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
  884. {
  885. struct kmem_cache_node *n = get_node(s, node);
  886. atomic_long_dec(&n->nr_slabs);
  887. atomic_long_sub(objects, &n->total_objects);
  888. }
  889. /* Object debug checks for alloc/free paths */
  890. static void setup_object_debug(struct kmem_cache *s, struct page *page,
  891. void *object)
  892. {
  893. if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)))
  894. return;
  895. init_object(s, object, SLUB_RED_INACTIVE);
  896. init_tracking(s, object);
  897. }
  898. static noinline int alloc_debug_processing(struct kmem_cache *s, struct page *page,
  899. void *object, unsigned long addr)
  900. {
  901. if (!check_slab(s, page))
  902. goto bad;
  903. if (!check_valid_pointer(s, page, object)) {
  904. object_err(s, page, object, "Freelist Pointer check fails");
  905. goto bad;
  906. }
  907. if (!check_object(s, page, object, SLUB_RED_INACTIVE))
  908. goto bad;
  909. /* Success perform special debug activities for allocs */
  910. if (s->flags & SLAB_STORE_USER)
  911. set_track(s, object, TRACK_ALLOC, addr);
  912. trace(s, page, object, 1);
  913. init_object(s, object, SLUB_RED_ACTIVE);
  914. return 1;
  915. bad:
  916. if (PageSlab(page)) {
  917. /*
  918. * If this is a slab page then lets do the best we can
  919. * to avoid issues in the future. Marking all objects
  920. * as used avoids touching the remaining objects.
  921. */
  922. slab_fix(s, "Marking all objects used");
  923. page->inuse = page->objects;
  924. page->freelist = NULL;
  925. }
  926. return 0;
  927. }
  928. static noinline int free_debug_processing(struct kmem_cache *s,
  929. struct page *page, void *object, unsigned long addr)
  930. {
  931. unsigned long flags;
  932. int rc = 0;
  933. local_irq_save(flags);
  934. slab_lock(page);
  935. if (!check_slab(s, page))
  936. goto fail;
  937. if (!check_valid_pointer(s, page, object)) {
  938. slab_err(s, page, "Invalid object pointer 0x%p", object);
  939. goto fail;
  940. }
  941. if (on_freelist(s, page, object)) {
  942. object_err(s, page, object, "Object already free");
  943. goto fail;
  944. }
  945. if (!check_object(s, page, object, SLUB_RED_ACTIVE))
  946. goto out;
  947. if (unlikely(s != page->slab)) {
  948. if (!PageSlab(page)) {
  949. slab_err(s, page, "Attempt to free object(0x%p) "
  950. "outside of slab", object);
  951. } else if (!page->slab) {
  952. printk(KERN_ERR
  953. "SLUB <none>: no slab for object 0x%p.\n",
  954. object);
  955. dump_stack();
  956. } else
  957. object_err(s, page, object,
  958. "page slab pointer corrupt.");
  959. goto fail;
  960. }
  961. if (s->flags & SLAB_STORE_USER)
  962. set_track(s, object, TRACK_FREE, addr);
  963. trace(s, page, object, 0);
  964. init_object(s, object, SLUB_RED_INACTIVE);
  965. rc = 1;
  966. out:
  967. slab_unlock(page);
  968. local_irq_restore(flags);
  969. return rc;
  970. fail:
  971. slab_fix(s, "Object at 0x%p not freed", object);
  972. goto out;
  973. }
  974. static int __init setup_slub_debug(char *str)
  975. {
  976. slub_debug = DEBUG_DEFAULT_FLAGS;
  977. if (*str++ != '=' || !*str)
  978. /*
  979. * No options specified. Switch on full debugging.
  980. */
  981. goto out;
  982. if (*str == ',')
  983. /*
  984. * No options but restriction on slabs. This means full
  985. * debugging for slabs matching a pattern.
  986. */
  987. goto check_slabs;
  988. if (tolower(*str) == 'o') {
  989. /*
  990. * Avoid enabling debugging on caches if its minimum order
  991. * would increase as a result.
  992. */
  993. disable_higher_order_debug = 1;
  994. goto out;
  995. }
  996. slub_debug = 0;
  997. if (*str == '-')
  998. /*
  999. * Switch off all debugging measures.
  1000. */
  1001. goto out;
  1002. /*
  1003. * Determine which debug features should be switched on
  1004. */
  1005. for (; *str && *str != ','; str++) {
  1006. switch (tolower(*str)) {
  1007. case 'f':
  1008. slub_debug |= SLAB_DEBUG_FREE;
  1009. break;
  1010. case 'z':
  1011. slub_debug |= SLAB_RED_ZONE;
  1012. break;
  1013. case 'p':
  1014. slub_debug |= SLAB_POISON;
  1015. break;
  1016. case 'u':
  1017. slub_debug |= SLAB_STORE_USER;
  1018. break;
  1019. case 't':
  1020. slub_debug |= SLAB_TRACE;
  1021. break;
  1022. case 'a':
  1023. slub_debug |= SLAB_FAILSLAB;
  1024. break;
  1025. default:
  1026. printk(KERN_ERR "slub_debug option '%c' "
  1027. "unknown. skipped\n", *str);
  1028. }
  1029. }
  1030. check_slabs:
  1031. if (*str == ',')
  1032. slub_debug_slabs = str + 1;
  1033. out:
  1034. return 1;
  1035. }
  1036. __setup("slub_debug", setup_slub_debug);
  1037. static unsigned long kmem_cache_flags(unsigned long object_size,
  1038. unsigned long flags, const char *name,
  1039. void (*ctor)(void *))
  1040. {
  1041. /*
  1042. * Enable debugging if selected on the kernel commandline.
  1043. */
  1044. if (slub_debug && (!slub_debug_slabs ||
  1045. !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs))))
  1046. flags |= slub_debug;
  1047. return flags;
  1048. }
  1049. #else
  1050. static inline void setup_object_debug(struct kmem_cache *s,
  1051. struct page *page, void *object) {}
  1052. static inline int alloc_debug_processing(struct kmem_cache *s,
  1053. struct page *page, void *object, unsigned long addr) { return 0; }
  1054. static inline int free_debug_processing(struct kmem_cache *s,
  1055. struct page *page, void *object, unsigned long addr) { return 0; }
  1056. static inline int slab_pad_check(struct kmem_cache *s, struct page *page)
  1057. { return 1; }
  1058. static inline int check_object(struct kmem_cache *s, struct page *page,
  1059. void *object, u8 val) { return 1; }
  1060. static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n,
  1061. struct page *page) {}
  1062. static inline void remove_full(struct kmem_cache *s, struct page *page) {}
  1063. static inline unsigned long kmem_cache_flags(unsigned long object_size,
  1064. unsigned long flags, const char *name,
  1065. void (*ctor)(void *))
  1066. {
  1067. return flags;
  1068. }
  1069. #define slub_debug 0
  1070. #define disable_higher_order_debug 0
  1071. static inline unsigned long slabs_node(struct kmem_cache *s, int node)
  1072. { return 0; }
  1073. static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
  1074. { return 0; }
  1075. static inline void inc_slabs_node(struct kmem_cache *s, int node,
  1076. int objects) {}
  1077. static inline void dec_slabs_node(struct kmem_cache *s, int node,
  1078. int objects) {}
  1079. static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags)
  1080. { return 0; }
  1081. static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags,
  1082. void *object) {}
  1083. static inline void slab_free_hook(struct kmem_cache *s, void *x) {}
  1084. #endif /* CONFIG_SLUB_DEBUG */
  1085. /*
  1086. * Slab allocation and freeing
  1087. */
  1088. static inline struct page *alloc_slab_page(gfp_t flags, int node,
  1089. struct kmem_cache_order_objects oo)
  1090. {
  1091. int order = oo_order(oo);
  1092. flags |= __GFP_NOTRACK;
  1093. if (node == NUMA_NO_NODE)
  1094. return alloc_pages(flags, order);
  1095. else
  1096. return alloc_pages_exact_node(node, flags, order);
  1097. }
  1098. static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
  1099. {
  1100. struct page *page;
  1101. struct kmem_cache_order_objects oo = s->oo;
  1102. gfp_t alloc_gfp;
  1103. flags &= gfp_allowed_mask;
  1104. if (flags & __GFP_WAIT)
  1105. local_irq_enable();
  1106. flags |= s->allocflags;
  1107. /*
  1108. * Let the initial higher-order allocation fail under memory pressure
  1109. * so we fall-back to the minimum order allocation.
  1110. */
  1111. alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
  1112. page = alloc_slab_page(alloc_gfp, node, oo);
  1113. if (unlikely(!page)) {
  1114. oo = s->min;
  1115. /*
  1116. * Allocation may have failed due to fragmentation.
  1117. * Try a lower order alloc if possible
  1118. */
  1119. page = alloc_slab_page(flags, node, oo);
  1120. if (page)
  1121. stat(s, ORDER_FALLBACK);
  1122. }
  1123. if (flags & __GFP_WAIT)
  1124. local_irq_disable();
  1125. if (!page)
  1126. return NULL;
  1127. if (kmemcheck_enabled
  1128. && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) {
  1129. int pages = 1 << oo_order(oo);
  1130. kmemcheck_alloc_shadow(page, oo_order(oo), flags, node);
  1131. /*
  1132. * Objects from caches that have a constructor don't get
  1133. * cleared when they're allocated, so we need to do it here.
  1134. */
  1135. if (s->ctor)
  1136. kmemcheck_mark_uninitialized_pages(page, pages);
  1137. else
  1138. kmemcheck_mark_unallocated_pages(page, pages);
  1139. }
  1140. page->objects = oo_objects(oo);
  1141. mod_zone_page_state(page_zone(page),
  1142. (s->flags & SLAB_RECLAIM_ACCOUNT) ?
  1143. NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
  1144. 1 << oo_order(oo));
  1145. return page;
  1146. }
  1147. static void setup_object(struct kmem_cache *s, struct page *page,
  1148. void *object)
  1149. {
  1150. setup_object_debug(s, page, object);
  1151. if (unlikely(s->ctor))
  1152. s->ctor(object);
  1153. }
  1154. static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
  1155. {
  1156. struct page *page;
  1157. void *start;
  1158. void *last;
  1159. void *p;
  1160. BUG_ON(flags & GFP_SLAB_BUG_MASK);
  1161. page = allocate_slab(s,
  1162. flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
  1163. if (!page)
  1164. goto out;
  1165. inc_slabs_node(s, page_to_nid(page), page->objects);
  1166. page->slab = s;
  1167. __SetPageSlab(page);
  1168. start = page_address(page);
  1169. if (unlikely(s->flags & SLAB_POISON))
  1170. memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page));
  1171. last = start;
  1172. for_each_object(p, s, start, page->objects) {
  1173. setup_object(s, page, last);
  1174. set_freepointer(s, last, p);
  1175. last = p;
  1176. }
  1177. setup_object(s, page, last);
  1178. set_freepointer(s, last, NULL);
  1179. page->freelist = start;
  1180. page->inuse = page->objects;
  1181. page->frozen = 1;
  1182. out:
  1183. return page;
  1184. }
  1185. static void __free_slab(struct kmem_cache *s, struct page *page)
  1186. {
  1187. int order = compound_order(page);
  1188. int pages = 1 << order;
  1189. if (kmem_cache_debug(s)) {
  1190. void *p;
  1191. slab_pad_check(s, page);
  1192. for_each_object(p, s, page_address(page),
  1193. page->objects)
  1194. check_object(s, page, p, SLUB_RED_INACTIVE);
  1195. }
  1196. kmemcheck_free_shadow(page, compound_order(page));
  1197. mod_zone_page_state(page_zone(page),
  1198. (s->flags & SLAB_RECLAIM_ACCOUNT) ?
  1199. NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
  1200. -pages);
  1201. __ClearPageSlab(page);
  1202. reset_page_mapcount(page);
  1203. if (current->reclaim_state)
  1204. current->reclaim_state->reclaimed_slab += pages;
  1205. __free_pages(page, order);
  1206. }
  1207. #define need_reserve_slab_rcu \
  1208. (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head))
  1209. static void rcu_free_slab(struct rcu_head *h)
  1210. {
  1211. struct page *page;
  1212. if (need_reserve_slab_rcu)
  1213. page = virt_to_head_page(h);
  1214. else
  1215. page = container_of((struct list_head *)h, struct page, lru);
  1216. __free_slab(page->slab, page);
  1217. }
  1218. static void free_slab(struct kmem_cache *s, struct page *page)
  1219. {
  1220. if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) {
  1221. struct rcu_head *head;
  1222. if (need_reserve_slab_rcu) {
  1223. int order = compound_order(page);
  1224. int offset = (PAGE_SIZE << order) - s->reserved;
  1225. VM_BUG_ON(s->reserved != sizeof(*head));
  1226. head = page_address(page) + offset;
  1227. } else {
  1228. /*
  1229. * RCU free overloads the RCU head over the LRU
  1230. */
  1231. head = (void *)&page->lru;
  1232. }
  1233. call_rcu(head, rcu_free_slab);
  1234. } else
  1235. __free_slab(s, page);
  1236. }
  1237. static void discard_slab(struct kmem_cache *s, struct page *page)
  1238. {
  1239. dec_slabs_node(s, page_to_nid(page), page->objects);
  1240. free_slab(s, page);
  1241. }
  1242. /*
  1243. * Management of partially allocated slabs.
  1244. *
  1245. * list_lock must be held.
  1246. */
  1247. static inline void add_partial(struct kmem_cache_node *n,
  1248. struct page *page, int tail)
  1249. {
  1250. n->nr_partial++;
  1251. if (tail == DEACTIVATE_TO_TAIL)
  1252. list_add_tail(&page->lru, &n->partial);
  1253. else
  1254. list_add(&page->lru, &n->partial);
  1255. }
  1256. /*
  1257. * list_lock must be held.
  1258. */
  1259. static inline void remove_partial(struct kmem_cache_node *n,
  1260. struct page *page)
  1261. {
  1262. list_del(&page->lru);
  1263. n->nr_partial--;
  1264. }
  1265. /*
  1266. * Remove slab from the partial list, freeze it and
  1267. * return the pointer to the freelist.
  1268. *
  1269. * Returns a list of objects or NULL if it fails.
  1270. *
  1271. * Must hold list_lock since we modify the partial list.
  1272. */
  1273. static inline void *acquire_slab(struct kmem_cache *s,
  1274. struct kmem_cache_node *n, struct page *page,
  1275. int mode)
  1276. {
  1277. void *freelist;
  1278. unsigned long counters;
  1279. struct page new;
  1280. /*
  1281. * Zap the freelist and set the frozen bit.
  1282. * The old freelist is the list of objects for the
  1283. * per cpu allocation list.
  1284. */
  1285. freelist = page->freelist;
  1286. counters = page->counters;
  1287. new.counters = counters;
  1288. if (mode) {
  1289. new.inuse = page->objects;
  1290. new.freelist = NULL;
  1291. } else {
  1292. new.freelist = freelist;
  1293. }
  1294. VM_BUG_ON(new.frozen);
  1295. new.frozen = 1;
  1296. if (!__cmpxchg_double_slab(s, page,
  1297. freelist, counters,
  1298. new.freelist, new.counters,
  1299. "acquire_slab"))
  1300. return NULL;
  1301. remove_partial(n, page);
  1302. WARN_ON(!freelist);
  1303. return freelist;
  1304. }
  1305. static int put_cpu_partial(struct kmem_cache *s, struct page *page, int drain);
  1306. /*
  1307. * Try to allocate a partial slab from a specific node.
  1308. */
  1309. static void *get_partial_node(struct kmem_cache *s,
  1310. struct kmem_cache_node *n, struct kmem_cache_cpu *c)
  1311. {
  1312. struct page *page, *page2;
  1313. void *object = NULL;
  1314. /*
  1315. * Racy check. If we mistakenly see no partial slabs then we
  1316. * just allocate an empty slab. If we mistakenly try to get a
  1317. * partial slab and there is none available then get_partials()
  1318. * will return NULL.
  1319. */
  1320. if (!n || !n->nr_partial)
  1321. return NULL;
  1322. spin_lock(&n->list_lock);
  1323. list_for_each_entry_safe(page, page2, &n->partial, lru) {
  1324. void *t = acquire_slab(s, n, page, object == NULL);
  1325. int available;
  1326. if (!t)
  1327. break;
  1328. if (!object) {
  1329. c->page = page;
  1330. stat(s, ALLOC_FROM_PARTIAL);
  1331. object = t;
  1332. available = page->objects - page->inuse;
  1333. } else {
  1334. available = put_cpu_partial(s, page, 0);
  1335. stat(s, CPU_PARTIAL_NODE);
  1336. }
  1337. if (kmem_cache_debug(s) || available > s->cpu_partial / 2)
  1338. break;
  1339. }
  1340. spin_unlock(&n->list_lock);
  1341. return object;
  1342. }
  1343. /*
  1344. * Get a page from somewhere. Search in increasing NUMA distances.
  1345. */
  1346. static void *get_any_partial(struct kmem_cache *s, gfp_t flags,
  1347. struct kmem_cache_cpu *c)
  1348. {
  1349. #ifdef CONFIG_NUMA
  1350. struct zonelist *zonelist;
  1351. struct zoneref *z;
  1352. struct zone *zone;
  1353. enum zone_type high_zoneidx = gfp_zone(flags);
  1354. void *object;
  1355. unsigned int cpuset_mems_cookie;
  1356. /*
  1357. * The defrag ratio allows a configuration of the tradeoffs between
  1358. * inter node defragmentation and node local allocations. A lower
  1359. * defrag_ratio increases the tendency to do local allocations
  1360. * instead of attempting to obtain partial slabs from other nodes.
  1361. *
  1362. * If the defrag_ratio is set to 0 then kmalloc() always
  1363. * returns node local objects. If the ratio is higher then kmalloc()
  1364. * may return off node objects because partial slabs are obtained
  1365. * from other nodes and filled up.
  1366. *
  1367. * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
  1368. * defrag_ratio = 1000) then every (well almost) allocation will
  1369. * first attempt to defrag slab caches on other nodes. This means
  1370. * scanning over all nodes to look for partial slabs which may be
  1371. * expensive if we do it every time we are trying to find a slab
  1372. * with available objects.
  1373. */
  1374. if (!s->remote_node_defrag_ratio ||
  1375. get_cycles() % 1024 > s->remote_node_defrag_ratio)
  1376. return NULL;
  1377. do {
  1378. cpuset_mems_cookie = get_mems_allowed();
  1379. zonelist = node_zonelist(slab_node(), flags);
  1380. for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
  1381. struct kmem_cache_node *n;
  1382. n = get_node(s, zone_to_nid(zone));
  1383. if (n && cpuset_zone_allowed_hardwall(zone, flags) &&
  1384. n->nr_partial > s->min_partial) {
  1385. object = get_partial_node(s, n, c);
  1386. if (object) {
  1387. /*
  1388. * Return the object even if
  1389. * put_mems_allowed indicated that
  1390. * the cpuset mems_allowed was
  1391. * updated in parallel. It's a
  1392. * harmless race between the alloc
  1393. * and the cpuset update.
  1394. */
  1395. put_mems_allowed(cpuset_mems_cookie);
  1396. return object;
  1397. }
  1398. }
  1399. }
  1400. } while (!put_mems_allowed(cpuset_mems_cookie));
  1401. #endif
  1402. return NULL;
  1403. }
  1404. /*
  1405. * Get a partial page, lock it and return it.
  1406. */
  1407. static void *get_partial(struct kmem_cache *s, gfp_t flags, int node,
  1408. struct kmem_cache_cpu *c)
  1409. {
  1410. void *object;
  1411. int searchnode = (node == NUMA_NO_NODE) ? numa_node_id() : node;
  1412. object = get_partial_node(s, get_node(s, searchnode), c);
  1413. if (object || node != NUMA_NO_NODE)
  1414. return object;
  1415. return get_any_partial(s, flags, c);
  1416. }
  1417. #ifdef CONFIG_PREEMPT
  1418. /*
  1419. * Calculate the next globally unique transaction for disambiguiation
  1420. * during cmpxchg. The transactions start with the cpu number and are then
  1421. * incremented by CONFIG_NR_CPUS.
  1422. */
  1423. #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS)
  1424. #else
  1425. /*
  1426. * No preemption supported therefore also no need to check for
  1427. * different cpus.
  1428. */
  1429. #define TID_STEP 1
  1430. #endif
  1431. static inline unsigned long next_tid(unsigned long tid)
  1432. {
  1433. return tid + TID_STEP;
  1434. }
  1435. static inline unsigned int tid_to_cpu(unsigned long tid)
  1436. {
  1437. return tid % TID_STEP;
  1438. }
  1439. static inline unsigned long tid_to_event(unsigned long tid)
  1440. {
  1441. return tid / TID_STEP;
  1442. }
  1443. static inline unsigned int init_tid(int cpu)
  1444. {
  1445. return cpu;
  1446. }
  1447. static inline void note_cmpxchg_failure(const char *n,
  1448. const struct kmem_cache *s, unsigned long tid)
  1449. {
  1450. #ifdef SLUB_DEBUG_CMPXCHG
  1451. unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid);
  1452. printk(KERN_INFO "%s %s: cmpxchg redo ", n, s->name);
  1453. #ifdef CONFIG_PREEMPT
  1454. if (tid_to_cpu(tid) != tid_to_cpu(actual_tid))
  1455. printk("due to cpu change %d -> %d\n",
  1456. tid_to_cpu(tid), tid_to_cpu(actual_tid));
  1457. else
  1458. #endif
  1459. if (tid_to_event(tid) != tid_to_event(actual_tid))
  1460. printk("due to cpu running other code. Event %ld->%ld\n",
  1461. tid_to_event(tid), tid_to_event(actual_tid));
  1462. else
  1463. printk("for unknown reason: actual=%lx was=%lx target=%lx\n",
  1464. actual_tid, tid, next_tid(tid));
  1465. #endif
  1466. stat(s, CMPXCHG_DOUBLE_CPU_FAIL);
  1467. }
  1468. void init_kmem_cache_cpus(struct kmem_cache *s)
  1469. {
  1470. int cpu;
  1471. for_each_possible_cpu(cpu)
  1472. per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu);
  1473. }
  1474. /*
  1475. * Remove the cpu slab
  1476. */
  1477. static void deactivate_slab(struct kmem_cache *s, struct page *page, void *freelist)
  1478. {
  1479. enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE };
  1480. struct kmem_cache_node *n = get_node(s, page_to_nid(page));
  1481. int lock = 0;
  1482. enum slab_modes l = M_NONE, m = M_NONE;
  1483. void *nextfree;
  1484. int tail = DEACTIVATE_TO_HEAD;
  1485. struct page new;
  1486. struct page old;
  1487. if (page->freelist) {
  1488. stat(s, DEACTIVATE_REMOTE_FREES);
  1489. tail = DEACTIVATE_TO_TAIL;
  1490. }
  1491. /*
  1492. * Stage one: Free all available per cpu objects back
  1493. * to the page freelist while it is still frozen. Leave the
  1494. * last one.
  1495. *
  1496. * There is no need to take the list->lock because the page
  1497. * is still frozen.
  1498. */
  1499. while (freelist && (nextfree = get_freepointer(s, freelist))) {
  1500. void *prior;
  1501. unsigned long counters;
  1502. do {
  1503. prior = page->freelist;
  1504. counters = page->counters;
  1505. set_freepointer(s, freelist, prior);
  1506. new.counters = counters;
  1507. new.inuse--;
  1508. VM_BUG_ON(!new.frozen);
  1509. } while (!__cmpxchg_double_slab(s, page,
  1510. prior, counters,
  1511. freelist, new.counters,
  1512. "drain percpu freelist"));
  1513. freelist = nextfree;
  1514. }
  1515. /*
  1516. * Stage two: Ensure that the page is unfrozen while the
  1517. * list presence reflects the actual number of objects
  1518. * during unfreeze.
  1519. *
  1520. * We setup the list membership and then perform a cmpxchg
  1521. * with the count. If there is a mismatch then the page
  1522. * is not unfrozen but the page is on the wrong list.
  1523. *
  1524. * Then we restart the process which may have to remove
  1525. * the page from the list that we just put it on again
  1526. * because the number of objects in the slab may have
  1527. * changed.
  1528. */
  1529. redo:
  1530. old.freelist = page->freelist;
  1531. old.counters = page->counters;
  1532. VM_BUG_ON(!old.frozen);
  1533. /* Determine target state of the slab */
  1534. new.counters = old.counters;
  1535. if (freelist) {
  1536. new.inuse--;
  1537. set_freepointer(s, freelist, old.freelist);
  1538. new.freelist = freelist;
  1539. } else
  1540. new.freelist = old.freelist;
  1541. new.frozen = 0;
  1542. if (!new.inuse && n->nr_partial > s->min_partial)
  1543. m = M_FREE;
  1544. else if (new.freelist) {
  1545. m = M_PARTIAL;
  1546. if (!lock) {
  1547. lock = 1;
  1548. /*
  1549. * Taking the spinlock removes the possiblity
  1550. * that acquire_slab() will see a slab page that
  1551. * is frozen
  1552. */
  1553. spin_lock(&n->list_lock);
  1554. }
  1555. } else {
  1556. m = M_FULL;
  1557. if (kmem_cache_debug(s) && !lock) {
  1558. lock = 1;
  1559. /*
  1560. * This also ensures that the scanning of full
  1561. * slabs from diagnostic functions will not see
  1562. * any frozen slabs.
  1563. */
  1564. spin_lock(&n->list_lock);
  1565. }
  1566. }
  1567. if (l != m) {
  1568. if (l == M_PARTIAL)
  1569. remove_partial(n, page);
  1570. else if (l == M_FULL)
  1571. remove_full(s, page);
  1572. if (m == M_PARTIAL) {
  1573. add_partial(n, page, tail);
  1574. stat(s, tail);
  1575. } else if (m == M_FULL) {
  1576. stat(s, DEACTIVATE_FULL);
  1577. add_full(s, n, page);
  1578. }
  1579. }
  1580. l = m;
  1581. if (!__cmpxchg_double_slab(s, page,
  1582. old.freelist, old.counters,
  1583. new.freelist, new.counters,
  1584. "unfreezing slab"))
  1585. goto redo;
  1586. if (lock)
  1587. spin_unlock(&n->list_lock);
  1588. if (m == M_FREE) {
  1589. stat(s, DEACTIVATE_EMPTY);
  1590. discard_slab(s, page);
  1591. stat(s, FREE_SLAB);
  1592. }
  1593. }
  1594. /*
  1595. * Unfreeze all the cpu partial slabs.
  1596. *
  1597. * This function must be called with interrupt disabled.
  1598. */
  1599. static void unfreeze_partials(struct kmem_cache *s)
  1600. {
  1601. struct kmem_cache_node *n = NULL;
  1602. struct kmem_cache_cpu *c = this_cpu_ptr(s->cpu_slab);
  1603. struct page *page, *discard_page = NULL;
  1604. while ((page = c->partial)) {
  1605. enum slab_modes { M_PARTIAL, M_FREE };
  1606. enum slab_modes l, m;
  1607. struct page new;
  1608. struct page old;
  1609. c->partial = page->next;
  1610. l = M_FREE;
  1611. do {
  1612. old.freelist = page->freelist;
  1613. old.counters = page->counters;
  1614. VM_BUG_ON(!old.frozen);
  1615. new.counters = old.counters;
  1616. new.freelist = old.freelist;
  1617. new.frozen = 0;
  1618. if (!new.inuse && (!n || n->nr_partial > s->min_partial))
  1619. m = M_FREE;
  1620. else {
  1621. struct kmem_cache_node *n2 = get_node(s,
  1622. page_to_nid(page));
  1623. m = M_PARTIAL;
  1624. if (n != n2) {
  1625. if (n)
  1626. spin_unlock(&n->list_lock);
  1627. n = n2;
  1628. spin_lock(&n->list_lock);
  1629. }
  1630. }
  1631. if (l != m) {
  1632. if (l == M_PARTIAL) {
  1633. remove_partial(n, page);
  1634. stat(s, FREE_REMOVE_PARTIAL);
  1635. } else {
  1636. add_partial(n, page,
  1637. DEACTIVATE_TO_TAIL);
  1638. stat(s, FREE_ADD_PARTIAL);
  1639. }
  1640. l = m;
  1641. }
  1642. } while (!__cmpxchg_double_slab(s, page,
  1643. old.freelist, old.counters,
  1644. new.freelist, new.counters,
  1645. "unfreezing slab"));
  1646. if (m == M_FREE) {
  1647. page->next = discard_page;
  1648. discard_page = page;
  1649. }
  1650. }
  1651. if (n)
  1652. spin_unlock(&n->list_lock);
  1653. while (discard_page) {
  1654. page = discard_page;
  1655. discard_page = discard_page->next;
  1656. stat(s, DEACTIVATE_EMPTY);
  1657. discard_slab(s, page);
  1658. stat(s, FREE_SLAB);
  1659. }
  1660. }
  1661. /*
  1662. * Put a page that was just frozen (in __slab_free) into a partial page
  1663. * slot if available. This is done without interrupts disabled and without
  1664. * preemption disabled. The cmpxchg is racy and may put the partial page
  1665. * onto a random cpus partial slot.
  1666. *
  1667. * If we did not find a slot then simply move all the partials to the
  1668. * per node partial list.
  1669. */
  1670. int put_cpu_partial(struct kmem_cache *s, struct page *page, int drain)
  1671. {
  1672. struct page *oldpage;
  1673. int pages;
  1674. int pobjects;
  1675. do {
  1676. pages = 0;
  1677. pobjects = 0;
  1678. oldpage = this_cpu_read(s->cpu_slab->partial);
  1679. if (oldpage) {
  1680. pobjects = oldpage->pobjects;
  1681. pages = oldpage->pages;
  1682. if (drain && pobjects > s->cpu_partial) {
  1683. unsigned long flags;
  1684. /*
  1685. * partial array is full. Move the existing
  1686. * set to the per node partial list.
  1687. */
  1688. local_irq_save(flags);
  1689. unfreeze_partials(s);
  1690. local_irq_restore(flags);
  1691. pobjects = 0;
  1692. pages = 0;
  1693. stat(s, CPU_PARTIAL_DRAIN);
  1694. }
  1695. }
  1696. pages++;
  1697. pobjects += page->objects - page->inuse;
  1698. page->pages = pages;
  1699. page->pobjects = pobjects;
  1700. page->next = oldpage;
  1701. } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) != oldpage);
  1702. return pobjects;
  1703. }
  1704. static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
  1705. {
  1706. stat(s, CPUSLAB_FLUSH);
  1707. deactivate_slab(s, c->page, c->freelist);
  1708. c->tid = next_tid(c->tid);
  1709. c->page = NULL;
  1710. c->freelist = NULL;
  1711. }
  1712. /*
  1713. * Flush cpu slab.
  1714. *
  1715. * Called from IPI handler with interrupts disabled.
  1716. */
  1717. static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
  1718. {
  1719. struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
  1720. if (likely(c)) {
  1721. if (c->page)
  1722. flush_slab(s, c);
  1723. unfreeze_partials(s);
  1724. }
  1725. }
  1726. static void flush_cpu_slab(void *d)
  1727. {
  1728. struct kmem_cache *s = d;
  1729. __flush_cpu_slab(s, smp_processor_id());
  1730. }
  1731. static bool has_cpu_slab(int cpu, void *info)
  1732. {
  1733. struct kmem_cache *s = info;
  1734. struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
  1735. return c->page || c->partial;
  1736. }
  1737. static void flush_all(struct kmem_cache *s)
  1738. {
  1739. on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC);
  1740. }
  1741. /*
  1742. * Check if the objects in a per cpu structure fit numa
  1743. * locality expectations.
  1744. */
  1745. static inline int node_match(struct page *page, int node)
  1746. {
  1747. #ifdef CONFIG_NUMA
  1748. if (node != NUMA_NO_NODE && page_to_nid(page) != node)
  1749. return 0;
  1750. #endif
  1751. return 1;
  1752. }
  1753. static int count_free(struct page *page)
  1754. {
  1755. return page->objects - page->inuse;
  1756. }
  1757. static unsigned long count_partial(struct kmem_cache_node *n,
  1758. int (*get_count)(struct page *))
  1759. {
  1760. unsigned long flags;
  1761. unsigned long x = 0;
  1762. struct page *page;
  1763. spin_lock_irqsave(&n->list_lock, flags);
  1764. list_for_each_entry(page, &n->partial, lru)
  1765. x += get_count(page);
  1766. spin_unlock_irqrestore(&n->list_lock, flags);
  1767. return x;
  1768. }
  1769. static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
  1770. {
  1771. #ifdef CONFIG_SLUB_DEBUG
  1772. return atomic_long_read(&n->total_objects);
  1773. #else
  1774. return 0;
  1775. #endif
  1776. }
  1777. static noinline void
  1778. slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
  1779. {
  1780. int node;
  1781. printk(KERN_WARNING
  1782. "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
  1783. nid, gfpflags);
  1784. printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, "
  1785. "default order: %d, min order: %d\n", s->name, s->object_size,
  1786. s->size, oo_order(s->oo), oo_order(s->min));
  1787. if (oo_order(s->min) > get_order(s->object_size))
  1788. printk(KERN_WARNING " %s debugging increased min order, use "
  1789. "slub_debug=O to disable.\n", s->name);
  1790. for_each_online_node(node) {
  1791. struct kmem_cache_node *n = get_node(s, node);
  1792. unsigned long nr_slabs;
  1793. unsigned long nr_objs;
  1794. unsigned long nr_free;
  1795. if (!n)
  1796. continue;
  1797. nr_free = count_partial(n, count_free);
  1798. nr_slabs = node_nr_slabs(n);
  1799. nr_objs = node_nr_objs(n);
  1800. printk(KERN_WARNING
  1801. " node %d: slabs: %ld, objs: %ld, free: %ld\n",
  1802. node, nr_slabs, nr_objs, nr_free);
  1803. }
  1804. }
  1805. static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags,
  1806. int node, struct kmem_cache_cpu **pc)
  1807. {
  1808. void *freelist;
  1809. struct kmem_cache_cpu *c = *pc;
  1810. struct page *page;
  1811. freelist = get_partial(s, flags, node, c);
  1812. if (freelist)
  1813. return freelist;
  1814. page = new_slab(s, flags, node);
  1815. if (page) {
  1816. c = __this_cpu_ptr(s->cpu_slab);
  1817. if (c->page)
  1818. flush_slab(s, c);
  1819. /*
  1820. * No other reference to the page yet so we can
  1821. * muck around with it freely without cmpxchg
  1822. */
  1823. freelist = page->freelist;
  1824. page->freelist = NULL;
  1825. stat(s, ALLOC_SLAB);
  1826. c->page = page;
  1827. *pc = c;
  1828. } else
  1829. freelist = NULL;
  1830. return freelist;
  1831. }
  1832. /*
  1833. * Check the page->freelist of a page and either transfer the freelist to the per cpu freelist
  1834. * or deactivate the page.
  1835. *
  1836. * The page is still frozen if the return value is not NULL.
  1837. *
  1838. * If this function returns NULL then the page has been unfrozen.
  1839. *
  1840. * This function must be called with interrupt disabled.
  1841. */
  1842. static inline void *get_freelist(struct kmem_cache *s, struct page *page)
  1843. {
  1844. struct page new;
  1845. unsigned long counters;
  1846. void *freelist;
  1847. do {
  1848. freelist = page->freelist;
  1849. counters = page->counters;
  1850. new.counters = counters;
  1851. VM_BUG_ON(!new.frozen);
  1852. new.inuse = page->objects;
  1853. new.frozen = freelist != NULL;
  1854. } while (!__cmpxchg_double_slab(s, page,
  1855. freelist, counters,
  1856. NULL, new.counters,
  1857. "get_freelist"));
  1858. return freelist;
  1859. }
  1860. /*
  1861. * Slow path. The lockless freelist is empty or we need to perform
  1862. * debugging duties.
  1863. *
  1864. * Processing is still very fast if new objects have been freed to the
  1865. * regular freelist. In that case we simply take over the regular freelist
  1866. * as the lockless freelist and zap the regular freelist.
  1867. *
  1868. * If that is not working then we fall back to the partial lists. We take the
  1869. * first element of the freelist as the object to allocate now and move the
  1870. * rest of the freelist to the lockless freelist.
  1871. *
  1872. * And if we were unable to get a new slab from the partial slab lists then
  1873. * we need to allocate a new slab. This is the slowest path since it involves
  1874. * a call to the page allocator and the setup of a new slab.
  1875. */
  1876. static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
  1877. unsigned long addr, struct kmem_cache_cpu *c)
  1878. {
  1879. void *freelist;
  1880. struct page *page;
  1881. unsigned long flags;
  1882. local_irq_save(flags);
  1883. #ifdef CONFIG_PREEMPT
  1884. /*
  1885. * We may have been preempted and rescheduled on a different
  1886. * cpu before disabling interrupts. Need to reload cpu area
  1887. * pointer.
  1888. */
  1889. c = this_cpu_ptr(s->cpu_slab);
  1890. #endif
  1891. page = c->page;
  1892. if (!page)
  1893. goto new_slab;
  1894. redo:
  1895. if (unlikely(!node_match(page, node))) {
  1896. stat(s, ALLOC_NODE_MISMATCH);
  1897. deactivate_slab(s, page, c->freelist);
  1898. c->page = NULL;
  1899. c->freelist = NULL;
  1900. goto new_slab;
  1901. }
  1902. /* must check again c->freelist in case of cpu migration or IRQ */
  1903. freelist = c->freelist;
  1904. if (freelist)
  1905. goto load_freelist;
  1906. stat(s, ALLOC_SLOWPATH);
  1907. freelist = get_freelist(s, page);
  1908. if (!freelist) {
  1909. c->page = NULL;
  1910. stat(s, DEACTIVATE_BYPASS);
  1911. goto new_slab;
  1912. }
  1913. stat(s, ALLOC_REFILL);
  1914. load_freelist:
  1915. /*
  1916. * freelist is pointing to the list of objects to be used.
  1917. * page is pointing to the page from which the objects are obtained.
  1918. * That page must be frozen for per cpu allocations to work.
  1919. */
  1920. VM_BUG_ON(!c->page->frozen);
  1921. c->freelist = get_freepointer(s, freelist);
  1922. c->tid = next_tid(c->tid);
  1923. local_irq_restore(flags);
  1924. return freelist;
  1925. new_slab:
  1926. if (c->partial) {
  1927. page = c->page = c->partial;
  1928. c->partial = page->next;
  1929. stat(s, CPU_PARTIAL_ALLOC);
  1930. c->freelist = NULL;
  1931. goto redo;
  1932. }
  1933. freelist = new_slab_objects(s, gfpflags, node, &c);
  1934. if (unlikely(!freelist)) {
  1935. if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit())
  1936. slab_out_of_memory(s, gfpflags, node);
  1937. local_irq_restore(flags);
  1938. return NULL;
  1939. }
  1940. page = c->page;
  1941. if (likely(!kmem_cache_debug(s)))
  1942. goto load_freelist;
  1943. /* Only entered in the debug case */
  1944. if (!alloc_debug_processing(s, page, freelist, addr))
  1945. goto new_slab; /* Slab failed checks. Next slab needed */
  1946. deactivate_slab(s, page, get_freepointer(s, freelist));
  1947. c->page = NULL;
  1948. c->freelist = NULL;
  1949. local_irq_restore(flags);
  1950. return freelist;
  1951. }
  1952. /*
  1953. * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
  1954. * have the fastpath folded into their functions. So no function call
  1955. * overhead for requests that can be satisfied on the fastpath.
  1956. *
  1957. * The fastpath works by first checking if the lockless freelist can be used.
  1958. * If not then __slab_alloc is called for slow processing.
  1959. *
  1960. * Otherwise we can simply pick the next object from the lockless free list.
  1961. */
  1962. static __always_inline void *slab_alloc(struct kmem_cache *s,
  1963. gfp_t gfpflags, int node, unsigned long addr)
  1964. {
  1965. void **object;
  1966. struct kmem_cache_cpu *c;
  1967. struct page *page;
  1968. unsigned long tid;
  1969. if (slab_pre_alloc_hook(s, gfpflags))
  1970. return NULL;
  1971. redo:
  1972. /*
  1973. * Must read kmem_cache cpu data via this cpu ptr. Preemption is
  1974. * enabled. We may switch back and forth between cpus while
  1975. * reading from one cpu area. That does not matter as long
  1976. * as we end up on the original cpu again when doing the cmpxchg.
  1977. */
  1978. c = __this_cpu_ptr(s->cpu_slab);
  1979. /*
  1980. * The transaction ids are globally unique per cpu and per operation on
  1981. * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
  1982. * occurs on the right processor and that there was no operation on the
  1983. * linked list in between.
  1984. */
  1985. tid = c->tid;
  1986. barrier();
  1987. object = c->freelist;
  1988. page = c->page;
  1989. if (unlikely(!object || !node_match(page, node)))
  1990. object = __slab_alloc(s, gfpflags, node, addr, c);
  1991. else {
  1992. void *next_object = get_freepointer_safe(s, object);
  1993. /*
  1994. * The cmpxchg will only match if there was no additional
  1995. * operation and if we are on the right processor.
  1996. *
  1997. * The cmpxchg does the following atomically (without lock semantics!)
  1998. * 1. Relocate first pointer to the current per cpu area.
  1999. * 2. Verify that tid and freelist have not been changed
  2000. * 3. If they were not changed replace tid and freelist
  2001. *
  2002. * Since this is without lock semantics the protection is only against
  2003. * code executing on this cpu *not* from access by other cpus.
  2004. */
  2005. if (unlikely(!this_cpu_cmpxchg_double(
  2006. s->cpu_slab->freelist, s->cpu_slab->tid,
  2007. object, tid,
  2008. next_object, next_tid(tid)))) {
  2009. note_cmpxchg_failure("slab_alloc", s, tid);
  2010. goto redo;
  2011. }
  2012. prefetch_freepointer(s, next_object);
  2013. stat(s, ALLOC_FASTPATH);
  2014. }
  2015. if (unlikely(gfpflags & __GFP_ZERO) && object)
  2016. memset(object, 0, s->object_size);
  2017. slab_post_alloc_hook(s, gfpflags, object);
  2018. return object;
  2019. }
  2020. void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
  2021. {
  2022. void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_);
  2023. trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size, s->size, gfpflags);
  2024. return ret;
  2025. }
  2026. EXPORT_SYMBOL(kmem_cache_alloc);
  2027. #ifdef CONFIG_TRACING
  2028. void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
  2029. {
  2030. void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_);
  2031. trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags);
  2032. return ret;
  2033. }
  2034. EXPORT_SYMBOL(kmem_cache_alloc_trace);
  2035. void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
  2036. {
  2037. void *ret = kmalloc_order(size, flags, order);
  2038. trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
  2039. return ret;
  2040. }
  2041. EXPORT_SYMBOL(kmalloc_order_trace);
  2042. #endif
  2043. #ifdef CONFIG_NUMA
  2044. void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
  2045. {
  2046. void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
  2047. trace_kmem_cache_alloc_node(_RET_IP_, ret,
  2048. s->object_size, s->size, gfpflags, node);
  2049. return ret;
  2050. }
  2051. EXPORT_SYMBOL(kmem_cache_alloc_node);
  2052. #ifdef CONFIG_TRACING
  2053. void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
  2054. gfp_t gfpflags,
  2055. int node, size_t size)
  2056. {
  2057. void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
  2058. trace_kmalloc_node(_RET_IP_, ret,
  2059. size, s->size, gfpflags, node);
  2060. return ret;
  2061. }
  2062. EXPORT_SYMBOL(kmem_cache_alloc_node_trace);
  2063. #endif
  2064. #endif
  2065. /*
  2066. * Slow patch handling. This may still be called frequently since objects
  2067. * have a longer lifetime than the cpu slabs in most processing loads.
  2068. *
  2069. * So we still attempt to reduce cache line usage. Just take the slab
  2070. * lock and free the item. If there is no additional partial page
  2071. * handling required then we can return immediately.
  2072. */
  2073. static void __slab_free(struct kmem_cache *s, struct page *page,
  2074. void *x, unsigned long addr)
  2075. {
  2076. void *prior;
  2077. void **object = (void *)x;
  2078. int was_frozen;
  2079. int inuse;
  2080. struct page new;
  2081. unsigned long counters;
  2082. struct kmem_cache_node *n = NULL;
  2083. unsigned long uninitialized_var(flags);
  2084. stat(s, FREE_SLOWPATH);
  2085. if (kmem_cache_debug(s) && !free_debug_processing(s, page, x, addr))
  2086. return;
  2087. do {
  2088. prior = page->freelist;
  2089. counters = page->counters;
  2090. set_freepointer(s, object, prior);
  2091. new.counters = counters;
  2092. was_frozen = new.frozen;
  2093. new.inuse--;
  2094. if ((!new.inuse || !prior) && !was_frozen && !n) {
  2095. if (!kmem_cache_debug(s) && !prior)
  2096. /*
  2097. * Slab was on no list before and will be partially empty
  2098. * We can defer the list move and instead freeze it.
  2099. */
  2100. new.frozen = 1;
  2101. else { /* Needs to be taken off a list */
  2102. n = get_node(s, page_to_nid(page));
  2103. /*
  2104. * Speculatively acquire the list_lock.
  2105. * If the cmpxchg does not succeed then we may
  2106. * drop the list_lock without any processing.
  2107. *
  2108. * Otherwise the list_lock will synchronize with
  2109. * other processors updating the list of slabs.
  2110. */
  2111. spin_lock_irqsave(&n->list_lock, flags);
  2112. }
  2113. }
  2114. inuse = new.inuse;
  2115. } while (!cmpxchg_double_slab(s, page,
  2116. prior, counters,
  2117. object, new.counters,
  2118. "__slab_free"));
  2119. if (likely(!n)) {
  2120. /*
  2121. * If we just froze the page then put it onto the
  2122. * per cpu partial list.
  2123. */
  2124. if (new.frozen && !was_frozen) {
  2125. put_cpu_partial(s, page, 1);
  2126. stat(s, CPU_PARTIAL_FREE);
  2127. }
  2128. /*
  2129. * The list lock was not taken therefore no list
  2130. * activity can be necessary.
  2131. */
  2132. if (was_frozen)
  2133. stat(s, FREE_FROZEN);
  2134. return;
  2135. }
  2136. /*
  2137. * was_frozen may have been set after we acquired the list_lock in
  2138. * an earlier loop. So we need to check it here again.
  2139. */
  2140. if (was_frozen)
  2141. stat(s, FREE_FROZEN);
  2142. else {
  2143. if (unlikely(!inuse && n->nr_partial > s->min_partial))
  2144. goto slab_empty;
  2145. /*
  2146. * Objects left in the slab. If it was not on the partial list before
  2147. * then add it.
  2148. */
  2149. if (unlikely(!prior)) {
  2150. remove_full(s, page);
  2151. add_partial(n, page, DEACTIVATE_TO_TAIL);
  2152. stat(s, FREE_ADD_PARTIAL);
  2153. }
  2154. }
  2155. spin_unlock_irqrestore(&n->list_lock, flags);
  2156. return;
  2157. slab_empty:
  2158. if (prior) {
  2159. /*
  2160. * Slab on the partial list.
  2161. */
  2162. remove_partial(n, page);
  2163. stat(s, FREE_REMOVE_PARTIAL);
  2164. } else
  2165. /* Slab must be on the full list */
  2166. remove_full(s, page);
  2167. spin_unlock_irqrestore(&n->list_lock, flags);
  2168. stat(s, FREE_SLAB);
  2169. discard_slab(s, page);
  2170. }
  2171. /*
  2172. * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
  2173. * can perform fastpath freeing without additional function calls.
  2174. *
  2175. * The fastpath is only possible if we are freeing to the current cpu slab
  2176. * of this processor. This typically the case if we have just allocated
  2177. * the item before.
  2178. *
  2179. * If fastpath is not possible then fall back to __slab_free where we deal
  2180. * with all sorts of special processing.
  2181. */
  2182. static __always_inline void slab_free(struct kmem_cache *s,
  2183. struct page *page, void *x, unsigned long addr)
  2184. {
  2185. void **object = (void *)x;
  2186. struct kmem_cache_cpu *c;
  2187. unsigned long tid;
  2188. slab_free_hook(s, x);
  2189. redo:
  2190. /*
  2191. * Determine the currently cpus per cpu slab.
  2192. * The cpu may change afterward. However that does not matter since
  2193. * data is retrieved via this pointer. If we are on the same cpu
  2194. * during the cmpxchg then the free will succedd.
  2195. */
  2196. c = __this_cpu_ptr(s->cpu_slab);
  2197. tid = c->tid;
  2198. barrier();
  2199. if (likely(page == c->page)) {
  2200. set_freepointer(s, object, c->freelist);
  2201. if (unlikely(!this_cpu_cmpxchg_double(
  2202. s->cpu_slab->freelist, s->cpu_slab->tid,
  2203. c->freelist, tid,
  2204. object, next_tid(tid)))) {
  2205. note_cmpxchg_failure("slab_free", s, tid);
  2206. goto redo;
  2207. }
  2208. stat(s, FREE_FASTPATH);
  2209. } else
  2210. __slab_free(s, page, x, addr);
  2211. }
  2212. void kmem_cache_free(struct kmem_cache *s, void *x)
  2213. {
  2214. struct page *page;
  2215. page = virt_to_head_page(x);
  2216. slab_free(s, page, x, _RET_IP_);
  2217. trace_kmem_cache_free(_RET_IP_, x);
  2218. }
  2219. EXPORT_SYMBOL(kmem_cache_free);
  2220. /*
  2221. * Object placement in a slab is made very easy because we always start at
  2222. * offset 0. If we tune the size of the object to the alignment then we can
  2223. * get the required alignment by putting one properly sized object after
  2224. * another.
  2225. *
  2226. * Notice that the allocation order determines the sizes of the per cpu
  2227. * caches. Each processor has always one slab available for allocations.
  2228. * Increasing the allocation order reduces the number of times that slabs
  2229. * must be moved on and off the partial lists and is therefore a factor in
  2230. * locking overhead.
  2231. */
  2232. /*
  2233. * Mininum / Maximum order of slab pages. This influences locking overhead
  2234. * and slab fragmentation. A higher order reduces the number of partial slabs
  2235. * and increases the number of allocations possible without having to
  2236. * take the list_lock.
  2237. */
  2238. static int slub_min_order;
  2239. static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
  2240. static int slub_min_objects;
  2241. /*
  2242. * Merge control. If this is set then no merging of slab caches will occur.
  2243. * (Could be removed. This was introduced to pacify the merge skeptics.)
  2244. */
  2245. static int slub_nomerge;
  2246. /*
  2247. * Calculate the order of allocation given an slab object size.
  2248. *
  2249. * The order of allocation has significant impact on performance and other
  2250. * system components. Generally order 0 allocations should be preferred since
  2251. * order 0 does not cause fragmentation in the page allocator. Larger objects
  2252. * be problematic to put into order 0 slabs because there may be too much
  2253. * unused space left. We go to a higher order if more than 1/16th of the slab
  2254. * would be wasted.
  2255. *
  2256. * In order to reach satisfactory performance we must ensure that a minimum
  2257. * number of objects is in one slab. Otherwise we may generate too much
  2258. * activity on the partial lists which requires taking the list_lock. This is
  2259. * less a concern for large slabs though which are rarely used.
  2260. *
  2261. * slub_max_order specifies the order where we begin to stop considering the
  2262. * number of objects in a slab as critical. If we reach slub_max_order then
  2263. * we try to keep the page order as low as possible. So we accept more waste
  2264. * of space in favor of a small page order.
  2265. *
  2266. * Higher order allocations also allow the placement of more objects in a
  2267. * slab and thereby reduce object handling overhead. If the user has
  2268. * requested a higher mininum order then we start with that one instead of
  2269. * the smallest order which will fit the object.
  2270. */
  2271. static inline int slab_order(int size, int min_objects,
  2272. int max_order, int fract_leftover, int reserved)
  2273. {
  2274. int order;
  2275. int rem;
  2276. int min_order = slub_min_order;
  2277. if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE)
  2278. return get_order(size * MAX_OBJS_PER_PAGE) - 1;
  2279. for (order = max(min_order,
  2280. fls(min_objects * size - 1) - PAGE_SHIFT);
  2281. order <= max_order; order++) {
  2282. unsigned long slab_size = PAGE_SIZE << order;
  2283. if (slab_size < min_objects * size + reserved)
  2284. continue;
  2285. rem = (slab_size - reserved) % size;
  2286. if (rem <= slab_size / fract_leftover)
  2287. break;
  2288. }
  2289. return order;
  2290. }
  2291. static inline int calculate_order(int size, int reserved)
  2292. {
  2293. int order;
  2294. int min_objects;
  2295. int fraction;
  2296. int max_objects;
  2297. /*
  2298. * Attempt to find best configuration for a slab. This
  2299. * works by first attempting to generate a layout with
  2300. * the best configuration and backing off gradually.
  2301. *
  2302. * First we reduce the acceptable waste in a slab. Then
  2303. * we reduce the minimum objects required in a slab.
  2304. */
  2305. min_objects = slub_min_objects;
  2306. if (!min_objects)
  2307. min_objects = 4 * (fls(nr_cpu_ids) + 1);
  2308. max_objects = order_objects(slub_max_order, size, reserved);
  2309. min_objects = min(min_objects, max_objects);
  2310. while (min_objects > 1) {
  2311. fraction = 16;
  2312. while (fraction >= 4) {
  2313. order = slab_order(size, min_objects,
  2314. slub_max_order, fraction, reserved);
  2315. if (order <= slub_max_order)
  2316. return order;
  2317. fraction /= 2;
  2318. }
  2319. min_objects--;
  2320. }
  2321. /*
  2322. * We were unable to place multiple objects in a slab. Now
  2323. * lets see if we can place a single object there.
  2324. */
  2325. order = slab_order(size, 1, slub_max_order, 1, reserved);
  2326. if (order <= slub_max_order)
  2327. return order;
  2328. /*
  2329. * Doh this slab cannot be placed using slub_max_order.
  2330. */
  2331. order = slab_order(size, 1, MAX_ORDER, 1, reserved);
  2332. if (order < MAX_ORDER)
  2333. return order;
  2334. return -ENOSYS;
  2335. }
  2336. /*
  2337. * Figure out what the alignment of the objects will be.
  2338. */
  2339. static unsigned long calculate_alignment(unsigned long flags,
  2340. unsigned long align, unsigned long size)
  2341. {
  2342. /*
  2343. * If the user wants hardware cache aligned objects then follow that
  2344. * suggestion if the object is sufficiently large.
  2345. *
  2346. * The hardware cache alignment cannot override the specified
  2347. * alignment though. If that is greater then use it.
  2348. */
  2349. if (flags & SLAB_HWCACHE_ALIGN) {
  2350. unsigned long ralign = cache_line_size();
  2351. while (size <= ralign / 2)
  2352. ralign /= 2;
  2353. align = max(align, ralign);
  2354. }
  2355. if (align < ARCH_SLAB_MINALIGN)
  2356. align = ARCH_SLAB_MINALIGN;
  2357. return ALIGN(align, sizeof(void *));
  2358. }
  2359. static void
  2360. init_kmem_cache_node(struct kmem_cache_node *n)
  2361. {
  2362. n->nr_partial = 0;
  2363. spin_lock_init(&n->list_lock);
  2364. INIT_LIST_HEAD(&n->partial);
  2365. #ifdef CONFIG_SLUB_DEBUG
  2366. atomic_long_set(&n->nr_slabs, 0);
  2367. atomic_long_set(&n->total_objects, 0);
  2368. INIT_LIST_HEAD(&n->full);
  2369. #endif
  2370. }
  2371. static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
  2372. {
  2373. BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
  2374. SLUB_PAGE_SHIFT * sizeof(struct kmem_cache_cpu));
  2375. /*
  2376. * Must align to double word boundary for the double cmpxchg
  2377. * instructions to work; see __pcpu_double_call_return_bool().
  2378. */
  2379. s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu),
  2380. 2 * sizeof(void *));
  2381. if (!s->cpu_slab)
  2382. return 0;
  2383. init_kmem_cache_cpus(s);
  2384. return 1;
  2385. }
  2386. static struct kmem_cache *kmem_cache_node;
  2387. /*
  2388. * No kmalloc_node yet so do it by hand. We know that this is the first
  2389. * slab on the node for this slabcache. There are no concurrent accesses
  2390. * possible.
  2391. *
  2392. * Note that this function only works on the kmalloc_node_cache
  2393. * when allocating for the kmalloc_node_cache. This is used for bootstrapping
  2394. * memory on a fresh node that has no slab structures yet.
  2395. */
  2396. static void early_kmem_cache_node_alloc(int node)
  2397. {
  2398. struct page *page;
  2399. struct kmem_cache_node *n;
  2400. BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node));
  2401. page = new_slab(kmem_cache_node, GFP_NOWAIT, node);
  2402. BUG_ON(!page);
  2403. if (page_to_nid(page) != node) {
  2404. printk(KERN_ERR "SLUB: Unable to allocate memory from "
  2405. "node %d\n", node);
  2406. printk(KERN_ERR "SLUB: Allocating a useless per node structure "
  2407. "in order to be able to continue\n");
  2408. }
  2409. n = page->freelist;
  2410. BUG_ON(!n);
  2411. page->freelist = get_freepointer(kmem_cache_node, n);
  2412. page->inuse = 1;
  2413. page->frozen = 0;
  2414. kmem_cache_node->node[node] = n;
  2415. #ifdef CONFIG_SLUB_DEBUG
  2416. init_object(kmem_cache_node, n, SLUB_RED_ACTIVE);
  2417. init_tracking(kmem_cache_node, n);
  2418. #endif
  2419. init_kmem_cache_node(n);
  2420. inc_slabs_node(kmem_cache_node, node, page->objects);
  2421. add_partial(n, page, DEACTIVATE_TO_HEAD);
  2422. }
  2423. static void free_kmem_cache_nodes(struct kmem_cache *s)
  2424. {
  2425. int node;
  2426. for_each_node_state(node, N_NORMAL_MEMORY) {
  2427. struct kmem_cache_node *n = s->node[node];
  2428. if (n)
  2429. kmem_cache_free(kmem_cache_node, n);
  2430. s->node[node] = NULL;
  2431. }
  2432. }
  2433. static int init_kmem_cache_nodes(struct kmem_cache *s)
  2434. {
  2435. int node;
  2436. for_each_node_state(node, N_NORMAL_MEMORY) {
  2437. struct kmem_cache_node *n;
  2438. if (slab_state == DOWN) {
  2439. early_kmem_cache_node_alloc(node);
  2440. continue;
  2441. }
  2442. n = kmem_cache_alloc_node(kmem_cache_node,
  2443. GFP_KERNEL, node);
  2444. if (!n) {
  2445. free_kmem_cache_nodes(s);
  2446. return 0;
  2447. }
  2448. s->node[node] = n;
  2449. init_kmem_cache_node(n);
  2450. }
  2451. return 1;
  2452. }
  2453. static void set_min_partial(struct kmem_cache *s, unsigned long min)
  2454. {
  2455. if (min < MIN_PARTIAL)
  2456. min = MIN_PARTIAL;
  2457. else if (min > MAX_PARTIAL)
  2458. min = MAX_PARTIAL;
  2459. s->min_partial = min;
  2460. }
  2461. /*
  2462. * calculate_sizes() determines the order and the distribution of data within
  2463. * a slab object.
  2464. */
  2465. static int calculate_sizes(struct kmem_cache *s, int forced_order)
  2466. {
  2467. unsigned long flags = s->flags;
  2468. unsigned long size = s->object_size;
  2469. unsigned long align = s->align;
  2470. int order;
  2471. /*
  2472. * Round up object size to the next word boundary. We can only
  2473. * place the free pointer at word boundaries and this determines
  2474. * the possible location of the free pointer.
  2475. */
  2476. size = ALIGN(size, sizeof(void *));
  2477. #ifdef CONFIG_SLUB_DEBUG
  2478. /*
  2479. * Determine if we can poison the object itself. If the user of
  2480. * the slab may touch the object after free or before allocation
  2481. * then we should never poison the object itself.
  2482. */
  2483. if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) &&
  2484. !s->ctor)
  2485. s->flags |= __OBJECT_POISON;
  2486. else
  2487. s->flags &= ~__OBJECT_POISON;
  2488. /*
  2489. * If we are Redzoning then check if there is some space between the
  2490. * end of the object and the free pointer. If not then add an
  2491. * additional word to have some bytes to store Redzone information.
  2492. */
  2493. if ((flags & SLAB_RED_ZONE) && size == s->object_size)
  2494. size += sizeof(void *);
  2495. #endif
  2496. /*
  2497. * With that we have determined the number of bytes in actual use
  2498. * by the object. This is the potential offset to the free pointer.
  2499. */
  2500. s->inuse = size;
  2501. if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) ||
  2502. s->ctor)) {
  2503. /*
  2504. * Relocate free pointer after the object if it is not
  2505. * permitted to overwrite the first word of the object on
  2506. * kmem_cache_free.
  2507. *
  2508. * This is the case if we do RCU, have a constructor or
  2509. * destructor or are poisoning the objects.
  2510. */
  2511. s->offset = size;
  2512. size += sizeof(void *);
  2513. }
  2514. #ifdef CONFIG_SLUB_DEBUG
  2515. if (flags & SLAB_STORE_USER)
  2516. /*
  2517. * Need to store information about allocs and frees after
  2518. * the object.
  2519. */
  2520. size += 2 * sizeof(struct track);
  2521. if (flags & SLAB_RED_ZONE)
  2522. /*
  2523. * Add some empty padding so that we can catch
  2524. * overwrites from earlier objects rather than let
  2525. * tracking information or the free pointer be
  2526. * corrupted if a user writes before the start
  2527. * of the object.
  2528. */
  2529. size += sizeof(void *);
  2530. #endif
  2531. /*
  2532. * Determine the alignment based on various parameters that the
  2533. * user specified and the dynamic determination of cache line size
  2534. * on bootup.
  2535. */
  2536. align = calculate_alignment(flags, align, s->object_size);
  2537. s->align = align;
  2538. /*
  2539. * SLUB stores one object immediately after another beginning from
  2540. * offset 0. In order to align the objects we have to simply size
  2541. * each object to conform to the alignment.
  2542. */
  2543. size = ALIGN(size, align);
  2544. s->size = size;
  2545. if (forced_order >= 0)
  2546. order = forced_order;
  2547. else
  2548. order = calculate_order(size, s->reserved);
  2549. if (order < 0)
  2550. return 0;
  2551. s->allocflags = 0;
  2552. if (order)
  2553. s->allocflags |= __GFP_COMP;
  2554. if (s->flags & SLAB_CACHE_DMA)
  2555. s->allocflags |= SLUB_DMA;
  2556. if (s->flags & SLAB_RECLAIM_ACCOUNT)
  2557. s->allocflags |= __GFP_RECLAIMABLE;
  2558. /*
  2559. * Determine the number of objects per slab
  2560. */
  2561. s->oo = oo_make(order, size, s->reserved);
  2562. s->min = oo_make(get_order(size), size, s->reserved);
  2563. if (oo_objects(s->oo) > oo_objects(s->max))
  2564. s->max = s->oo;
  2565. return !!oo_objects(s->oo);
  2566. }
  2567. static int kmem_cache_open(struct kmem_cache *s,
  2568. const char *name, size_t size,
  2569. size_t align, unsigned long flags,
  2570. void (*ctor)(void *))
  2571. {
  2572. memset(s, 0, kmem_size);
  2573. s->name = name;
  2574. s->ctor = ctor;
  2575. s->object_size = size;
  2576. s->align = align;
  2577. s->flags = kmem_cache_flags(size, flags, name, ctor);
  2578. s->reserved = 0;
  2579. if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU))
  2580. s->reserved = sizeof(struct rcu_head);
  2581. if (!calculate_sizes(s, -1))
  2582. goto error;
  2583. if (disable_higher_order_debug) {
  2584. /*
  2585. * Disable debugging flags that store metadata if the min slab
  2586. * order increased.
  2587. */
  2588. if (get_order(s->size) > get_order(s->object_size)) {
  2589. s->flags &= ~DEBUG_METADATA_FLAGS;
  2590. s->offset = 0;
  2591. if (!calculate_sizes(s, -1))
  2592. goto error;
  2593. }
  2594. }
  2595. #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
  2596. defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
  2597. if (system_has_cmpxchg_double() && (s->flags & SLAB_DEBUG_FLAGS) == 0)
  2598. /* Enable fast mode */
  2599. s->flags |= __CMPXCHG_DOUBLE;
  2600. #endif
  2601. /*
  2602. * The larger the object size is, the more pages we want on the partial
  2603. * list to avoid pounding the page allocator excessively.
  2604. */
  2605. set_min_partial(s, ilog2(s->size) / 2);
  2606. /*
  2607. * cpu_partial determined the maximum number of objects kept in the
  2608. * per cpu partial lists of a processor.
  2609. *
  2610. * Per cpu partial lists mainly contain slabs that just have one
  2611. * object freed. If they are used for allocation then they can be
  2612. * filled up again with minimal effort. The slab will never hit the
  2613. * per node partial lists and therefore no locking will be required.
  2614. *
  2615. * This setting also determines
  2616. *
  2617. * A) The number of objects from per cpu partial slabs dumped to the
  2618. * per node list when we reach the limit.
  2619. * B) The number of objects in cpu partial slabs to extract from the
  2620. * per node list when we run out of per cpu objects. We only fetch 50%
  2621. * to keep some capacity around for frees.
  2622. */
  2623. if (kmem_cache_debug(s))
  2624. s->cpu_partial = 0;
  2625. else if (s->size >= PAGE_SIZE)
  2626. s->cpu_partial = 2;
  2627. else if (s->size >= 1024)
  2628. s->cpu_partial = 6;
  2629. else if (s->size >= 256)
  2630. s->cpu_partial = 13;
  2631. else
  2632. s->cpu_partial = 30;
  2633. s->refcount = 1;
  2634. #ifdef CONFIG_NUMA
  2635. s->remote_node_defrag_ratio = 1000;
  2636. #endif
  2637. if (!init_kmem_cache_nodes(s))
  2638. goto error;
  2639. if (alloc_kmem_cache_cpus(s))
  2640. return 1;
  2641. free_kmem_cache_nodes(s);
  2642. error:
  2643. if (flags & SLAB_PANIC)
  2644. panic("Cannot create slab %s size=%lu realsize=%u "
  2645. "order=%u offset=%u flags=%lx\n",
  2646. s->name, (unsigned long)size, s->size, oo_order(s->oo),
  2647. s->offset, flags);
  2648. return 0;
  2649. }
  2650. /*
  2651. * Determine the size of a slab object
  2652. */
  2653. unsigned int kmem_cache_size(struct kmem_cache *s)
  2654. {
  2655. return s->object_size;
  2656. }
  2657. EXPORT_SYMBOL(kmem_cache_size);
  2658. static void list_slab_objects(struct kmem_cache *s, struct page *page,
  2659. const char *text)
  2660. {
  2661. #ifdef CONFIG_SLUB_DEBUG
  2662. void *addr = page_address(page);
  2663. void *p;
  2664. unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) *
  2665. sizeof(long), GFP_ATOMIC);
  2666. if (!map)
  2667. return;
  2668. slab_err(s, page, "%s", text);
  2669. slab_lock(page);
  2670. get_map(s, page, map);
  2671. for_each_object(p, s, addr, page->objects) {
  2672. if (!test_bit(slab_index(p, s, addr), map)) {
  2673. printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n",
  2674. p, p - addr);
  2675. print_tracking(s, p);
  2676. }
  2677. }
  2678. slab_unlock(page);
  2679. kfree(map);
  2680. #endif
  2681. }
  2682. /*
  2683. * Attempt to free all partial slabs on a node.
  2684. * This is called from kmem_cache_close(). We must be the last thread
  2685. * using the cache and therefore we do not need to lock anymore.
  2686. */
  2687. static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
  2688. {
  2689. struct page *page, *h;
  2690. list_for_each_entry_safe(page, h, &n->partial, lru) {
  2691. if (!page->inuse) {
  2692. remove_partial(n, page);
  2693. discard_slab(s, page);
  2694. } else {
  2695. list_slab_objects(s, page,
  2696. "Objects remaining on kmem_cache_close()");
  2697. }
  2698. }
  2699. }
  2700. /*
  2701. * Release all resources used by a slab cache.
  2702. */
  2703. static inline int kmem_cache_close(struct kmem_cache *s)
  2704. {
  2705. int node;
  2706. flush_all(s);
  2707. free_percpu(s->cpu_slab);
  2708. /* Attempt to free all objects */
  2709. for_each_node_state(node, N_NORMAL_MEMORY) {
  2710. struct kmem_cache_node *n = get_node(s, node);
  2711. free_partial(s, n);
  2712. if (n->nr_partial || slabs_node(s, node))
  2713. return 1;
  2714. }
  2715. free_kmem_cache_nodes(s);
  2716. return 0;
  2717. }
  2718. /*
  2719. * Close a cache and release the kmem_cache structure
  2720. * (must be used for caches created using kmem_cache_create)
  2721. */
  2722. void kmem_cache_destroy(struct kmem_cache *s)
  2723. {
  2724. down_write(&slub_lock);
  2725. s->refcount--;
  2726. if (!s->refcount) {
  2727. list_del(&s->list);
  2728. up_write(&slub_lock);
  2729. if (kmem_cache_close(s)) {
  2730. printk(KERN_ERR "SLUB %s: %s called for cache that "
  2731. "still has objects.\n", s->name, __func__);
  2732. dump_stack();
  2733. }
  2734. if (s->flags & SLAB_DESTROY_BY_RCU)
  2735. rcu_barrier();
  2736. sysfs_slab_remove(s);
  2737. } else
  2738. up_write(&slub_lock);
  2739. }
  2740. EXPORT_SYMBOL(kmem_cache_destroy);
  2741. /********************************************************************
  2742. * Kmalloc subsystem
  2743. *******************************************************************/
  2744. struct kmem_cache *kmalloc_caches[SLUB_PAGE_SHIFT];
  2745. EXPORT_SYMBOL(kmalloc_caches);
  2746. static struct kmem_cache *kmem_cache;
  2747. #ifdef CONFIG_ZONE_DMA
  2748. static struct kmem_cache *kmalloc_dma_caches[SLUB_PAGE_SHIFT];
  2749. #endif
  2750. static int __init setup_slub_min_order(char *str)
  2751. {
  2752. get_option(&str, &slub_min_order);
  2753. return 1;
  2754. }
  2755. __setup("slub_min_order=", setup_slub_min_order);
  2756. static int __init setup_slub_max_order(char *str)
  2757. {
  2758. get_option(&str, &slub_max_order);
  2759. slub_max_order = min(slub_max_order, MAX_ORDER - 1);
  2760. return 1;
  2761. }
  2762. __setup("slub_max_order=", setup_slub_max_order);
  2763. static int __init setup_slub_min_objects(char *str)
  2764. {
  2765. get_option(&str, &slub_min_objects);
  2766. return 1;
  2767. }
  2768. __setup("slub_min_objects=", setup_slub_min_objects);
  2769. static int __init setup_slub_nomerge(char *str)
  2770. {
  2771. slub_nomerge = 1;
  2772. return 1;
  2773. }
  2774. __setup("slub_nomerge", setup_slub_nomerge);
  2775. static struct kmem_cache *__init create_kmalloc_cache(const char *name,
  2776. int size, unsigned int flags)
  2777. {
  2778. struct kmem_cache *s;
  2779. s = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
  2780. /*
  2781. * This function is called with IRQs disabled during early-boot on
  2782. * single CPU so there's no need to take slub_lock here.
  2783. */
  2784. if (!kmem_cache_open(s, name, size, ARCH_KMALLOC_MINALIGN,
  2785. flags, NULL))
  2786. goto panic;
  2787. list_add(&s->list, &slab_caches);
  2788. return s;
  2789. panic:
  2790. panic("Creation of kmalloc slab %s size=%d failed.\n", name, size);
  2791. return NULL;
  2792. }
  2793. /*
  2794. * Conversion table for small slabs sizes / 8 to the index in the
  2795. * kmalloc array. This is necessary for slabs < 192 since we have non power
  2796. * of two cache sizes there. The size of larger slabs can be determined using
  2797. * fls.
  2798. */
  2799. static s8 size_index[24] = {
  2800. 3, /* 8 */
  2801. 4, /* 16 */
  2802. 5, /* 24 */
  2803. 5, /* 32 */
  2804. 6, /* 40 */
  2805. 6, /* 48 */
  2806. 6, /* 56 */
  2807. 6, /* 64 */
  2808. 1, /* 72 */
  2809. 1, /* 80 */
  2810. 1, /* 88 */
  2811. 1, /* 96 */
  2812. 7, /* 104 */
  2813. 7, /* 112 */
  2814. 7, /* 120 */
  2815. 7, /* 128 */
  2816. 2, /* 136 */
  2817. 2, /* 144 */
  2818. 2, /* 152 */
  2819. 2, /* 160 */
  2820. 2, /* 168 */
  2821. 2, /* 176 */
  2822. 2, /* 184 */
  2823. 2 /* 192 */
  2824. };
  2825. static inline int size_index_elem(size_t bytes)
  2826. {
  2827. return (bytes - 1) / 8;
  2828. }
  2829. static struct kmem_cache *get_slab(size_t size, gfp_t flags)
  2830. {
  2831. int index;
  2832. if (size <= 192) {
  2833. if (!size)
  2834. return ZERO_SIZE_PTR;
  2835. index = size_index[size_index_elem(size)];
  2836. } else
  2837. index = fls(size - 1);
  2838. #ifdef CONFIG_ZONE_DMA
  2839. if (unlikely((flags & SLUB_DMA)))
  2840. return kmalloc_dma_caches[index];
  2841. #endif
  2842. return kmalloc_caches[index];
  2843. }
  2844. void *__kmalloc(size_t size, gfp_t flags)
  2845. {
  2846. struct kmem_cache *s;
  2847. void *ret;
  2848. if (unlikely(size > SLUB_MAX_SIZE))
  2849. return kmalloc_large(size, flags);
  2850. s = get_slab(size, flags);
  2851. if (unlikely(ZERO_OR_NULL_PTR(s)))
  2852. return s;
  2853. ret = slab_alloc(s, flags, NUMA_NO_NODE, _RET_IP_);
  2854. trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
  2855. return ret;
  2856. }
  2857. EXPORT_SYMBOL(__kmalloc);
  2858. #ifdef CONFIG_NUMA
  2859. static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
  2860. {
  2861. struct page *page;
  2862. void *ptr = NULL;
  2863. flags |= __GFP_COMP | __GFP_NOTRACK;
  2864. page = alloc_pages_node(node, flags, get_order(size));
  2865. if (page)
  2866. ptr = page_address(page);
  2867. kmemleak_alloc(ptr, size, 1, flags);
  2868. return ptr;
  2869. }
  2870. void *__kmalloc_node(size_t size, gfp_t flags, int node)
  2871. {
  2872. struct kmem_cache *s;
  2873. void *ret;
  2874. if (unlikely(size > SLUB_MAX_SIZE)) {
  2875. ret = kmalloc_large_node(size, flags, node);
  2876. trace_kmalloc_node(_RET_IP_, ret,
  2877. size, PAGE_SIZE << get_order(size),
  2878. flags, node);
  2879. return ret;
  2880. }
  2881. s = get_slab(size, flags);
  2882. if (unlikely(ZERO_OR_NULL_PTR(s)))
  2883. return s;
  2884. ret = slab_alloc(s, flags, node, _RET_IP_);
  2885. trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
  2886. return ret;
  2887. }
  2888. EXPORT_SYMBOL(__kmalloc_node);
  2889. #endif
  2890. size_t ksize(const void *object)
  2891. {
  2892. struct page *page;
  2893. if (unlikely(object == ZERO_SIZE_PTR))
  2894. return 0;
  2895. page = virt_to_head_page(object);
  2896. if (unlikely(!PageSlab(page))) {
  2897. WARN_ON(!PageCompound(page));
  2898. return PAGE_SIZE << compound_order(page);
  2899. }
  2900. return slab_ksize(page->slab);
  2901. }
  2902. EXPORT_SYMBOL(ksize);
  2903. #ifdef CONFIG_SLUB_DEBUG
  2904. bool verify_mem_not_deleted(const void *x)
  2905. {
  2906. struct page *page;
  2907. void *object = (void *)x;
  2908. unsigned long flags;
  2909. bool rv;
  2910. if (unlikely(ZERO_OR_NULL_PTR(x)))
  2911. return false;
  2912. local_irq_save(flags);
  2913. page = virt_to_head_page(x);
  2914. if (unlikely(!PageSlab(page))) {
  2915. /* maybe it was from stack? */
  2916. rv = true;
  2917. goto out_unlock;
  2918. }
  2919. slab_lock(page);
  2920. if (on_freelist(page->slab, page, object)) {
  2921. object_err(page->slab, page, object, "Object is on free-list");
  2922. rv = false;
  2923. } else {
  2924. rv = true;
  2925. }
  2926. slab_unlock(page);
  2927. out_unlock:
  2928. local_irq_restore(flags);
  2929. return rv;
  2930. }
  2931. EXPORT_SYMBOL(verify_mem_not_deleted);
  2932. #endif
  2933. void kfree(const void *x)
  2934. {
  2935. struct page *page;
  2936. void *object = (void *)x;
  2937. trace_kfree(_RET_IP_, x);
  2938. if (unlikely(ZERO_OR_NULL_PTR(x)))
  2939. return;
  2940. page = virt_to_head_page(x);
  2941. if (unlikely(!PageSlab(page))) {
  2942. BUG_ON(!PageCompound(page));
  2943. kmemleak_free(x);
  2944. put_page(page);
  2945. return;
  2946. }
  2947. slab_free(page->slab, page, object, _RET_IP_);
  2948. }
  2949. EXPORT_SYMBOL(kfree);
  2950. /*
  2951. * kmem_cache_shrink removes empty slabs from the partial lists and sorts
  2952. * the remaining slabs by the number of items in use. The slabs with the
  2953. * most items in use come first. New allocations will then fill those up
  2954. * and thus they can be removed from the partial lists.
  2955. *
  2956. * The slabs with the least items are placed last. This results in them
  2957. * being allocated from last increasing the chance that the last objects
  2958. * are freed in them.
  2959. */
  2960. int kmem_cache_shrink(struct kmem_cache *s)
  2961. {
  2962. int node;
  2963. int i;
  2964. struct kmem_cache_node *n;
  2965. struct page *page;
  2966. struct page *t;
  2967. int objects = oo_objects(s->max);
  2968. struct list_head *slabs_by_inuse =
  2969. kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL);
  2970. unsigned long flags;
  2971. if (!slabs_by_inuse)
  2972. return -ENOMEM;
  2973. flush_all(s);
  2974. for_each_node_state(node, N_NORMAL_MEMORY) {
  2975. n = get_node(s, node);
  2976. if (!n->nr_partial)
  2977. continue;
  2978. for (i = 0; i < objects; i++)
  2979. INIT_LIST_HEAD(slabs_by_inuse + i);
  2980. spin_lock_irqsave(&n->list_lock, flags);
  2981. /*
  2982. * Build lists indexed by the items in use in each slab.
  2983. *
  2984. * Note that concurrent frees may occur while we hold the
  2985. * list_lock. page->inuse here is the upper limit.
  2986. */
  2987. list_for_each_entry_safe(page, t, &n->partial, lru) {
  2988. list_move(&page->lru, slabs_by_inuse + page->inuse);
  2989. if (!page->inuse)
  2990. n->nr_partial--;
  2991. }
  2992. /*
  2993. * Rebuild the partial list with the slabs filled up most
  2994. * first and the least used slabs at the end.
  2995. */
  2996. for (i = objects - 1; i > 0; i--)
  2997. list_splice(slabs_by_inuse + i, n->partial.prev);
  2998. spin_unlock_irqrestore(&n->list_lock, flags);
  2999. /* Release empty slabs */
  3000. list_for_each_entry_safe(page, t, slabs_by_inuse, lru)
  3001. discard_slab(s, page);
  3002. }
  3003. kfree(slabs_by_inuse);
  3004. return 0;
  3005. }
  3006. EXPORT_SYMBOL(kmem_cache_shrink);
  3007. #if defined(CONFIG_MEMORY_HOTPLUG)
  3008. static int slab_mem_going_offline_callback(void *arg)
  3009. {
  3010. struct kmem_cache *s;
  3011. down_read(&slub_lock);
  3012. list_for_each_entry(s, &slab_caches, list)
  3013. kmem_cache_shrink(s);
  3014. up_read(&slub_lock);
  3015. return 0;
  3016. }
  3017. static void slab_mem_offline_callback(void *arg)
  3018. {
  3019. struct kmem_cache_node *n;
  3020. struct kmem_cache *s;
  3021. struct memory_notify *marg = arg;
  3022. int offline_node;
  3023. offline_node = marg->status_change_nid;
  3024. /*
  3025. * If the node still has available memory. we need kmem_cache_node
  3026. * for it yet.
  3027. */
  3028. if (offline_node < 0)
  3029. return;
  3030. down_read(&slub_lock);
  3031. list_for_each_entry(s, &slab_caches, list) {
  3032. n = get_node(s, offline_node);
  3033. if (n) {
  3034. /*
  3035. * if n->nr_slabs > 0, slabs still exist on the node
  3036. * that is going down. We were unable to free them,
  3037. * and offline_pages() function shouldn't call this
  3038. * callback. So, we must fail.
  3039. */
  3040. BUG_ON(slabs_node(s, offline_node));
  3041. s->node[offline_node] = NULL;
  3042. kmem_cache_free(kmem_cache_node, n);
  3043. }
  3044. }
  3045. up_read(&slub_lock);
  3046. }
  3047. static int slab_mem_going_online_callback(void *arg)
  3048. {
  3049. struct kmem_cache_node *n;
  3050. struct kmem_cache *s;
  3051. struct memory_notify *marg = arg;
  3052. int nid = marg->status_change_nid;
  3053. int ret = 0;
  3054. /*
  3055. * If the node's memory is already available, then kmem_cache_node is
  3056. * already created. Nothing to do.
  3057. */
  3058. if (nid < 0)
  3059. return 0;
  3060. /*
  3061. * We are bringing a node online. No memory is available yet. We must
  3062. * allocate a kmem_cache_node structure in order to bring the node
  3063. * online.
  3064. */
  3065. down_read(&slub_lock);
  3066. list_for_each_entry(s, &slab_caches, list) {
  3067. /*
  3068. * XXX: kmem_cache_alloc_node will fallback to other nodes
  3069. * since memory is not yet available from the node that
  3070. * is brought up.
  3071. */
  3072. n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL);
  3073. if (!n) {
  3074. ret = -ENOMEM;
  3075. goto out;
  3076. }
  3077. init_kmem_cache_node(n);
  3078. s->node[nid] = n;
  3079. }
  3080. out:
  3081. up_read(&slub_lock);
  3082. return ret;
  3083. }
  3084. static int slab_memory_callback(struct notifier_block *self,
  3085. unsigned long action, void *arg)
  3086. {
  3087. int ret = 0;
  3088. switch (action) {
  3089. case MEM_GOING_ONLINE:
  3090. ret = slab_mem_going_online_callback(arg);
  3091. break;
  3092. case MEM_GOING_OFFLINE:
  3093. ret = slab_mem_going_offline_callback(arg);
  3094. break;
  3095. case MEM_OFFLINE:
  3096. case MEM_CANCEL_ONLINE:
  3097. slab_mem_offline_callback(arg);
  3098. break;
  3099. case MEM_ONLINE:
  3100. case MEM_CANCEL_OFFLINE:
  3101. break;
  3102. }
  3103. if (ret)
  3104. ret = notifier_from_errno(ret);
  3105. else
  3106. ret = NOTIFY_OK;
  3107. return ret;
  3108. }
  3109. #endif /* CONFIG_MEMORY_HOTPLUG */
  3110. /********************************************************************
  3111. * Basic setup of slabs
  3112. *******************************************************************/
  3113. /*
  3114. * Used for early kmem_cache structures that were allocated using
  3115. * the page allocator
  3116. */
  3117. static void __init kmem_cache_bootstrap_fixup(struct kmem_cache *s)
  3118. {
  3119. int node;
  3120. list_add(&s->list, &slab_caches);
  3121. s->refcount = -1;
  3122. for_each_node_state(node, N_NORMAL_MEMORY) {
  3123. struct kmem_cache_node *n = get_node(s, node);
  3124. struct page *p;
  3125. if (n) {
  3126. list_for_each_entry(p, &n->partial, lru)
  3127. p->slab = s;
  3128. #ifdef CONFIG_SLUB_DEBUG
  3129. list_for_each_entry(p, &n->full, lru)
  3130. p->slab = s;
  3131. #endif
  3132. }
  3133. }
  3134. }
  3135. void __init kmem_cache_init(void)
  3136. {
  3137. int i;
  3138. int caches = 0;
  3139. struct kmem_cache *temp_kmem_cache;
  3140. int order;
  3141. struct kmem_cache *temp_kmem_cache_node;
  3142. unsigned long kmalloc_size;
  3143. if (debug_guardpage_minorder())
  3144. slub_max_order = 0;
  3145. kmem_size = offsetof(struct kmem_cache, node) +
  3146. nr_node_ids * sizeof(struct kmem_cache_node *);
  3147. /* Allocate two kmem_caches from the page allocator */
  3148. kmalloc_size = ALIGN(kmem_size, cache_line_size());
  3149. order = get_order(2 * kmalloc_size);
  3150. kmem_cache = (void *)__get_free_pages(GFP_NOWAIT, order);
  3151. /*
  3152. * Must first have the slab cache available for the allocations of the
  3153. * struct kmem_cache_node's. There is special bootstrap code in
  3154. * kmem_cache_open for slab_state == DOWN.
  3155. */
  3156. kmem_cache_node = (void *)kmem_cache + kmalloc_size;
  3157. kmem_cache_open(kmem_cache_node, "kmem_cache_node",
  3158. sizeof(struct kmem_cache_node),
  3159. 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
  3160. hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI);
  3161. /* Able to allocate the per node structures */
  3162. slab_state = PARTIAL;
  3163. temp_kmem_cache = kmem_cache;
  3164. kmem_cache_open(kmem_cache, "kmem_cache", kmem_size,
  3165. 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
  3166. kmem_cache = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
  3167. memcpy(kmem_cache, temp_kmem_cache, kmem_size);
  3168. /*
  3169. * Allocate kmem_cache_node properly from the kmem_cache slab.
  3170. * kmem_cache_node is separately allocated so no need to
  3171. * update any list pointers.
  3172. */
  3173. temp_kmem_cache_node = kmem_cache_node;
  3174. kmem_cache_node = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
  3175. memcpy(kmem_cache_node, temp_kmem_cache_node, kmem_size);
  3176. kmem_cache_bootstrap_fixup(kmem_cache_node);
  3177. caches++;
  3178. kmem_cache_bootstrap_fixup(kmem_cache);
  3179. caches++;
  3180. /* Free temporary boot structure */
  3181. free_pages((unsigned long)temp_kmem_cache, order);
  3182. /* Now we can use the kmem_cache to allocate kmalloc slabs */
  3183. /*
  3184. * Patch up the size_index table if we have strange large alignment
  3185. * requirements for the kmalloc array. This is only the case for
  3186. * MIPS it seems. The standard arches will not generate any code here.
  3187. *
  3188. * Largest permitted alignment is 256 bytes due to the way we
  3189. * handle the index determination for the smaller caches.
  3190. *
  3191. * Make sure that nothing crazy happens if someone starts tinkering
  3192. * around with ARCH_KMALLOC_MINALIGN
  3193. */
  3194. BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
  3195. (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
  3196. for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
  3197. int elem = size_index_elem(i);
  3198. if (elem >= ARRAY_SIZE(size_index))
  3199. break;
  3200. size_index[elem] = KMALLOC_SHIFT_LOW;
  3201. }
  3202. if (KMALLOC_MIN_SIZE == 64) {
  3203. /*
  3204. * The 96 byte size cache is not used if the alignment
  3205. * is 64 byte.
  3206. */
  3207. for (i = 64 + 8; i <= 96; i += 8)
  3208. size_index[size_index_elem(i)] = 7;
  3209. } else if (KMALLOC_MIN_SIZE == 128) {
  3210. /*
  3211. * The 192 byte sized cache is not used if the alignment
  3212. * is 128 byte. Redirect kmalloc to use the 256 byte cache
  3213. * instead.
  3214. */
  3215. for (i = 128 + 8; i <= 192; i += 8)
  3216. size_index[size_index_elem(i)] = 8;
  3217. }
  3218. /* Caches that are not of the two-to-the-power-of size */
  3219. if (KMALLOC_MIN_SIZE <= 32) {
  3220. kmalloc_caches[1] = create_kmalloc_cache("kmalloc-96", 96, 0);
  3221. caches++;
  3222. }
  3223. if (KMALLOC_MIN_SIZE <= 64) {
  3224. kmalloc_caches[2] = create_kmalloc_cache("kmalloc-192", 192, 0);
  3225. caches++;
  3226. }
  3227. for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
  3228. kmalloc_caches[i] = create_kmalloc_cache("kmalloc", 1 << i, 0);
  3229. caches++;
  3230. }
  3231. slab_state = UP;
  3232. /* Provide the correct kmalloc names now that the caches are up */
  3233. if (KMALLOC_MIN_SIZE <= 32) {
  3234. kmalloc_caches[1]->name = kstrdup(kmalloc_caches[1]->name, GFP_NOWAIT);
  3235. BUG_ON(!kmalloc_caches[1]->name);
  3236. }
  3237. if (KMALLOC_MIN_SIZE <= 64) {
  3238. kmalloc_caches[2]->name = kstrdup(kmalloc_caches[2]->name, GFP_NOWAIT);
  3239. BUG_ON(!kmalloc_caches[2]->name);
  3240. }
  3241. for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
  3242. char *s = kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i);
  3243. BUG_ON(!s);
  3244. kmalloc_caches[i]->name = s;
  3245. }
  3246. #ifdef CONFIG_SMP
  3247. register_cpu_notifier(&slab_notifier);
  3248. #endif
  3249. #ifdef CONFIG_ZONE_DMA
  3250. for (i = 0; i < SLUB_PAGE_SHIFT; i++) {
  3251. struct kmem_cache *s = kmalloc_caches[i];
  3252. if (s && s->size) {
  3253. char *name = kasprintf(GFP_NOWAIT,
  3254. "dma-kmalloc-%d", s->object_size);
  3255. BUG_ON(!name);
  3256. kmalloc_dma_caches[i] = create_kmalloc_cache(name,
  3257. s->object_size, SLAB_CACHE_DMA);
  3258. }
  3259. }
  3260. #endif
  3261. printk(KERN_INFO
  3262. "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
  3263. " CPUs=%d, Nodes=%d\n",
  3264. caches, cache_line_size(),
  3265. slub_min_order, slub_max_order, slub_min_objects,
  3266. nr_cpu_ids, nr_node_ids);
  3267. }
  3268. void __init kmem_cache_init_late(void)
  3269. {
  3270. }
  3271. /*
  3272. * Find a mergeable slab cache
  3273. */
  3274. static int slab_unmergeable(struct kmem_cache *s)
  3275. {
  3276. if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE))
  3277. return 1;
  3278. if (s->ctor)
  3279. return 1;
  3280. /*
  3281. * We may have set a slab to be unmergeable during bootstrap.
  3282. */
  3283. if (s->refcount < 0)
  3284. return 1;
  3285. return 0;
  3286. }
  3287. static struct kmem_cache *find_mergeable(size_t size,
  3288. size_t align, unsigned long flags, const char *name,
  3289. void (*ctor)(void *))
  3290. {
  3291. struct kmem_cache *s;
  3292. if (slub_nomerge || (flags & SLUB_NEVER_MERGE))
  3293. return NULL;
  3294. if (ctor)
  3295. return NULL;
  3296. size = ALIGN(size, sizeof(void *));
  3297. align = calculate_alignment(flags, align, size);
  3298. size = ALIGN(size, align);
  3299. flags = kmem_cache_flags(size, flags, name, NULL);
  3300. list_for_each_entry(s, &slab_caches, list) {
  3301. if (slab_unmergeable(s))
  3302. continue;
  3303. if (size > s->size)
  3304. continue;
  3305. if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME))
  3306. continue;
  3307. /*
  3308. * Check if alignment is compatible.
  3309. * Courtesy of Adrian Drzewiecki
  3310. */
  3311. if ((s->size & ~(align - 1)) != s->size)
  3312. continue;
  3313. if (s->size - size >= sizeof(void *))
  3314. continue;
  3315. return s;
  3316. }
  3317. return NULL;
  3318. }
  3319. struct kmem_cache *kmem_cache_create(const char *name, size_t size,
  3320. size_t align, unsigned long flags, void (*ctor)(void *))
  3321. {
  3322. struct kmem_cache *s;
  3323. char *n;
  3324. if (WARN_ON(!name))
  3325. return NULL;
  3326. down_write(&slub_lock);
  3327. s = find_mergeable(size, align, flags, name, ctor);
  3328. if (s) {
  3329. s->refcount++;
  3330. /*
  3331. * Adjust the object sizes so that we clear
  3332. * the complete object on kzalloc.
  3333. */
  3334. s->object_size = max(s->object_size, (int)size);
  3335. s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *)));
  3336. if (sysfs_slab_alias(s, name)) {
  3337. s->refcount--;
  3338. goto err;
  3339. }
  3340. up_write(&slub_lock);
  3341. return s;
  3342. }
  3343. n = kstrdup(name, GFP_KERNEL);
  3344. if (!n)
  3345. goto err;
  3346. s = kmalloc(kmem_size, GFP_KERNEL);
  3347. if (s) {
  3348. if (kmem_cache_open(s, n,
  3349. size, align, flags, ctor)) {
  3350. list_add(&s->list, &slab_caches);
  3351. up_write(&slub_lock);
  3352. if (sysfs_slab_add(s)) {
  3353. down_write(&slub_lock);
  3354. list_del(&s->list);
  3355. kfree(n);
  3356. kfree(s);
  3357. goto err;
  3358. }
  3359. return s;
  3360. }
  3361. kfree(s);
  3362. }
  3363. kfree(n);
  3364. err:
  3365. up_write(&slub_lock);
  3366. if (flags & SLAB_PANIC)
  3367. panic("Cannot create slabcache %s\n", name);
  3368. else
  3369. s = NULL;
  3370. return s;
  3371. }
  3372. EXPORT_SYMBOL(kmem_cache_create);
  3373. #ifdef CONFIG_SMP
  3374. /*
  3375. * Use the cpu notifier to insure that the cpu slabs are flushed when
  3376. * necessary.
  3377. */
  3378. static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb,
  3379. unsigned long action, void *hcpu)
  3380. {
  3381. long cpu = (long)hcpu;
  3382. struct kmem_cache *s;
  3383. unsigned long flags;
  3384. switch (action) {
  3385. case CPU_UP_CANCELED:
  3386. case CPU_UP_CANCELED_FROZEN:
  3387. case CPU_DEAD:
  3388. case CPU_DEAD_FROZEN:
  3389. down_read(&slub_lock);
  3390. list_for_each_entry(s, &slab_caches, list) {
  3391. local_irq_save(flags);
  3392. __flush_cpu_slab(s, cpu);
  3393. local_irq_restore(flags);
  3394. }
  3395. up_read(&slub_lock);
  3396. break;
  3397. default:
  3398. break;
  3399. }
  3400. return NOTIFY_OK;
  3401. }
  3402. static struct notifier_block __cpuinitdata slab_notifier = {
  3403. .notifier_call = slab_cpuup_callback
  3404. };
  3405. #endif
  3406. void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
  3407. {
  3408. struct kmem_cache *s;
  3409. void *ret;
  3410. if (unlikely(size > SLUB_MAX_SIZE))
  3411. return kmalloc_large(size, gfpflags);
  3412. s = get_slab(size, gfpflags);
  3413. if (unlikely(ZERO_OR_NULL_PTR(s)))
  3414. return s;
  3415. ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, caller);
  3416. /* Honor the call site pointer we received. */
  3417. trace_kmalloc(caller, ret, size, s->size, gfpflags);
  3418. return ret;
  3419. }
  3420. #ifdef CONFIG_NUMA
  3421. void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
  3422. int node, unsigned long caller)
  3423. {
  3424. struct kmem_cache *s;
  3425. void *ret;
  3426. if (unlikely(size > SLUB_MAX_SIZE)) {
  3427. ret = kmalloc_large_node(size, gfpflags, node);
  3428. trace_kmalloc_node(caller, ret,
  3429. size, PAGE_SIZE << get_order(size),
  3430. gfpflags, node);
  3431. return ret;
  3432. }
  3433. s = get_slab(size, gfpflags);
  3434. if (unlikely(ZERO_OR_NULL_PTR(s)))
  3435. return s;
  3436. ret = slab_alloc(s, gfpflags, node, caller);
  3437. /* Honor the call site pointer we received. */
  3438. trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
  3439. return ret;
  3440. }
  3441. #endif
  3442. #ifdef CONFIG_SYSFS
  3443. static int count_inuse(struct page *page)
  3444. {
  3445. return page->inuse;
  3446. }
  3447. static int count_total(struct page *page)
  3448. {
  3449. return page->objects;
  3450. }
  3451. #endif
  3452. #ifdef CONFIG_SLUB_DEBUG
  3453. static int validate_slab(struct kmem_cache *s, struct page *page,
  3454. unsigned long *map)
  3455. {
  3456. void *p;
  3457. void *addr = page_address(page);
  3458. if (!check_slab(s, page) ||
  3459. !on_freelist(s, page, NULL))
  3460. return 0;
  3461. /* Now we know that a valid freelist exists */
  3462. bitmap_zero(map, page->objects);
  3463. get_map(s, page, map);
  3464. for_each_object(p, s, addr, page->objects) {
  3465. if (test_bit(slab_index(p, s, addr), map))
  3466. if (!check_object(s, page, p, SLUB_RED_INACTIVE))
  3467. return 0;
  3468. }
  3469. for_each_object(p, s, addr, page->objects)
  3470. if (!test_bit(slab_index(p, s, addr), map))
  3471. if (!check_object(s, page, p, SLUB_RED_ACTIVE))
  3472. return 0;
  3473. return 1;
  3474. }
  3475. static void validate_slab_slab(struct kmem_cache *s, struct page *page,
  3476. unsigned long *map)
  3477. {
  3478. slab_lock(page);
  3479. validate_slab(s, page, map);
  3480. slab_unlock(page);
  3481. }
  3482. static int validate_slab_node(struct kmem_cache *s,
  3483. struct kmem_cache_node *n, unsigned long *map)
  3484. {
  3485. unsigned long count = 0;
  3486. struct page *page;
  3487. unsigned long flags;
  3488. spin_lock_irqsave(&n->list_lock, flags);
  3489. list_for_each_entry(page, &n->partial, lru) {
  3490. validate_slab_slab(s, page, map);
  3491. count++;
  3492. }
  3493. if (count != n->nr_partial)
  3494. printk(KERN_ERR "SLUB %s: %ld partial slabs counted but "
  3495. "counter=%ld\n", s->name, count, n->nr_partial);
  3496. if (!(s->flags & SLAB_STORE_USER))
  3497. goto out;
  3498. list_for_each_entry(page, &n->full, lru) {
  3499. validate_slab_slab(s, page, map);
  3500. count++;
  3501. }
  3502. if (count != atomic_long_read(&n->nr_slabs))
  3503. printk(KERN_ERR "SLUB: %s %ld slabs counted but "
  3504. "counter=%ld\n", s->name, count,
  3505. atomic_long_read(&n->nr_slabs));
  3506. out:
  3507. spin_unlock_irqrestore(&n->list_lock, flags);
  3508. return count;
  3509. }
  3510. static long validate_slab_cache(struct kmem_cache *s)
  3511. {
  3512. int node;
  3513. unsigned long count = 0;
  3514. unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
  3515. sizeof(unsigned long), GFP_KERNEL);
  3516. if (!map)
  3517. return -ENOMEM;
  3518. flush_all(s);
  3519. for_each_node_state(node, N_NORMAL_MEMORY) {
  3520. struct kmem_cache_node *n = get_node(s, node);
  3521. count += validate_slab_node(s, n, map);
  3522. }
  3523. kfree(map);
  3524. return count;
  3525. }
  3526. /*
  3527. * Generate lists of code addresses where slabcache objects are allocated
  3528. * and freed.
  3529. */
  3530. struct location {
  3531. unsigned long count;
  3532. unsigned long addr;
  3533. long long sum_time;
  3534. long min_time;
  3535. long max_time;
  3536. long min_pid;
  3537. long max_pid;
  3538. DECLARE_BITMAP(cpus, NR_CPUS);
  3539. nodemask_t nodes;
  3540. };
  3541. struct loc_track {
  3542. unsigned long max;
  3543. unsigned long count;
  3544. struct location *loc;
  3545. };
  3546. static void free_loc_track(struct loc_track *t)
  3547. {
  3548. if (t->max)
  3549. free_pages((unsigned long)t->loc,
  3550. get_order(sizeof(struct location) * t->max));
  3551. }
  3552. static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
  3553. {
  3554. struct location *l;
  3555. int order;
  3556. order = get_order(sizeof(struct location) * max);
  3557. l = (void *)__get_free_pages(flags, order);
  3558. if (!l)
  3559. return 0;
  3560. if (t->count) {
  3561. memcpy(l, t->loc, sizeof(struct location) * t->count);
  3562. free_loc_track(t);
  3563. }
  3564. t->max = max;
  3565. t->loc = l;
  3566. return 1;
  3567. }
  3568. static int add_location(struct loc_track *t, struct kmem_cache *s,
  3569. const struct track *track)
  3570. {
  3571. long start, end, pos;
  3572. struct location *l;
  3573. unsigned long caddr;
  3574. unsigned long age = jiffies - track->when;
  3575. start = -1;
  3576. end = t->count;
  3577. for ( ; ; ) {
  3578. pos = start + (end - start + 1) / 2;
  3579. /*
  3580. * There is nothing at "end". If we end up there
  3581. * we need to add something to before end.
  3582. */
  3583. if (pos == end)
  3584. break;
  3585. caddr = t->loc[pos].addr;
  3586. if (track->addr == caddr) {
  3587. l = &t->loc[pos];
  3588. l->count++;
  3589. if (track->when) {
  3590. l->sum_time += age;
  3591. if (age < l->min_time)
  3592. l->min_time = age;
  3593. if (age > l->max_time)
  3594. l->max_time = age;
  3595. if (track->pid < l->min_pid)
  3596. l->min_pid = track->pid;
  3597. if (track->pid > l->max_pid)
  3598. l->max_pid = track->pid;
  3599. cpumask_set_cpu(track->cpu,
  3600. to_cpumask(l->cpus));
  3601. }
  3602. node_set(page_to_nid(virt_to_page(track)), l->nodes);
  3603. return 1;
  3604. }
  3605. if (track->addr < caddr)
  3606. end = pos;
  3607. else
  3608. start = pos;
  3609. }
  3610. /*
  3611. * Not found. Insert new tracking element.
  3612. */
  3613. if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
  3614. return 0;
  3615. l = t->loc + pos;
  3616. if (pos < t->count)
  3617. memmove(l + 1, l,
  3618. (t->count - pos) * sizeof(struct location));
  3619. t->count++;
  3620. l->count = 1;
  3621. l->addr = track->addr;
  3622. l->sum_time = age;
  3623. l->min_time = age;
  3624. l->max_time = age;
  3625. l->min_pid = track->pid;
  3626. l->max_pid = track->pid;
  3627. cpumask_clear(to_cpumask(l->cpus));
  3628. cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
  3629. nodes_clear(l->nodes);
  3630. node_set(page_to_nid(virt_to_page(track)), l->nodes);
  3631. return 1;
  3632. }
  3633. static void process_slab(struct loc_track *t, struct kmem_cache *s,
  3634. struct page *page, enum track_item alloc,
  3635. unsigned long *map)
  3636. {
  3637. void *addr = page_address(page);
  3638. void *p;
  3639. bitmap_zero(map, page->objects);
  3640. get_map(s, page, map);
  3641. for_each_object(p, s, addr, page->objects)
  3642. if (!test_bit(slab_index(p, s, addr), map))
  3643. add_location(t, s, get_track(s, p, alloc));
  3644. }
  3645. static int list_locations(struct kmem_cache *s, char *buf,
  3646. enum track_item alloc)
  3647. {
  3648. int len = 0;
  3649. unsigned long i;
  3650. struct loc_track t = { 0, 0, NULL };
  3651. int node;
  3652. unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
  3653. sizeof(unsigned long), GFP_KERNEL);
  3654. if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
  3655. GFP_TEMPORARY)) {
  3656. kfree(map);
  3657. return sprintf(buf, "Out of memory\n");
  3658. }
  3659. /* Push back cpu slabs */
  3660. flush_all(s);
  3661. for_each_node_state(node, N_NORMAL_MEMORY) {
  3662. struct kmem_cache_node *n = get_node(s, node);
  3663. unsigned long flags;
  3664. struct page *page;
  3665. if (!atomic_long_read(&n->nr_slabs))
  3666. continue;
  3667. spin_lock_irqsave(&n->list_lock, flags);
  3668. list_for_each_entry(page, &n->partial, lru)
  3669. process_slab(&t, s, page, alloc, map);
  3670. list_for_each_entry(page, &n->full, lru)
  3671. process_slab(&t, s, page, alloc, map);
  3672. spin_unlock_irqrestore(&n->list_lock, flags);
  3673. }
  3674. for (i = 0; i < t.count; i++) {
  3675. struct location *l = &t.loc[i];
  3676. if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100)
  3677. break;
  3678. len += sprintf(buf + len, "%7ld ", l->count);
  3679. if (l->addr)
  3680. len += sprintf(buf + len, "%pS", (void *)l->addr);
  3681. else
  3682. len += sprintf(buf + len, "<not-available>");
  3683. if (l->sum_time != l->min_time) {
  3684. len += sprintf(buf + len, " age=%ld/%ld/%ld",
  3685. l->min_time,
  3686. (long)div_u64(l->sum_time, l->count),
  3687. l->max_time);
  3688. } else
  3689. len += sprintf(buf + len, " age=%ld",
  3690. l->min_time);
  3691. if (l->min_pid != l->max_pid)
  3692. len += sprintf(buf + len, " pid=%ld-%ld",
  3693. l->min_pid, l->max_pid);
  3694. else
  3695. len += sprintf(buf + len, " pid=%ld",
  3696. l->min_pid);
  3697. if (num_online_cpus() > 1 &&
  3698. !cpumask_empty(to_cpumask(l->cpus)) &&
  3699. len < PAGE_SIZE - 60) {
  3700. len += sprintf(buf + len, " cpus=");
  3701. len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50,
  3702. to_cpumask(l->cpus));
  3703. }
  3704. if (nr_online_nodes > 1 && !nodes_empty(l->nodes) &&
  3705. len < PAGE_SIZE - 60) {
  3706. len += sprintf(buf + len, " nodes=");
  3707. len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50,
  3708. l->nodes);
  3709. }
  3710. len += sprintf(buf + len, "\n");
  3711. }
  3712. free_loc_track(&t);
  3713. kfree(map);
  3714. if (!t.count)
  3715. len += sprintf(buf, "No data\n");
  3716. return len;
  3717. }
  3718. #endif
  3719. #ifdef SLUB_RESILIENCY_TEST
  3720. static void resiliency_test(void)
  3721. {
  3722. u8 *p;
  3723. BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || SLUB_PAGE_SHIFT < 10);
  3724. printk(KERN_ERR "SLUB resiliency testing\n");
  3725. printk(KERN_ERR "-----------------------\n");
  3726. printk(KERN_ERR "A. Corruption after allocation\n");
  3727. p = kzalloc(16, GFP_KERNEL);
  3728. p[16] = 0x12;
  3729. printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer"
  3730. " 0x12->0x%p\n\n", p + 16);
  3731. validate_slab_cache(kmalloc_caches[4]);
  3732. /* Hmmm... The next two are dangerous */
  3733. p = kzalloc(32, GFP_KERNEL);
  3734. p[32 + sizeof(void *)] = 0x34;
  3735. printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab"
  3736. " 0x34 -> -0x%p\n", p);
  3737. printk(KERN_ERR
  3738. "If allocated object is overwritten then not detectable\n\n");
  3739. validate_slab_cache(kmalloc_caches[5]);
  3740. p = kzalloc(64, GFP_KERNEL);
  3741. p += 64 + (get_cycles() & 0xff) * sizeof(void *);
  3742. *p = 0x56;
  3743. printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
  3744. p);
  3745. printk(KERN_ERR
  3746. "If allocated object is overwritten then not detectable\n\n");
  3747. validate_slab_cache(kmalloc_caches[6]);
  3748. printk(KERN_ERR "\nB. Corruption after free\n");
  3749. p = kzalloc(128, GFP_KERNEL);
  3750. kfree(p);
  3751. *p = 0x78;
  3752. printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
  3753. validate_slab_cache(kmalloc_caches[7]);
  3754. p = kzalloc(256, GFP_KERNEL);
  3755. kfree(p);
  3756. p[50] = 0x9a;
  3757. printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
  3758. p);
  3759. validate_slab_cache(kmalloc_caches[8]);
  3760. p = kzalloc(512, GFP_KERNEL);
  3761. kfree(p);
  3762. p[512] = 0xab;
  3763. printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
  3764. validate_slab_cache(kmalloc_caches[9]);
  3765. }
  3766. #else
  3767. #ifdef CONFIG_SYSFS
  3768. static void resiliency_test(void) {};
  3769. #endif
  3770. #endif
  3771. #ifdef CONFIG_SYSFS
  3772. enum slab_stat_type {
  3773. SL_ALL, /* All slabs */
  3774. SL_PARTIAL, /* Only partially allocated slabs */
  3775. SL_CPU, /* Only slabs used for cpu caches */
  3776. SL_OBJECTS, /* Determine allocated objects not slabs */
  3777. SL_TOTAL /* Determine object capacity not slabs */
  3778. };
  3779. #define SO_ALL (1 << SL_ALL)
  3780. #define SO_PARTIAL (1 << SL_PARTIAL)
  3781. #define SO_CPU (1 << SL_CPU)
  3782. #define SO_OBJECTS (1 << SL_OBJECTS)
  3783. #define SO_TOTAL (1 << SL_TOTAL)
  3784. static ssize_t show_slab_objects(struct kmem_cache *s,
  3785. char *buf, unsigned long flags)
  3786. {
  3787. unsigned long total = 0;
  3788. int node;
  3789. int x;
  3790. unsigned long *nodes;
  3791. unsigned long *per_cpu;
  3792. nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL);
  3793. if (!nodes)
  3794. return -ENOMEM;
  3795. per_cpu = nodes + nr_node_ids;
  3796. if (flags & SO_CPU) {
  3797. int cpu;
  3798. for_each_possible_cpu(cpu) {
  3799. struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
  3800. int node;
  3801. struct page *page;
  3802. page = ACCESS_ONCE(c->page);
  3803. if (!page)
  3804. continue;
  3805. node = page_to_nid(page);
  3806. if (flags & SO_TOTAL)
  3807. x = page->objects;
  3808. else if (flags & SO_OBJECTS)
  3809. x = page->inuse;
  3810. else
  3811. x = 1;
  3812. total += x;
  3813. nodes[node] += x;
  3814. page = ACCESS_ONCE(c->partial);
  3815. if (page) {
  3816. x = page->pobjects;
  3817. total += x;
  3818. nodes[node] += x;
  3819. }
  3820. per_cpu[node]++;
  3821. }
  3822. }
  3823. lock_memory_hotplug();
  3824. #ifdef CONFIG_SLUB_DEBUG
  3825. if (flags & SO_ALL) {
  3826. for_each_node_state(node, N_NORMAL_MEMORY) {
  3827. struct kmem_cache_node *n = get_node(s, node);
  3828. if (flags & SO_TOTAL)
  3829. x = atomic_long_read(&n->total_objects);
  3830. else if (flags & SO_OBJECTS)
  3831. x = atomic_long_read(&n->total_objects) -
  3832. count_partial(n, count_free);
  3833. else
  3834. x = atomic_long_read(&n->nr_slabs);
  3835. total += x;
  3836. nodes[node] += x;
  3837. }
  3838. } else
  3839. #endif
  3840. if (flags & SO_PARTIAL) {
  3841. for_each_node_state(node, N_NORMAL_MEMORY) {
  3842. struct kmem_cache_node *n = get_node(s, node);
  3843. if (flags & SO_TOTAL)
  3844. x = count_partial(n, count_total);
  3845. else if (flags & SO_OBJECTS)
  3846. x = count_partial(n, count_inuse);
  3847. else
  3848. x = n->nr_partial;
  3849. total += x;
  3850. nodes[node] += x;
  3851. }
  3852. }
  3853. x = sprintf(buf, "%lu", total);
  3854. #ifdef CONFIG_NUMA
  3855. for_each_node_state(node, N_NORMAL_MEMORY)
  3856. if (nodes[node])
  3857. x += sprintf(buf + x, " N%d=%lu",
  3858. node, nodes[node]);
  3859. #endif
  3860. unlock_memory_hotplug();
  3861. kfree(nodes);
  3862. return x + sprintf(buf + x, "\n");
  3863. }
  3864. #ifdef CONFIG_SLUB_DEBUG
  3865. static int any_slab_objects(struct kmem_cache *s)
  3866. {
  3867. int node;
  3868. for_each_online_node(node) {
  3869. struct kmem_cache_node *n = get_node(s, node);
  3870. if (!n)
  3871. continue;
  3872. if (atomic_long_read(&n->total_objects))
  3873. return 1;
  3874. }
  3875. return 0;
  3876. }
  3877. #endif
  3878. #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
  3879. #define to_slab(n) container_of(n, struct kmem_cache, kobj)
  3880. struct slab_attribute {
  3881. struct attribute attr;
  3882. ssize_t (*show)(struct kmem_cache *s, char *buf);
  3883. ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
  3884. };
  3885. #define SLAB_ATTR_RO(_name) \
  3886. static struct slab_attribute _name##_attr = \
  3887. __ATTR(_name, 0400, _name##_show, NULL)
  3888. #define SLAB_ATTR(_name) \
  3889. static struct slab_attribute _name##_attr = \
  3890. __ATTR(_name, 0600, _name##_show, _name##_store)
  3891. static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
  3892. {
  3893. return sprintf(buf, "%d\n", s->size);
  3894. }
  3895. SLAB_ATTR_RO(slab_size);
  3896. static ssize_t align_show(struct kmem_cache *s, char *buf)
  3897. {
  3898. return sprintf(buf, "%d\n", s->align);
  3899. }
  3900. SLAB_ATTR_RO(align);
  3901. static ssize_t object_size_show(struct kmem_cache *s, char *buf)
  3902. {
  3903. return sprintf(buf, "%d\n", s->object_size);
  3904. }
  3905. SLAB_ATTR_RO(object_size);
  3906. static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
  3907. {
  3908. return sprintf(buf, "%d\n", oo_objects(s->oo));
  3909. }
  3910. SLAB_ATTR_RO(objs_per_slab);
  3911. static ssize_t order_store(struct kmem_cache *s,
  3912. const char *buf, size_t length)
  3913. {
  3914. unsigned long order;
  3915. int err;
  3916. err = strict_strtoul(buf, 10, &order);
  3917. if (err)
  3918. return err;
  3919. if (order > slub_max_order || order < slub_min_order)
  3920. return -EINVAL;
  3921. calculate_sizes(s, order);
  3922. return length;
  3923. }
  3924. static ssize_t order_show(struct kmem_cache *s, char *buf)
  3925. {
  3926. return sprintf(buf, "%d\n", oo_order(s->oo));
  3927. }
  3928. SLAB_ATTR(order);
  3929. static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
  3930. {
  3931. return sprintf(buf, "%lu\n", s->min_partial);
  3932. }
  3933. static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
  3934. size_t length)
  3935. {
  3936. unsigned long min;
  3937. int err;
  3938. err = strict_strtoul(buf, 10, &min);
  3939. if (err)
  3940. return err;
  3941. set_min_partial(s, min);
  3942. return length;
  3943. }
  3944. SLAB_ATTR(min_partial);
  3945. static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf)
  3946. {
  3947. return sprintf(buf, "%u\n", s->cpu_partial);
  3948. }
  3949. static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf,
  3950. size_t length)
  3951. {
  3952. unsigned long objects;
  3953. int err;
  3954. err = strict_strtoul(buf, 10, &objects);
  3955. if (err)
  3956. return err;
  3957. if (objects && kmem_cache_debug(s))
  3958. return -EINVAL;
  3959. s->cpu_partial = objects;
  3960. flush_all(s);
  3961. return length;
  3962. }
  3963. SLAB_ATTR(cpu_partial);
  3964. static ssize_t ctor_show(struct kmem_cache *s, char *buf)
  3965. {
  3966. if (!s->ctor)
  3967. return 0;
  3968. return sprintf(buf, "%pS\n", s->ctor);
  3969. }
  3970. SLAB_ATTR_RO(ctor);
  3971. static ssize_t aliases_show(struct kmem_cache *s, char *buf)
  3972. {
  3973. return sprintf(buf, "%d\n", s->refcount - 1);
  3974. }
  3975. SLAB_ATTR_RO(aliases);
  3976. static ssize_t partial_show(struct kmem_cache *s, char *buf)
  3977. {
  3978. return show_slab_objects(s, buf, SO_PARTIAL);
  3979. }
  3980. SLAB_ATTR_RO(partial);
  3981. static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
  3982. {
  3983. return show_slab_objects(s, buf, SO_CPU);
  3984. }
  3985. SLAB_ATTR_RO(cpu_slabs);
  3986. static ssize_t objects_show(struct kmem_cache *s, char *buf)
  3987. {
  3988. return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
  3989. }
  3990. SLAB_ATTR_RO(objects);
  3991. static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
  3992. {
  3993. return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
  3994. }
  3995. SLAB_ATTR_RO(objects_partial);
  3996. static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf)
  3997. {
  3998. int objects = 0;
  3999. int pages = 0;
  4000. int cpu;
  4001. int len;
  4002. for_each_online_cpu(cpu) {
  4003. struct page *page = per_cpu_ptr(s->cpu_slab, cpu)->partial;
  4004. if (page) {
  4005. pages += page->pages;
  4006. objects += page->pobjects;
  4007. }
  4008. }
  4009. len = sprintf(buf, "%d(%d)", objects, pages);
  4010. #ifdef CONFIG_SMP
  4011. for_each_online_cpu(cpu) {
  4012. struct page *page = per_cpu_ptr(s->cpu_slab, cpu) ->partial;
  4013. if (page && len < PAGE_SIZE - 20)
  4014. len += sprintf(buf + len, " C%d=%d(%d)", cpu,
  4015. page->pobjects, page->pages);
  4016. }
  4017. #endif
  4018. return len + sprintf(buf + len, "\n");
  4019. }
  4020. SLAB_ATTR_RO(slabs_cpu_partial);
  4021. static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
  4022. {
  4023. return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
  4024. }
  4025. static ssize_t reclaim_account_store(struct kmem_cache *s,
  4026. const char *buf, size_t length)
  4027. {
  4028. s->flags &= ~SLAB_RECLAIM_ACCOUNT;
  4029. if (buf[0] == '1')
  4030. s->flags |= SLAB_RECLAIM_ACCOUNT;
  4031. return length;
  4032. }
  4033. SLAB_ATTR(reclaim_account);
  4034. static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
  4035. {
  4036. return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
  4037. }
  4038. SLAB_ATTR_RO(hwcache_align);
  4039. #ifdef CONFIG_ZONE_DMA
  4040. static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
  4041. {
  4042. return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
  4043. }
  4044. SLAB_ATTR_RO(cache_dma);
  4045. #endif
  4046. static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
  4047. {
  4048. return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU));
  4049. }
  4050. SLAB_ATTR_RO(destroy_by_rcu);
  4051. static ssize_t reserved_show(struct kmem_cache *s, char *buf)
  4052. {
  4053. return sprintf(buf, "%d\n", s->reserved);
  4054. }
  4055. SLAB_ATTR_RO(reserved);
  4056. #ifdef CONFIG_SLUB_DEBUG
  4057. static ssize_t slabs_show(struct kmem_cache *s, char *buf)
  4058. {
  4059. return show_slab_objects(s, buf, SO_ALL);
  4060. }
  4061. SLAB_ATTR_RO(slabs);
  4062. static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
  4063. {
  4064. return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
  4065. }
  4066. SLAB_ATTR_RO(total_objects);
  4067. static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
  4068. {
  4069. return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE));
  4070. }
  4071. static ssize_t sanity_checks_store(struct kmem_cache *s,
  4072. const char *buf, size_t length)
  4073. {
  4074. s->flags &= ~SLAB_DEBUG_FREE;
  4075. if (buf[0] == '1') {
  4076. s->flags &= ~__CMPXCHG_DOUBLE;
  4077. s->flags |= SLAB_DEBUG_FREE;
  4078. }
  4079. return length;
  4080. }
  4081. SLAB_ATTR(sanity_checks);
  4082. static ssize_t trace_show(struct kmem_cache *s, char *buf)
  4083. {
  4084. return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
  4085. }
  4086. static ssize_t trace_store(struct kmem_cache *s, const char *buf,
  4087. size_t length)
  4088. {
  4089. s->flags &= ~SLAB_TRACE;
  4090. if (buf[0] == '1') {
  4091. s->flags &= ~__CMPXCHG_DOUBLE;
  4092. s->flags |= SLAB_TRACE;
  4093. }
  4094. return length;
  4095. }
  4096. SLAB_ATTR(trace);
  4097. static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
  4098. {
  4099. return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
  4100. }
  4101. static ssize_t red_zone_store(struct kmem_cache *s,
  4102. const char *buf, size_t length)
  4103. {
  4104. if (any_slab_objects(s))
  4105. return -EBUSY;
  4106. s->flags &= ~SLAB_RED_ZONE;
  4107. if (buf[0] == '1') {
  4108. s->flags &= ~__CMPXCHG_DOUBLE;
  4109. s->flags |= SLAB_RED_ZONE;
  4110. }
  4111. calculate_sizes(s, -1);
  4112. return length;
  4113. }
  4114. SLAB_ATTR(red_zone);
  4115. static ssize_t poison_show(struct kmem_cache *s, char *buf)
  4116. {
  4117. return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
  4118. }
  4119. static ssize_t poison_store(struct kmem_cache *s,
  4120. const char *buf, size_t length)
  4121. {
  4122. if (any_slab_objects(s))
  4123. return -EBUSY;
  4124. s->flags &= ~SLAB_POISON;
  4125. if (buf[0] == '1') {
  4126. s->flags &= ~__CMPXCHG_DOUBLE;
  4127. s->flags |= SLAB_POISON;
  4128. }
  4129. calculate_sizes(s, -1);
  4130. return length;
  4131. }
  4132. SLAB_ATTR(poison);
  4133. static ssize_t store_user_show(struct kmem_cache *s, char *buf)
  4134. {
  4135. return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
  4136. }
  4137. static ssize_t store_user_store(struct kmem_cache *s,
  4138. const char *buf, size_t length)
  4139. {
  4140. if (any_slab_objects(s))
  4141. return -EBUSY;
  4142. s->flags &= ~SLAB_STORE_USER;
  4143. if (buf[0] == '1') {
  4144. s->flags &= ~__CMPXCHG_DOUBLE;
  4145. s->flags |= SLAB_STORE_USER;
  4146. }
  4147. calculate_sizes(s, -1);
  4148. return length;
  4149. }
  4150. SLAB_ATTR(store_user);
  4151. static ssize_t validate_show(struct kmem_cache *s, char *buf)
  4152. {
  4153. return 0;
  4154. }
  4155. static ssize_t validate_store(struct kmem_cache *s,
  4156. const char *buf, size_t length)
  4157. {
  4158. int ret = -EINVAL;
  4159. if (buf[0] == '1') {
  4160. ret = validate_slab_cache(s);
  4161. if (ret >= 0)
  4162. ret = length;
  4163. }
  4164. return ret;
  4165. }
  4166. SLAB_ATTR(validate);
  4167. static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
  4168. {
  4169. if (!(s->flags & SLAB_STORE_USER))
  4170. return -ENOSYS;
  4171. return list_locations(s, buf, TRACK_ALLOC);
  4172. }
  4173. SLAB_ATTR_RO(alloc_calls);
  4174. static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
  4175. {
  4176. if (!(s->flags & SLAB_STORE_USER))
  4177. return -ENOSYS;
  4178. return list_locations(s, buf, TRACK_FREE);
  4179. }
  4180. SLAB_ATTR_RO(free_calls);
  4181. #endif /* CONFIG_SLUB_DEBUG */
  4182. #ifdef CONFIG_FAILSLAB
  4183. static ssize_t failslab_show(struct kmem_cache *s, char *buf)
  4184. {
  4185. return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
  4186. }
  4187. static ssize_t failslab_store(struct kmem_cache *s, const char *buf,
  4188. size_t length)
  4189. {
  4190. s->flags &= ~SLAB_FAILSLAB;
  4191. if (buf[0] == '1')
  4192. s->flags |= SLAB_FAILSLAB;
  4193. return length;
  4194. }
  4195. SLAB_ATTR(failslab);
  4196. #endif
  4197. static ssize_t shrink_show(struct kmem_cache *s, char *buf)
  4198. {
  4199. return 0;
  4200. }
  4201. static ssize_t shrink_store(struct kmem_cache *s,
  4202. const char *buf, size_t length)
  4203. {
  4204. if (buf[0] == '1') {
  4205. int rc = kmem_cache_shrink(s);
  4206. if (rc)
  4207. return rc;
  4208. } else
  4209. return -EINVAL;
  4210. return length;
  4211. }
  4212. SLAB_ATTR(shrink);
  4213. #ifdef CONFIG_NUMA
  4214. static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
  4215. {
  4216. return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10);
  4217. }
  4218. static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
  4219. const char *buf, size_t length)
  4220. {
  4221. unsigned long ratio;
  4222. int err;
  4223. err = strict_strtoul(buf, 10, &ratio);
  4224. if (err)
  4225. return err;
  4226. if (ratio <= 100)
  4227. s->remote_node_defrag_ratio = ratio * 10;
  4228. return length;
  4229. }
  4230. SLAB_ATTR(remote_node_defrag_ratio);
  4231. #endif
  4232. #ifdef CONFIG_SLUB_STATS
  4233. static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
  4234. {
  4235. unsigned long sum = 0;
  4236. int cpu;
  4237. int len;
  4238. int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL);
  4239. if (!data)
  4240. return -ENOMEM;
  4241. for_each_online_cpu(cpu) {
  4242. unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
  4243. data[cpu] = x;
  4244. sum += x;
  4245. }
  4246. len = sprintf(buf, "%lu", sum);
  4247. #ifdef CONFIG_SMP
  4248. for_each_online_cpu(cpu) {
  4249. if (data[cpu] && len < PAGE_SIZE - 20)
  4250. len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
  4251. }
  4252. #endif
  4253. kfree(data);
  4254. return len + sprintf(buf + len, "\n");
  4255. }
  4256. static void clear_stat(struct kmem_cache *s, enum stat_item si)
  4257. {
  4258. int cpu;
  4259. for_each_online_cpu(cpu)
  4260. per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
  4261. }
  4262. #define STAT_ATTR(si, text) \
  4263. static ssize_t text##_show(struct kmem_cache *s, char *buf) \
  4264. { \
  4265. return show_stat(s, buf, si); \
  4266. } \
  4267. static ssize_t text##_store(struct kmem_cache *s, \
  4268. const char *buf, size_t length) \
  4269. { \
  4270. if (buf[0] != '0') \
  4271. return -EINVAL; \
  4272. clear_stat(s, si); \
  4273. return length; \
  4274. } \
  4275. SLAB_ATTR(text); \
  4276. STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
  4277. STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
  4278. STAT_ATTR(FREE_FASTPATH, free_fastpath);
  4279. STAT_ATTR(FREE_SLOWPATH, free_slowpath);
  4280. STAT_ATTR(FREE_FROZEN, free_frozen);
  4281. STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
  4282. STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
  4283. STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
  4284. STAT_ATTR(ALLOC_SLAB, alloc_slab);
  4285. STAT_ATTR(ALLOC_REFILL, alloc_refill);
  4286. STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch);
  4287. STAT_ATTR(FREE_SLAB, free_slab);
  4288. STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
  4289. STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
  4290. STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
  4291. STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
  4292. STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
  4293. STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
  4294. STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass);
  4295. STAT_ATTR(ORDER_FALLBACK, order_fallback);
  4296. STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail);
  4297. STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail);
  4298. STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc);
  4299. STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free);
  4300. STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node);
  4301. STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain);
  4302. #endif
  4303. static struct attribute *slab_attrs[] = {
  4304. &slab_size_attr.attr,
  4305. &object_size_attr.attr,
  4306. &objs_per_slab_attr.attr,
  4307. &order_attr.attr,
  4308. &min_partial_attr.attr,
  4309. &cpu_partial_attr.attr,
  4310. &objects_attr.attr,
  4311. &objects_partial_attr.attr,
  4312. &partial_attr.attr,
  4313. &cpu_slabs_attr.attr,
  4314. &ctor_attr.attr,
  4315. &aliases_attr.attr,
  4316. &align_attr.attr,
  4317. &hwcache_align_attr.attr,
  4318. &reclaim_account_attr.attr,
  4319. &destroy_by_rcu_attr.attr,
  4320. &shrink_attr.attr,
  4321. &reserved_attr.attr,
  4322. &slabs_cpu_partial_attr.attr,
  4323. #ifdef CONFIG_SLUB_DEBUG
  4324. &total_objects_attr.attr,
  4325. &slabs_attr.attr,
  4326. &sanity_checks_attr.attr,
  4327. &trace_attr.attr,
  4328. &red_zone_attr.attr,
  4329. &poison_attr.attr,
  4330. &store_user_attr.attr,
  4331. &validate_attr.attr,
  4332. &alloc_calls_attr.attr,
  4333. &free_calls_attr.attr,
  4334. #endif
  4335. #ifdef CONFIG_ZONE_DMA
  4336. &cache_dma_attr.attr,
  4337. #endif
  4338. #ifdef CONFIG_NUMA
  4339. &remote_node_defrag_ratio_attr.attr,
  4340. #endif
  4341. #ifdef CONFIG_SLUB_STATS
  4342. &alloc_fastpath_attr.attr,
  4343. &alloc_slowpath_attr.attr,
  4344. &free_fastpath_attr.attr,
  4345. &free_slowpath_attr.attr,
  4346. &free_frozen_attr.attr,
  4347. &free_add_partial_attr.attr,
  4348. &free_remove_partial_attr.attr,
  4349. &alloc_from_partial_attr.attr,
  4350. &alloc_slab_attr.attr,
  4351. &alloc_refill_attr.attr,
  4352. &alloc_node_mismatch_attr.attr,
  4353. &free_slab_attr.attr,
  4354. &cpuslab_flush_attr.attr,
  4355. &deactivate_full_attr.attr,
  4356. &deactivate_empty_attr.attr,
  4357. &deactivate_to_head_attr.attr,
  4358. &deactivate_to_tail_attr.attr,
  4359. &deactivate_remote_frees_attr.attr,
  4360. &deactivate_bypass_attr.attr,
  4361. &order_fallback_attr.attr,
  4362. &cmpxchg_double_fail_attr.attr,
  4363. &cmpxchg_double_cpu_fail_attr.attr,
  4364. &cpu_partial_alloc_attr.attr,
  4365. &cpu_partial_free_attr.attr,
  4366. &cpu_partial_node_attr.attr,
  4367. &cpu_partial_drain_attr.attr,
  4368. #endif
  4369. #ifdef CONFIG_FAILSLAB
  4370. &failslab_attr.attr,
  4371. #endif
  4372. NULL
  4373. };
  4374. static struct attribute_group slab_attr_group = {
  4375. .attrs = slab_attrs,
  4376. };
  4377. static ssize_t slab_attr_show(struct kobject *kobj,
  4378. struct attribute *attr,
  4379. char *buf)
  4380. {
  4381. struct slab_attribute *attribute;
  4382. struct kmem_cache *s;
  4383. int err;
  4384. attribute = to_slab_attr(attr);
  4385. s = to_slab(kobj);
  4386. if (!attribute->show)
  4387. return -EIO;
  4388. err = attribute->show(s, buf);
  4389. return err;
  4390. }
  4391. static ssize_t slab_attr_store(struct kobject *kobj,
  4392. struct attribute *attr,
  4393. const char *buf, size_t len)
  4394. {
  4395. struct slab_attribute *attribute;
  4396. struct kmem_cache *s;
  4397. int err;
  4398. attribute = to_slab_attr(attr);
  4399. s = to_slab(kobj);
  4400. if (!attribute->store)
  4401. return -EIO;
  4402. err = attribute->store(s, buf, len);
  4403. return err;
  4404. }
  4405. static void kmem_cache_release(struct kobject *kobj)
  4406. {
  4407. struct kmem_cache *s = to_slab(kobj);
  4408. kfree(s->name);
  4409. kfree(s);
  4410. }
  4411. static const struct sysfs_ops slab_sysfs_ops = {
  4412. .show = slab_attr_show,
  4413. .store = slab_attr_store,
  4414. };
  4415. static struct kobj_type slab_ktype = {
  4416. .sysfs_ops = &slab_sysfs_ops,
  4417. .release = kmem_cache_release
  4418. };
  4419. static int uevent_filter(struct kset *kset, struct kobject *kobj)
  4420. {
  4421. struct kobj_type *ktype = get_ktype(kobj);
  4422. if (ktype == &slab_ktype)
  4423. return 1;
  4424. return 0;
  4425. }
  4426. static const struct kset_uevent_ops slab_uevent_ops = {
  4427. .filter = uevent_filter,
  4428. };
  4429. static struct kset *slab_kset;
  4430. #define ID_STR_LENGTH 64
  4431. /* Create a unique string id for a slab cache:
  4432. *
  4433. * Format :[flags-]size
  4434. */
  4435. static char *create_unique_id(struct kmem_cache *s)
  4436. {
  4437. char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
  4438. char *p = name;
  4439. BUG_ON(!name);
  4440. *p++ = ':';
  4441. /*
  4442. * First flags affecting slabcache operations. We will only
  4443. * get here for aliasable slabs so we do not need to support
  4444. * too many flags. The flags here must cover all flags that
  4445. * are matched during merging to guarantee that the id is
  4446. * unique.
  4447. */
  4448. if (s->flags & SLAB_CACHE_DMA)
  4449. *p++ = 'd';
  4450. if (s->flags & SLAB_RECLAIM_ACCOUNT)
  4451. *p++ = 'a';
  4452. if (s->flags & SLAB_DEBUG_FREE)
  4453. *p++ = 'F';
  4454. if (!(s->flags & SLAB_NOTRACK))
  4455. *p++ = 't';
  4456. if (p != name + 1)
  4457. *p++ = '-';
  4458. p += sprintf(p, "%07d", s->size);
  4459. BUG_ON(p > name + ID_STR_LENGTH - 1);
  4460. return name;
  4461. }
  4462. static int sysfs_slab_add(struct kmem_cache *s)
  4463. {
  4464. int err;
  4465. const char *name;
  4466. int unmergeable;
  4467. if (slab_state < SYSFS)
  4468. /* Defer until later */
  4469. return 0;
  4470. unmergeable = slab_unmergeable(s);
  4471. if (unmergeable) {
  4472. /*
  4473. * Slabcache can never be merged so we can use the name proper.
  4474. * This is typically the case for debug situations. In that
  4475. * case we can catch duplicate names easily.
  4476. */
  4477. sysfs_remove_link(&slab_kset->kobj, s->name);
  4478. name = s->name;
  4479. } else {
  4480. /*
  4481. * Create a unique name for the slab as a target
  4482. * for the symlinks.
  4483. */
  4484. name = create_unique_id(s);
  4485. }
  4486. s->kobj.kset = slab_kset;
  4487. err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name);
  4488. if (err) {
  4489. kobject_put(&s->kobj);
  4490. return err;
  4491. }
  4492. err = sysfs_create_group(&s->kobj, &slab_attr_group);
  4493. if (err) {
  4494. kobject_del(&s->kobj);
  4495. kobject_put(&s->kobj);
  4496. return err;
  4497. }
  4498. kobject_uevent(&s->kobj, KOBJ_ADD);
  4499. if (!unmergeable) {
  4500. /* Setup first alias */
  4501. sysfs_slab_alias(s, s->name);
  4502. kfree(name);
  4503. }
  4504. return 0;
  4505. }
  4506. static void sysfs_slab_remove(struct kmem_cache *s)
  4507. {
  4508. if (slab_state < SYSFS)
  4509. /*
  4510. * Sysfs has not been setup yet so no need to remove the
  4511. * cache from sysfs.
  4512. */
  4513. return;
  4514. kobject_uevent(&s->kobj, KOBJ_REMOVE);
  4515. kobject_del(&s->kobj);
  4516. kobject_put(&s->kobj);
  4517. }
  4518. /*
  4519. * Need to buffer aliases during bootup until sysfs becomes
  4520. * available lest we lose that information.
  4521. */
  4522. struct saved_alias {
  4523. struct kmem_cache *s;
  4524. const char *name;
  4525. struct saved_alias *next;
  4526. };
  4527. static struct saved_alias *alias_list;
  4528. static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
  4529. {
  4530. struct saved_alias *al;
  4531. if (slab_state == SYSFS) {
  4532. /*
  4533. * If we have a leftover link then remove it.
  4534. */
  4535. sysfs_remove_link(&slab_kset->kobj, name);
  4536. return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
  4537. }
  4538. al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
  4539. if (!al)
  4540. return -ENOMEM;
  4541. al->s = s;
  4542. al->name = name;
  4543. al->next = alias_list;
  4544. alias_list = al;
  4545. return 0;
  4546. }
  4547. static int __init slab_sysfs_init(void)
  4548. {
  4549. struct kmem_cache *s;
  4550. int err;
  4551. down_write(&slub_lock);
  4552. slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
  4553. if (!slab_kset) {
  4554. up_write(&slub_lock);
  4555. printk(KERN_ERR "Cannot register slab subsystem.\n");
  4556. return -ENOSYS;
  4557. }
  4558. slab_state = SYSFS;
  4559. list_for_each_entry(s, &slab_caches, list) {
  4560. err = sysfs_slab_add(s);
  4561. if (err)
  4562. printk(KERN_ERR "SLUB: Unable to add boot slab %s"
  4563. " to sysfs\n", s->name);
  4564. }
  4565. while (alias_list) {
  4566. struct saved_alias *al = alias_list;
  4567. alias_list = alias_list->next;
  4568. err = sysfs_slab_alias(al->s, al->name);
  4569. if (err)
  4570. printk(KERN_ERR "SLUB: Unable to add boot slab alias"
  4571. " %s to sysfs\n", s->name);
  4572. kfree(al);
  4573. }
  4574. up_write(&slub_lock);
  4575. resiliency_test();
  4576. return 0;
  4577. }
  4578. __initcall(slab_sysfs_init);
  4579. #endif /* CONFIG_SYSFS */
  4580. /*
  4581. * The /proc/slabinfo ABI
  4582. */
  4583. #ifdef CONFIG_SLABINFO
  4584. static void print_slabinfo_header(struct seq_file *m)
  4585. {
  4586. seq_puts(m, "slabinfo - version: 2.1\n");
  4587. seq_puts(m, "# name <active_objs> <num_objs> <object_size> "
  4588. "<objperslab> <pagesperslab>");
  4589. seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
  4590. seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
  4591. seq_putc(m, '\n');
  4592. }
  4593. static void *s_start(struct seq_file *m, loff_t *pos)
  4594. {
  4595. loff_t n = *pos;
  4596. down_read(&slub_lock);
  4597. if (!n)
  4598. print_slabinfo_header(m);
  4599. return seq_list_start(&slab_caches, *pos);
  4600. }
  4601. static void *s_next(struct seq_file *m, void *p, loff_t *pos)
  4602. {
  4603. return seq_list_next(p, &slab_caches, pos);
  4604. }
  4605. static void s_stop(struct seq_file *m, void *p)
  4606. {
  4607. up_read(&slub_lock);
  4608. }
  4609. static int s_show(struct seq_file *m, void *p)
  4610. {
  4611. unsigned long nr_partials = 0;
  4612. unsigned long nr_slabs = 0;
  4613. unsigned long nr_inuse = 0;
  4614. unsigned long nr_objs = 0;
  4615. unsigned long nr_free = 0;
  4616. struct kmem_cache *s;
  4617. int node;
  4618. s = list_entry(p, struct kmem_cache, list);
  4619. for_each_online_node(node) {
  4620. struct kmem_cache_node *n = get_node(s, node);
  4621. if (!n)
  4622. continue;
  4623. nr_partials += n->nr_partial;
  4624. nr_slabs += atomic_long_read(&n->nr_slabs);
  4625. nr_objs += atomic_long_read(&n->total_objects);
  4626. nr_free += count_partial(n, count_free);
  4627. }
  4628. nr_inuse = nr_objs - nr_free;
  4629. seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse,
  4630. nr_objs, s->size, oo_objects(s->oo),
  4631. (1 << oo_order(s->oo)));
  4632. seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0);
  4633. seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs,
  4634. 0UL);
  4635. seq_putc(m, '\n');
  4636. return 0;
  4637. }
  4638. static const struct seq_operations slabinfo_op = {
  4639. .start = s_start,
  4640. .next = s_next,
  4641. .stop = s_stop,
  4642. .show = s_show,
  4643. };
  4644. static int slabinfo_open(struct inode *inode, struct file *file)
  4645. {
  4646. return seq_open(file, &slabinfo_op);
  4647. }
  4648. static const struct file_operations proc_slabinfo_operations = {
  4649. .open = slabinfo_open,
  4650. .read = seq_read,
  4651. .llseek = seq_lseek,
  4652. .release = seq_release,
  4653. };
  4654. static int __init slab_proc_init(void)
  4655. {
  4656. proc_create("slabinfo", S_IRUSR, NULL, &proc_slabinfo_operations);
  4657. return 0;
  4658. }
  4659. module_init(slab_proc_init);
  4660. #endif /* CONFIG_SLABINFO */