page_alloc.c 121 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489149014911492149314941495149614971498149915001501150215031504150515061507150815091510151115121513151415151516151715181519152015211522152315241525152615271528152915301531153215331534153515361537153815391540154115421543154415451546154715481549155015511552155315541555155615571558155915601561156215631564156515661567156815691570157115721573157415751576157715781579158015811582158315841585158615871588158915901591159215931594159515961597159815991600160116021603160416051606160716081609161016111612161316141615161616171618161916201621162216231624162516261627162816291630163116321633163416351636163716381639164016411642164316441645164616471648164916501651165216531654165516561657165816591660166116621663166416651666166716681669167016711672167316741675167616771678167916801681168216831684168516861687168816891690169116921693169416951696169716981699170017011702170317041705170617071708170917101711171217131714171517161717171817191720172117221723172417251726172717281729173017311732173317341735173617371738173917401741174217431744174517461747174817491750175117521753175417551756175717581759176017611762176317641765176617671768176917701771177217731774177517761777177817791780178117821783178417851786178717881789179017911792179317941795179617971798179918001801180218031804180518061807180818091810181118121813181418151816181718181819182018211822182318241825182618271828182918301831183218331834183518361837183818391840184118421843184418451846184718481849185018511852185318541855185618571858185918601861186218631864186518661867186818691870187118721873187418751876187718781879188018811882188318841885188618871888188918901891189218931894189518961897189818991900190119021903190419051906190719081909191019111912191319141915191619171918191919201921192219231924192519261927192819291930193119321933193419351936193719381939194019411942194319441945194619471948194919501951195219531954195519561957195819591960196119621963196419651966196719681969197019711972197319741975197619771978197919801981198219831984198519861987198819891990199119921993199419951996199719981999200020012002200320042005200620072008200920102011201220132014201520162017201820192020202120222023202420252026202720282029203020312032203320342035203620372038203920402041204220432044204520462047204820492050205120522053205420552056205720582059206020612062206320642065206620672068206920702071207220732074207520762077207820792080208120822083208420852086208720882089209020912092209320942095209620972098209921002101210221032104210521062107210821092110211121122113211421152116211721182119212021212122212321242125212621272128212921302131213221332134213521362137213821392140214121422143214421452146214721482149215021512152215321542155215621572158215921602161216221632164216521662167216821692170217121722173217421752176217721782179218021812182218321842185218621872188218921902191219221932194219521962197219821992200220122022203220422052206220722082209221022112212221322142215221622172218221922202221222222232224222522262227222822292230223122322233223422352236223722382239224022412242224322442245224622472248224922502251225222532254225522562257225822592260226122622263226422652266226722682269227022712272227322742275227622772278227922802281228222832284228522862287228822892290229122922293229422952296229722982299230023012302230323042305230623072308230923102311231223132314231523162317231823192320232123222323232423252326232723282329233023312332233323342335233623372338233923402341234223432344234523462347234823492350235123522353235423552356235723582359236023612362236323642365236623672368236923702371237223732374237523762377237823792380238123822383238423852386238723882389239023912392239323942395239623972398239924002401240224032404240524062407240824092410241124122413241424152416241724182419242024212422242324242425242624272428242924302431243224332434243524362437243824392440244124422443244424452446244724482449245024512452245324542455245624572458245924602461246224632464246524662467246824692470247124722473247424752476247724782479248024812482248324842485248624872488248924902491249224932494249524962497249824992500250125022503250425052506250725082509251025112512251325142515251625172518251925202521252225232524252525262527252825292530253125322533253425352536253725382539254025412542254325442545254625472548254925502551255225532554255525562557255825592560256125622563256425652566256725682569257025712572257325742575257625772578257925802581258225832584258525862587258825892590259125922593259425952596259725982599260026012602260326042605260626072608260926102611261226132614261526162617261826192620262126222623262426252626262726282629263026312632263326342635263626372638263926402641264226432644264526462647264826492650265126522653265426552656265726582659266026612662266326642665266626672668266926702671267226732674267526762677267826792680268126822683268426852686268726882689269026912692269326942695269626972698269927002701270227032704270527062707270827092710271127122713271427152716271727182719272027212722272327242725272627272728272927302731273227332734273527362737273827392740274127422743274427452746274727482749275027512752275327542755275627572758275927602761276227632764276527662767276827692770277127722773277427752776277727782779278027812782278327842785278627872788278927902791279227932794279527962797279827992800280128022803280428052806280728082809281028112812281328142815281628172818281928202821282228232824282528262827282828292830283128322833283428352836283728382839284028412842284328442845284628472848284928502851285228532854285528562857285828592860286128622863286428652866286728682869287028712872287328742875287628772878287928802881288228832884288528862887288828892890289128922893289428952896289728982899290029012902290329042905290629072908290929102911291229132914291529162917291829192920292129222923292429252926292729282929293029312932293329342935293629372938293929402941294229432944294529462947294829492950295129522953295429552956295729582959296029612962296329642965296629672968296929702971297229732974297529762977297829792980298129822983298429852986298729882989299029912992299329942995299629972998299930003001300230033004300530063007300830093010301130123013301430153016301730183019302030213022302330243025302630273028302930303031303230333034303530363037303830393040304130423043304430453046304730483049305030513052305330543055305630573058305930603061306230633064306530663067306830693070307130723073307430753076307730783079308030813082308330843085308630873088308930903091309230933094309530963097309830993100310131023103310431053106310731083109311031113112311331143115311631173118311931203121312231233124312531263127312831293130313131323133313431353136313731383139314031413142314331443145314631473148314931503151315231533154315531563157315831593160316131623163316431653166316731683169317031713172317331743175317631773178317931803181318231833184318531863187318831893190319131923193319431953196319731983199320032013202320332043205320632073208320932103211321232133214321532163217321832193220322132223223322432253226322732283229323032313232323332343235323632373238323932403241324232433244324532463247324832493250325132523253325432553256325732583259326032613262326332643265326632673268326932703271327232733274327532763277327832793280328132823283328432853286328732883289329032913292329332943295329632973298329933003301330233033304330533063307330833093310331133123313331433153316331733183319332033213322332333243325332633273328332933303331333233333334333533363337333833393340334133423343334433453346334733483349335033513352335333543355335633573358335933603361336233633364336533663367336833693370337133723373337433753376337733783379338033813382338333843385338633873388338933903391339233933394339533963397339833993400340134023403340434053406340734083409341034113412341334143415341634173418341934203421342234233424342534263427342834293430343134323433343434353436343734383439344034413442344334443445344634473448344934503451345234533454345534563457345834593460346134623463346434653466346734683469347034713472347334743475347634773478347934803481348234833484348534863487348834893490349134923493349434953496349734983499350035013502350335043505350635073508350935103511351235133514351535163517351835193520352135223523352435253526352735283529353035313532353335343535353635373538353935403541354235433544354535463547354835493550355135523553355435553556355735583559356035613562356335643565356635673568356935703571357235733574357535763577357835793580358135823583358435853586358735883589359035913592359335943595359635973598359936003601360236033604360536063607360836093610361136123613361436153616361736183619362036213622362336243625362636273628362936303631363236333634363536363637363836393640364136423643364436453646364736483649365036513652365336543655365636573658365936603661366236633664366536663667366836693670367136723673367436753676367736783679368036813682368336843685368636873688368936903691369236933694369536963697369836993700370137023703370437053706370737083709371037113712371337143715371637173718371937203721372237233724372537263727372837293730373137323733373437353736373737383739374037413742374337443745374637473748374937503751375237533754375537563757375837593760376137623763376437653766376737683769377037713772377337743775377637773778377937803781378237833784378537863787378837893790379137923793379437953796379737983799380038013802380338043805380638073808380938103811381238133814381538163817381838193820382138223823382438253826382738283829383038313832383338343835383638373838383938403841384238433844384538463847384838493850385138523853385438553856385738583859386038613862386338643865386638673868386938703871387238733874387538763877387838793880388138823883388438853886388738883889389038913892389338943895389638973898389939003901390239033904390539063907390839093910391139123913391439153916391739183919392039213922392339243925392639273928392939303931393239333934393539363937393839393940394139423943394439453946394739483949395039513952395339543955395639573958395939603961396239633964396539663967396839693970397139723973397439753976397739783979398039813982398339843985398639873988398939903991399239933994399539963997399839994000400140024003400440054006400740084009401040114012401340144015401640174018401940204021402240234024402540264027402840294030403140324033403440354036403740384039404040414042404340444045404640474048404940504051405240534054405540564057405840594060406140624063406440654066406740684069407040714072407340744075407640774078407940804081408240834084408540864087408840894090409140924093409440954096409740984099410041014102410341044105410641074108410941104111411241134114411541164117411841194120412141224123412441254126412741284129413041314132413341344135413641374138413941404141414241434144414541464147414841494150415141524153415441554156415741584159416041614162416341644165416641674168416941704171417241734174417541764177417841794180418141824183418441854186418741884189419041914192419341944195419641974198419942004201420242034204420542064207420842094210421142124213421442154216421742184219422042214222422342244225422642274228422942304231423242334234423542364237423842394240424142424243424442454246424742484249425042514252425342544255425642574258425942604261426242634264426542664267426842694270427142724273427442754276427742784279428042814282428342844285428642874288428942904291429242934294429542964297429842994300430143024303430443054306430743084309431043114312431343144315431643174318431943204321432243234324432543264327432843294330433143324333433443354336433743384339434043414342434343444345434643474348434943504351435243534354435543564357435843594360436143624363436443654366436743684369437043714372437343744375437643774378437943804381438243834384438543864387438843894390439143924393439443954396439743984399440044014402440344044405440644074408440944104411441244134414441544164417441844194420442144224423442444254426442744284429443044314432443344344435443644374438
  1. /*
  2. * linux/mm/page_alloc.c
  3. *
  4. * Manages the free list, the system allocates free pages here.
  5. * Note that kmalloc() lives in slab.c
  6. *
  7. * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
  8. * Swap reorganised 29.12.95, Stephen Tweedie
  9. * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
  10. * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
  11. * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
  12. * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
  13. * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
  14. * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
  15. */
  16. #include <linux/stddef.h>
  17. #include <linux/mm.h>
  18. #include <linux/swap.h>
  19. #include <linux/interrupt.h>
  20. #include <linux/pagemap.h>
  21. #include <linux/bootmem.h>
  22. #include <linux/compiler.h>
  23. #include <linux/kernel.h>
  24. #include <linux/module.h>
  25. #include <linux/suspend.h>
  26. #include <linux/pagevec.h>
  27. #include <linux/blkdev.h>
  28. #include <linux/slab.h>
  29. #include <linux/notifier.h>
  30. #include <linux/topology.h>
  31. #include <linux/sysctl.h>
  32. #include <linux/cpu.h>
  33. #include <linux/cpuset.h>
  34. #include <linux/memory_hotplug.h>
  35. #include <linux/nodemask.h>
  36. #include <linux/vmalloc.h>
  37. #include <linux/mempolicy.h>
  38. #include <linux/stop_machine.h>
  39. #include <linux/sort.h>
  40. #include <linux/pfn.h>
  41. #include <linux/backing-dev.h>
  42. #include <linux/fault-inject.h>
  43. #include <asm/tlbflush.h>
  44. #include <asm/div64.h>
  45. #include "internal.h"
  46. /*
  47. * Array of node states.
  48. */
  49. nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
  50. [N_POSSIBLE] = NODE_MASK_ALL,
  51. [N_ONLINE] = { { [0] = 1UL } },
  52. #ifndef CONFIG_NUMA
  53. [N_NORMAL_MEMORY] = { { [0] = 1UL } },
  54. #ifdef CONFIG_HIGHMEM
  55. [N_HIGH_MEMORY] = { { [0] = 1UL } },
  56. #endif
  57. [N_CPU] = { { [0] = 1UL } },
  58. #endif /* NUMA */
  59. };
  60. EXPORT_SYMBOL(node_states);
  61. unsigned long totalram_pages __read_mostly;
  62. unsigned long totalreserve_pages __read_mostly;
  63. long nr_swap_pages;
  64. int percpu_pagelist_fraction;
  65. static void __free_pages_ok(struct page *page, unsigned int order);
  66. /*
  67. * results with 256, 32 in the lowmem_reserve sysctl:
  68. * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
  69. * 1G machine -> (16M dma, 784M normal, 224M high)
  70. * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
  71. * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
  72. * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
  73. *
  74. * TBD: should special case ZONE_DMA32 machines here - in those we normally
  75. * don't need any ZONE_NORMAL reservation
  76. */
  77. int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
  78. #ifdef CONFIG_ZONE_DMA
  79. 256,
  80. #endif
  81. #ifdef CONFIG_ZONE_DMA32
  82. 256,
  83. #endif
  84. #ifdef CONFIG_HIGHMEM
  85. 32,
  86. #endif
  87. 32,
  88. };
  89. EXPORT_SYMBOL(totalram_pages);
  90. static char * const zone_names[MAX_NR_ZONES] = {
  91. #ifdef CONFIG_ZONE_DMA
  92. "DMA",
  93. #endif
  94. #ifdef CONFIG_ZONE_DMA32
  95. "DMA32",
  96. #endif
  97. "Normal",
  98. #ifdef CONFIG_HIGHMEM
  99. "HighMem",
  100. #endif
  101. "Movable",
  102. };
  103. int min_free_kbytes = 1024;
  104. unsigned long __meminitdata nr_kernel_pages;
  105. unsigned long __meminitdata nr_all_pages;
  106. static unsigned long __meminitdata dma_reserve;
  107. #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
  108. /*
  109. * MAX_ACTIVE_REGIONS determines the maxmimum number of distinct
  110. * ranges of memory (RAM) that may be registered with add_active_range().
  111. * Ranges passed to add_active_range() will be merged if possible
  112. * so the number of times add_active_range() can be called is
  113. * related to the number of nodes and the number of holes
  114. */
  115. #ifdef CONFIG_MAX_ACTIVE_REGIONS
  116. /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
  117. #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
  118. #else
  119. #if MAX_NUMNODES >= 32
  120. /* If there can be many nodes, allow up to 50 holes per node */
  121. #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
  122. #else
  123. /* By default, allow up to 256 distinct regions */
  124. #define MAX_ACTIVE_REGIONS 256
  125. #endif
  126. #endif
  127. static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS];
  128. static int __meminitdata nr_nodemap_entries;
  129. static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
  130. static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
  131. #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
  132. static unsigned long __meminitdata node_boundary_start_pfn[MAX_NUMNODES];
  133. static unsigned long __meminitdata node_boundary_end_pfn[MAX_NUMNODES];
  134. #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
  135. unsigned long __initdata required_kernelcore;
  136. unsigned long __initdata required_movablecore;
  137. unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
  138. /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
  139. int movable_zone;
  140. EXPORT_SYMBOL(movable_zone);
  141. #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
  142. #if MAX_NUMNODES > 1
  143. int nr_node_ids __read_mostly = MAX_NUMNODES;
  144. EXPORT_SYMBOL(nr_node_ids);
  145. #endif
  146. int page_group_by_mobility_disabled __read_mostly;
  147. static inline int get_pageblock_migratetype(struct page *page)
  148. {
  149. if (unlikely(page_group_by_mobility_disabled))
  150. return MIGRATE_UNMOVABLE;
  151. return get_pageblock_flags_group(page, PB_migrate, PB_migrate_end);
  152. }
  153. static void set_pageblock_migratetype(struct page *page, int migratetype)
  154. {
  155. set_pageblock_flags_group(page, (unsigned long)migratetype,
  156. PB_migrate, PB_migrate_end);
  157. }
  158. static inline int allocflags_to_migratetype(gfp_t gfp_flags)
  159. {
  160. WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK);
  161. if (unlikely(page_group_by_mobility_disabled))
  162. return MIGRATE_UNMOVABLE;
  163. /* Cluster based on mobility */
  164. return (((gfp_flags & __GFP_MOVABLE) != 0) << 1) |
  165. ((gfp_flags & __GFP_RECLAIMABLE) != 0);
  166. }
  167. #ifdef CONFIG_DEBUG_VM
  168. static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
  169. {
  170. int ret = 0;
  171. unsigned seq;
  172. unsigned long pfn = page_to_pfn(page);
  173. do {
  174. seq = zone_span_seqbegin(zone);
  175. if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
  176. ret = 1;
  177. else if (pfn < zone->zone_start_pfn)
  178. ret = 1;
  179. } while (zone_span_seqretry(zone, seq));
  180. return ret;
  181. }
  182. static int page_is_consistent(struct zone *zone, struct page *page)
  183. {
  184. if (!pfn_valid_within(page_to_pfn(page)))
  185. return 0;
  186. if (zone != page_zone(page))
  187. return 0;
  188. return 1;
  189. }
  190. /*
  191. * Temporary debugging check for pages not lying within a given zone.
  192. */
  193. static int bad_range(struct zone *zone, struct page *page)
  194. {
  195. if (page_outside_zone_boundaries(zone, page))
  196. return 1;
  197. if (!page_is_consistent(zone, page))
  198. return 1;
  199. return 0;
  200. }
  201. #else
  202. static inline int bad_range(struct zone *zone, struct page *page)
  203. {
  204. return 0;
  205. }
  206. #endif
  207. static void bad_page(struct page *page)
  208. {
  209. printk(KERN_EMERG "Bad page state in process '%s'\n"
  210. KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
  211. KERN_EMERG "Trying to fix it up, but a reboot is needed\n"
  212. KERN_EMERG "Backtrace:\n",
  213. current->comm, page, (int)(2*sizeof(unsigned long)),
  214. (unsigned long)page->flags, page->mapping,
  215. page_mapcount(page), page_count(page));
  216. dump_stack();
  217. page->flags &= ~(1 << PG_lru |
  218. 1 << PG_private |
  219. 1 << PG_locked |
  220. 1 << PG_active |
  221. 1 << PG_dirty |
  222. 1 << PG_reclaim |
  223. 1 << PG_slab |
  224. 1 << PG_swapcache |
  225. 1 << PG_writeback |
  226. 1 << PG_buddy );
  227. set_page_count(page, 0);
  228. reset_page_mapcount(page);
  229. page->mapping = NULL;
  230. add_taint(TAINT_BAD_PAGE);
  231. }
  232. /*
  233. * Higher-order pages are called "compound pages". They are structured thusly:
  234. *
  235. * The first PAGE_SIZE page is called the "head page".
  236. *
  237. * The remaining PAGE_SIZE pages are called "tail pages".
  238. *
  239. * All pages have PG_compound set. All pages have their ->private pointing at
  240. * the head page (even the head page has this).
  241. *
  242. * The first tail page's ->lru.next holds the address of the compound page's
  243. * put_page() function. Its ->lru.prev holds the order of allocation.
  244. * This usage means that zero-order pages may not be compound.
  245. */
  246. static void free_compound_page(struct page *page)
  247. {
  248. __free_pages_ok(page, compound_order(page));
  249. }
  250. static void prep_compound_page(struct page *page, unsigned long order)
  251. {
  252. int i;
  253. int nr_pages = 1 << order;
  254. set_compound_page_dtor(page, free_compound_page);
  255. set_compound_order(page, order);
  256. __SetPageHead(page);
  257. for (i = 1; i < nr_pages; i++) {
  258. struct page *p = page + i;
  259. __SetPageTail(p);
  260. p->first_page = page;
  261. }
  262. }
  263. static void destroy_compound_page(struct page *page, unsigned long order)
  264. {
  265. int i;
  266. int nr_pages = 1 << order;
  267. if (unlikely(compound_order(page) != order))
  268. bad_page(page);
  269. if (unlikely(!PageHead(page)))
  270. bad_page(page);
  271. __ClearPageHead(page);
  272. for (i = 1; i < nr_pages; i++) {
  273. struct page *p = page + i;
  274. if (unlikely(!PageTail(p) |
  275. (p->first_page != page)))
  276. bad_page(page);
  277. __ClearPageTail(p);
  278. }
  279. }
  280. static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
  281. {
  282. int i;
  283. VM_BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM);
  284. /*
  285. * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
  286. * and __GFP_HIGHMEM from hard or soft interrupt context.
  287. */
  288. VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
  289. for (i = 0; i < (1 << order); i++)
  290. clear_highpage(page + i);
  291. }
  292. /*
  293. * function for dealing with page's order in buddy system.
  294. * zone->lock is already acquired when we use these.
  295. * So, we don't need atomic page->flags operations here.
  296. */
  297. static inline unsigned long page_order(struct page *page)
  298. {
  299. return page_private(page);
  300. }
  301. static inline void set_page_order(struct page *page, int order)
  302. {
  303. set_page_private(page, order);
  304. __SetPageBuddy(page);
  305. }
  306. static inline void rmv_page_order(struct page *page)
  307. {
  308. __ClearPageBuddy(page);
  309. set_page_private(page, 0);
  310. }
  311. /*
  312. * Locate the struct page for both the matching buddy in our
  313. * pair (buddy1) and the combined O(n+1) page they form (page).
  314. *
  315. * 1) Any buddy B1 will have an order O twin B2 which satisfies
  316. * the following equation:
  317. * B2 = B1 ^ (1 << O)
  318. * For example, if the starting buddy (buddy2) is #8 its order
  319. * 1 buddy is #10:
  320. * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
  321. *
  322. * 2) Any buddy B will have an order O+1 parent P which
  323. * satisfies the following equation:
  324. * P = B & ~(1 << O)
  325. *
  326. * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
  327. */
  328. static inline struct page *
  329. __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
  330. {
  331. unsigned long buddy_idx = page_idx ^ (1 << order);
  332. return page + (buddy_idx - page_idx);
  333. }
  334. static inline unsigned long
  335. __find_combined_index(unsigned long page_idx, unsigned int order)
  336. {
  337. return (page_idx & ~(1 << order));
  338. }
  339. /*
  340. * This function checks whether a page is free && is the buddy
  341. * we can do coalesce a page and its buddy if
  342. * (a) the buddy is not in a hole &&
  343. * (b) the buddy is in the buddy system &&
  344. * (c) a page and its buddy have the same order &&
  345. * (d) a page and its buddy are in the same zone.
  346. *
  347. * For recording whether a page is in the buddy system, we use PG_buddy.
  348. * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
  349. *
  350. * For recording page's order, we use page_private(page).
  351. */
  352. static inline int page_is_buddy(struct page *page, struct page *buddy,
  353. int order)
  354. {
  355. if (!pfn_valid_within(page_to_pfn(buddy)))
  356. return 0;
  357. if (page_zone_id(page) != page_zone_id(buddy))
  358. return 0;
  359. if (PageBuddy(buddy) && page_order(buddy) == order) {
  360. BUG_ON(page_count(buddy) != 0);
  361. return 1;
  362. }
  363. return 0;
  364. }
  365. /*
  366. * Freeing function for a buddy system allocator.
  367. *
  368. * The concept of a buddy system is to maintain direct-mapped table
  369. * (containing bit values) for memory blocks of various "orders".
  370. * The bottom level table contains the map for the smallest allocatable
  371. * units of memory (here, pages), and each level above it describes
  372. * pairs of units from the levels below, hence, "buddies".
  373. * At a high level, all that happens here is marking the table entry
  374. * at the bottom level available, and propagating the changes upward
  375. * as necessary, plus some accounting needed to play nicely with other
  376. * parts of the VM system.
  377. * At each level, we keep a list of pages, which are heads of continuous
  378. * free pages of length of (1 << order) and marked with PG_buddy. Page's
  379. * order is recorded in page_private(page) field.
  380. * So when we are allocating or freeing one, we can derive the state of the
  381. * other. That is, if we allocate a small block, and both were
  382. * free, the remainder of the region must be split into blocks.
  383. * If a block is freed, and its buddy is also free, then this
  384. * triggers coalescing into a block of larger size.
  385. *
  386. * -- wli
  387. */
  388. static inline void __free_one_page(struct page *page,
  389. struct zone *zone, unsigned int order)
  390. {
  391. unsigned long page_idx;
  392. int order_size = 1 << order;
  393. int migratetype = get_pageblock_migratetype(page);
  394. if (unlikely(PageCompound(page)))
  395. destroy_compound_page(page, order);
  396. page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
  397. VM_BUG_ON(page_idx & (order_size - 1));
  398. VM_BUG_ON(bad_range(zone, page));
  399. __mod_zone_page_state(zone, NR_FREE_PAGES, order_size);
  400. while (order < MAX_ORDER-1) {
  401. unsigned long combined_idx;
  402. struct page *buddy;
  403. buddy = __page_find_buddy(page, page_idx, order);
  404. if (!page_is_buddy(page, buddy, order))
  405. break; /* Move the buddy up one level. */
  406. list_del(&buddy->lru);
  407. zone->free_area[order].nr_free--;
  408. rmv_page_order(buddy);
  409. combined_idx = __find_combined_index(page_idx, order);
  410. page = page + (combined_idx - page_idx);
  411. page_idx = combined_idx;
  412. order++;
  413. }
  414. set_page_order(page, order);
  415. list_add(&page->lru,
  416. &zone->free_area[order].free_list[migratetype]);
  417. zone->free_area[order].nr_free++;
  418. }
  419. static inline int free_pages_check(struct page *page)
  420. {
  421. if (unlikely(page_mapcount(page) |
  422. (page->mapping != NULL) |
  423. (page_count(page) != 0) |
  424. (page->flags & (
  425. 1 << PG_lru |
  426. 1 << PG_private |
  427. 1 << PG_locked |
  428. 1 << PG_active |
  429. 1 << PG_slab |
  430. 1 << PG_swapcache |
  431. 1 << PG_writeback |
  432. 1 << PG_reserved |
  433. 1 << PG_buddy ))))
  434. bad_page(page);
  435. if (PageDirty(page))
  436. __ClearPageDirty(page);
  437. /*
  438. * For now, we report if PG_reserved was found set, but do not
  439. * clear it, and do not free the page. But we shall soon need
  440. * to do more, for when the ZERO_PAGE count wraps negative.
  441. */
  442. return PageReserved(page);
  443. }
  444. /*
  445. * Frees a list of pages.
  446. * Assumes all pages on list are in same zone, and of same order.
  447. * count is the number of pages to free.
  448. *
  449. * If the zone was previously in an "all pages pinned" state then look to
  450. * see if this freeing clears that state.
  451. *
  452. * And clear the zone's pages_scanned counter, to hold off the "all pages are
  453. * pinned" detection logic.
  454. */
  455. static void free_pages_bulk(struct zone *zone, int count,
  456. struct list_head *list, int order)
  457. {
  458. spin_lock(&zone->lock);
  459. zone->all_unreclaimable = 0;
  460. zone->pages_scanned = 0;
  461. while (count--) {
  462. struct page *page;
  463. VM_BUG_ON(list_empty(list));
  464. page = list_entry(list->prev, struct page, lru);
  465. /* have to delete it as __free_one_page list manipulates */
  466. list_del(&page->lru);
  467. __free_one_page(page, zone, order);
  468. }
  469. spin_unlock(&zone->lock);
  470. }
  471. static void free_one_page(struct zone *zone, struct page *page, int order)
  472. {
  473. spin_lock(&zone->lock);
  474. zone->all_unreclaimable = 0;
  475. zone->pages_scanned = 0;
  476. __free_one_page(page, zone, order);
  477. spin_unlock(&zone->lock);
  478. }
  479. static void __free_pages_ok(struct page *page, unsigned int order)
  480. {
  481. unsigned long flags;
  482. int i;
  483. int reserved = 0;
  484. for (i = 0 ; i < (1 << order) ; ++i)
  485. reserved += free_pages_check(page + i);
  486. if (reserved)
  487. return;
  488. if (!PageHighMem(page))
  489. debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
  490. arch_free_page(page, order);
  491. kernel_map_pages(page, 1 << order, 0);
  492. local_irq_save(flags);
  493. __count_vm_events(PGFREE, 1 << order);
  494. free_one_page(page_zone(page), page, order);
  495. local_irq_restore(flags);
  496. }
  497. /*
  498. * permit the bootmem allocator to evade page validation on high-order frees
  499. */
  500. void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order)
  501. {
  502. if (order == 0) {
  503. __ClearPageReserved(page);
  504. set_page_count(page, 0);
  505. set_page_refcounted(page);
  506. __free_page(page);
  507. } else {
  508. int loop;
  509. prefetchw(page);
  510. for (loop = 0; loop < BITS_PER_LONG; loop++) {
  511. struct page *p = &page[loop];
  512. if (loop + 1 < BITS_PER_LONG)
  513. prefetchw(p + 1);
  514. __ClearPageReserved(p);
  515. set_page_count(p, 0);
  516. }
  517. set_page_refcounted(page);
  518. __free_pages(page, order);
  519. }
  520. }
  521. /*
  522. * The order of subdivision here is critical for the IO subsystem.
  523. * Please do not alter this order without good reasons and regression
  524. * testing. Specifically, as large blocks of memory are subdivided,
  525. * the order in which smaller blocks are delivered depends on the order
  526. * they're subdivided in this function. This is the primary factor
  527. * influencing the order in which pages are delivered to the IO
  528. * subsystem according to empirical testing, and this is also justified
  529. * by considering the behavior of a buddy system containing a single
  530. * large block of memory acted on by a series of small allocations.
  531. * This behavior is a critical factor in sglist merging's success.
  532. *
  533. * -- wli
  534. */
  535. static inline void expand(struct zone *zone, struct page *page,
  536. int low, int high, struct free_area *area,
  537. int migratetype)
  538. {
  539. unsigned long size = 1 << high;
  540. while (high > low) {
  541. area--;
  542. high--;
  543. size >>= 1;
  544. VM_BUG_ON(bad_range(zone, &page[size]));
  545. list_add(&page[size].lru, &area->free_list[migratetype]);
  546. area->nr_free++;
  547. set_page_order(&page[size], high);
  548. }
  549. }
  550. /*
  551. * This page is about to be returned from the page allocator
  552. */
  553. static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
  554. {
  555. if (unlikely(page_mapcount(page) |
  556. (page->mapping != NULL) |
  557. (page_count(page) != 0) |
  558. (page->flags & (
  559. 1 << PG_lru |
  560. 1 << PG_private |
  561. 1 << PG_locked |
  562. 1 << PG_active |
  563. 1 << PG_dirty |
  564. 1 << PG_slab |
  565. 1 << PG_swapcache |
  566. 1 << PG_writeback |
  567. 1 << PG_reserved |
  568. 1 << PG_buddy ))))
  569. bad_page(page);
  570. /*
  571. * For now, we report if PG_reserved was found set, but do not
  572. * clear it, and do not allocate the page: as a safety net.
  573. */
  574. if (PageReserved(page))
  575. return 1;
  576. page->flags &= ~(1 << PG_uptodate | 1 << PG_error | 1 << PG_readahead |
  577. 1 << PG_referenced | 1 << PG_arch_1 |
  578. 1 << PG_owner_priv_1 | 1 << PG_mappedtodisk);
  579. set_page_private(page, 0);
  580. set_page_refcounted(page);
  581. arch_alloc_page(page, order);
  582. kernel_map_pages(page, 1 << order, 1);
  583. if (gfp_flags & __GFP_ZERO)
  584. prep_zero_page(page, order, gfp_flags);
  585. if (order && (gfp_flags & __GFP_COMP))
  586. prep_compound_page(page, order);
  587. return 0;
  588. }
  589. /*
  590. * Go through the free lists for the given migratetype and remove
  591. * the smallest available page from the freelists
  592. */
  593. static struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
  594. int migratetype)
  595. {
  596. unsigned int current_order;
  597. struct free_area * area;
  598. struct page *page;
  599. /* Find a page of the appropriate size in the preferred list */
  600. for (current_order = order; current_order < MAX_ORDER; ++current_order) {
  601. area = &(zone->free_area[current_order]);
  602. if (list_empty(&area->free_list[migratetype]))
  603. continue;
  604. page = list_entry(area->free_list[migratetype].next,
  605. struct page, lru);
  606. list_del(&page->lru);
  607. rmv_page_order(page);
  608. area->nr_free--;
  609. __mod_zone_page_state(zone, NR_FREE_PAGES, - (1UL << order));
  610. expand(zone, page, order, current_order, area, migratetype);
  611. return page;
  612. }
  613. return NULL;
  614. }
  615. /*
  616. * This array describes the order lists are fallen back to when
  617. * the free lists for the desirable migrate type are depleted
  618. */
  619. static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
  620. [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
  621. [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
  622. [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
  623. [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
  624. };
  625. /*
  626. * Move the free pages in a range to the free lists of the requested type.
  627. * Note that start_page and end_pages are not aligned in a MAX_ORDER_NR_PAGES
  628. * boundary. If alignment is required, use move_freepages_block()
  629. */
  630. int move_freepages(struct zone *zone,
  631. struct page *start_page, struct page *end_page,
  632. int migratetype)
  633. {
  634. struct page *page;
  635. unsigned long order;
  636. int blocks_moved = 0;
  637. #ifndef CONFIG_HOLES_IN_ZONE
  638. /*
  639. * page_zone is not safe to call in this context when
  640. * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
  641. * anyway as we check zone boundaries in move_freepages_block().
  642. * Remove at a later date when no bug reports exist related to
  643. * grouping pages by mobility
  644. */
  645. BUG_ON(page_zone(start_page) != page_zone(end_page));
  646. #endif
  647. for (page = start_page; page <= end_page;) {
  648. if (!pfn_valid_within(page_to_pfn(page))) {
  649. page++;
  650. continue;
  651. }
  652. if (!PageBuddy(page)) {
  653. page++;
  654. continue;
  655. }
  656. order = page_order(page);
  657. list_del(&page->lru);
  658. list_add(&page->lru,
  659. &zone->free_area[order].free_list[migratetype]);
  660. page += 1 << order;
  661. blocks_moved++;
  662. }
  663. return blocks_moved;
  664. }
  665. int move_freepages_block(struct zone *zone, struct page *page, int migratetype)
  666. {
  667. unsigned long start_pfn, end_pfn;
  668. struct page *start_page, *end_page;
  669. start_pfn = page_to_pfn(page);
  670. start_pfn = start_pfn & ~(MAX_ORDER_NR_PAGES-1);
  671. start_page = pfn_to_page(start_pfn);
  672. end_page = start_page + MAX_ORDER_NR_PAGES - 1;
  673. end_pfn = start_pfn + MAX_ORDER_NR_PAGES - 1;
  674. /* Do not cross zone boundaries */
  675. if (start_pfn < zone->zone_start_pfn)
  676. start_page = page;
  677. if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
  678. return 0;
  679. return move_freepages(zone, start_page, end_page, migratetype);
  680. }
  681. /* Return the page with the lowest PFN in the list */
  682. static struct page *min_page(struct list_head *list)
  683. {
  684. unsigned long min_pfn = -1UL;
  685. struct page *min_page = NULL, *page;;
  686. list_for_each_entry(page, list, lru) {
  687. unsigned long pfn = page_to_pfn(page);
  688. if (pfn < min_pfn) {
  689. min_pfn = pfn;
  690. min_page = page;
  691. }
  692. }
  693. return min_page;
  694. }
  695. /* Remove an element from the buddy allocator from the fallback list */
  696. static struct page *__rmqueue_fallback(struct zone *zone, int order,
  697. int start_migratetype)
  698. {
  699. struct free_area * area;
  700. int current_order;
  701. struct page *page;
  702. int migratetype, i;
  703. /* Find the largest possible block of pages in the other list */
  704. for (current_order = MAX_ORDER-1; current_order >= order;
  705. --current_order) {
  706. for (i = 0; i < MIGRATE_TYPES - 1; i++) {
  707. migratetype = fallbacks[start_migratetype][i];
  708. /* MIGRATE_RESERVE handled later if necessary */
  709. if (migratetype == MIGRATE_RESERVE)
  710. continue;
  711. area = &(zone->free_area[current_order]);
  712. if (list_empty(&area->free_list[migratetype]))
  713. continue;
  714. /* Bias kernel allocations towards low pfns */
  715. page = list_entry(area->free_list[migratetype].next,
  716. struct page, lru);
  717. if (unlikely(start_migratetype != MIGRATE_MOVABLE))
  718. page = min_page(&area->free_list[migratetype]);
  719. area->nr_free--;
  720. /*
  721. * If breaking a large block of pages, move all free
  722. * pages to the preferred allocation list. If falling
  723. * back for a reclaimable kernel allocation, be more
  724. * agressive about taking ownership of free pages
  725. */
  726. if (unlikely(current_order >= MAX_ORDER / 2) ||
  727. start_migratetype == MIGRATE_RECLAIMABLE) {
  728. unsigned long pages;
  729. pages = move_freepages_block(zone, page,
  730. start_migratetype);
  731. /* Claim the whole block if over half of it is free */
  732. if ((pages << current_order) >= (1 << (MAX_ORDER-2)))
  733. set_pageblock_migratetype(page,
  734. start_migratetype);
  735. migratetype = start_migratetype;
  736. }
  737. /* Remove the page from the freelists */
  738. list_del(&page->lru);
  739. rmv_page_order(page);
  740. __mod_zone_page_state(zone, NR_FREE_PAGES,
  741. -(1UL << order));
  742. if (current_order == MAX_ORDER - 1)
  743. set_pageblock_migratetype(page,
  744. start_migratetype);
  745. expand(zone, page, order, current_order, area, migratetype);
  746. return page;
  747. }
  748. }
  749. /* Use MIGRATE_RESERVE rather than fail an allocation */
  750. return __rmqueue_smallest(zone, order, MIGRATE_RESERVE);
  751. }
  752. /*
  753. * Do the hard work of removing an element from the buddy allocator.
  754. * Call me with the zone->lock already held.
  755. */
  756. static struct page *__rmqueue(struct zone *zone, unsigned int order,
  757. int migratetype)
  758. {
  759. struct page *page;
  760. page = __rmqueue_smallest(zone, order, migratetype);
  761. if (unlikely(!page))
  762. page = __rmqueue_fallback(zone, order, migratetype);
  763. return page;
  764. }
  765. /*
  766. * Obtain a specified number of elements from the buddy allocator, all under
  767. * a single hold of the lock, for efficiency. Add them to the supplied list.
  768. * Returns the number of new pages which were placed at *list.
  769. */
  770. static int rmqueue_bulk(struct zone *zone, unsigned int order,
  771. unsigned long count, struct list_head *list,
  772. int migratetype)
  773. {
  774. int i;
  775. spin_lock(&zone->lock);
  776. for (i = 0; i < count; ++i) {
  777. struct page *page = __rmqueue(zone, order, migratetype);
  778. if (unlikely(page == NULL))
  779. break;
  780. list_add(&page->lru, list);
  781. set_page_private(page, migratetype);
  782. }
  783. spin_unlock(&zone->lock);
  784. return i;
  785. }
  786. #ifdef CONFIG_NUMA
  787. /*
  788. * Called from the vmstat counter updater to drain pagesets of this
  789. * currently executing processor on remote nodes after they have
  790. * expired.
  791. *
  792. * Note that this function must be called with the thread pinned to
  793. * a single processor.
  794. */
  795. void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
  796. {
  797. unsigned long flags;
  798. int to_drain;
  799. local_irq_save(flags);
  800. if (pcp->count >= pcp->batch)
  801. to_drain = pcp->batch;
  802. else
  803. to_drain = pcp->count;
  804. free_pages_bulk(zone, to_drain, &pcp->list, 0);
  805. pcp->count -= to_drain;
  806. local_irq_restore(flags);
  807. }
  808. #endif
  809. static void __drain_pages(unsigned int cpu)
  810. {
  811. unsigned long flags;
  812. struct zone *zone;
  813. int i;
  814. for_each_zone(zone) {
  815. struct per_cpu_pageset *pset;
  816. if (!populated_zone(zone))
  817. continue;
  818. pset = zone_pcp(zone, cpu);
  819. for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
  820. struct per_cpu_pages *pcp;
  821. pcp = &pset->pcp[i];
  822. local_irq_save(flags);
  823. free_pages_bulk(zone, pcp->count, &pcp->list, 0);
  824. pcp->count = 0;
  825. local_irq_restore(flags);
  826. }
  827. }
  828. }
  829. #ifdef CONFIG_HIBERNATION
  830. void mark_free_pages(struct zone *zone)
  831. {
  832. unsigned long pfn, max_zone_pfn;
  833. unsigned long flags;
  834. int order, t;
  835. struct list_head *curr;
  836. if (!zone->spanned_pages)
  837. return;
  838. spin_lock_irqsave(&zone->lock, flags);
  839. max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
  840. for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
  841. if (pfn_valid(pfn)) {
  842. struct page *page = pfn_to_page(pfn);
  843. if (!swsusp_page_is_forbidden(page))
  844. swsusp_unset_page_free(page);
  845. }
  846. for_each_migratetype_order(order, t) {
  847. list_for_each(curr, &zone->free_area[order].free_list[t]) {
  848. unsigned long i;
  849. pfn = page_to_pfn(list_entry(curr, struct page, lru));
  850. for (i = 0; i < (1UL << order); i++)
  851. swsusp_set_page_free(pfn_to_page(pfn + i));
  852. }
  853. }
  854. spin_unlock_irqrestore(&zone->lock, flags);
  855. }
  856. #endif /* CONFIG_PM */
  857. /*
  858. * Spill all of this CPU's per-cpu pages back into the buddy allocator.
  859. */
  860. void drain_local_pages(void)
  861. {
  862. unsigned long flags;
  863. local_irq_save(flags);
  864. __drain_pages(smp_processor_id());
  865. local_irq_restore(flags);
  866. }
  867. void smp_drain_local_pages(void *arg)
  868. {
  869. drain_local_pages();
  870. }
  871. /*
  872. * Spill all the per-cpu pages from all CPUs back into the buddy allocator
  873. */
  874. void drain_all_local_pages(void)
  875. {
  876. unsigned long flags;
  877. local_irq_save(flags);
  878. __drain_pages(smp_processor_id());
  879. local_irq_restore(flags);
  880. smp_call_function(smp_drain_local_pages, NULL, 0, 1);
  881. }
  882. /*
  883. * Free a 0-order page
  884. */
  885. static void fastcall free_hot_cold_page(struct page *page, int cold)
  886. {
  887. struct zone *zone = page_zone(page);
  888. struct per_cpu_pages *pcp;
  889. unsigned long flags;
  890. if (PageAnon(page))
  891. page->mapping = NULL;
  892. if (free_pages_check(page))
  893. return;
  894. if (!PageHighMem(page))
  895. debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
  896. arch_free_page(page, 0);
  897. kernel_map_pages(page, 1, 0);
  898. pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
  899. local_irq_save(flags);
  900. __count_vm_event(PGFREE);
  901. list_add(&page->lru, &pcp->list);
  902. set_page_private(page, get_pageblock_migratetype(page));
  903. pcp->count++;
  904. if (pcp->count >= pcp->high) {
  905. free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
  906. pcp->count -= pcp->batch;
  907. }
  908. local_irq_restore(flags);
  909. put_cpu();
  910. }
  911. void fastcall free_hot_page(struct page *page)
  912. {
  913. free_hot_cold_page(page, 0);
  914. }
  915. void fastcall free_cold_page(struct page *page)
  916. {
  917. free_hot_cold_page(page, 1);
  918. }
  919. /*
  920. * split_page takes a non-compound higher-order page, and splits it into
  921. * n (1<<order) sub-pages: page[0..n]
  922. * Each sub-page must be freed individually.
  923. *
  924. * Note: this is probably too low level an operation for use in drivers.
  925. * Please consult with lkml before using this in your driver.
  926. */
  927. void split_page(struct page *page, unsigned int order)
  928. {
  929. int i;
  930. VM_BUG_ON(PageCompound(page));
  931. VM_BUG_ON(!page_count(page));
  932. for (i = 1; i < (1 << order); i++)
  933. set_page_refcounted(page + i);
  934. }
  935. /*
  936. * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
  937. * we cheat by calling it from here, in the order > 0 path. Saves a branch
  938. * or two.
  939. */
  940. static struct page *buffered_rmqueue(struct zonelist *zonelist,
  941. struct zone *zone, int order, gfp_t gfp_flags)
  942. {
  943. unsigned long flags;
  944. struct page *page;
  945. int cold = !!(gfp_flags & __GFP_COLD);
  946. int cpu;
  947. int migratetype = allocflags_to_migratetype(gfp_flags);
  948. again:
  949. cpu = get_cpu();
  950. if (likely(order == 0)) {
  951. struct per_cpu_pages *pcp;
  952. pcp = &zone_pcp(zone, cpu)->pcp[cold];
  953. local_irq_save(flags);
  954. if (!pcp->count) {
  955. pcp->count = rmqueue_bulk(zone, 0,
  956. pcp->batch, &pcp->list, migratetype);
  957. if (unlikely(!pcp->count))
  958. goto failed;
  959. }
  960. /* Find a page of the appropriate migrate type */
  961. list_for_each_entry(page, &pcp->list, lru)
  962. if (page_private(page) == migratetype)
  963. break;
  964. /* Allocate more to the pcp list if necessary */
  965. if (unlikely(&page->lru == &pcp->list)) {
  966. pcp->count += rmqueue_bulk(zone, 0,
  967. pcp->batch, &pcp->list, migratetype);
  968. page = list_entry(pcp->list.next, struct page, lru);
  969. }
  970. list_del(&page->lru);
  971. pcp->count--;
  972. } else {
  973. spin_lock_irqsave(&zone->lock, flags);
  974. page = __rmqueue(zone, order, migratetype);
  975. spin_unlock(&zone->lock);
  976. if (!page)
  977. goto failed;
  978. }
  979. __count_zone_vm_events(PGALLOC, zone, 1 << order);
  980. zone_statistics(zonelist, zone);
  981. local_irq_restore(flags);
  982. put_cpu();
  983. VM_BUG_ON(bad_range(zone, page));
  984. if (prep_new_page(page, order, gfp_flags))
  985. goto again;
  986. return page;
  987. failed:
  988. local_irq_restore(flags);
  989. put_cpu();
  990. return NULL;
  991. }
  992. #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
  993. #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
  994. #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
  995. #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
  996. #define ALLOC_HARDER 0x10 /* try to alloc harder */
  997. #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
  998. #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
  999. #ifdef CONFIG_FAIL_PAGE_ALLOC
  1000. static struct fail_page_alloc_attr {
  1001. struct fault_attr attr;
  1002. u32 ignore_gfp_highmem;
  1003. u32 ignore_gfp_wait;
  1004. u32 min_order;
  1005. #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
  1006. struct dentry *ignore_gfp_highmem_file;
  1007. struct dentry *ignore_gfp_wait_file;
  1008. struct dentry *min_order_file;
  1009. #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
  1010. } fail_page_alloc = {
  1011. .attr = FAULT_ATTR_INITIALIZER,
  1012. .ignore_gfp_wait = 1,
  1013. .ignore_gfp_highmem = 1,
  1014. .min_order = 1,
  1015. };
  1016. static int __init setup_fail_page_alloc(char *str)
  1017. {
  1018. return setup_fault_attr(&fail_page_alloc.attr, str);
  1019. }
  1020. __setup("fail_page_alloc=", setup_fail_page_alloc);
  1021. static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
  1022. {
  1023. if (order < fail_page_alloc.min_order)
  1024. return 0;
  1025. if (gfp_mask & __GFP_NOFAIL)
  1026. return 0;
  1027. if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
  1028. return 0;
  1029. if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
  1030. return 0;
  1031. return should_fail(&fail_page_alloc.attr, 1 << order);
  1032. }
  1033. #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
  1034. static int __init fail_page_alloc_debugfs(void)
  1035. {
  1036. mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
  1037. struct dentry *dir;
  1038. int err;
  1039. err = init_fault_attr_dentries(&fail_page_alloc.attr,
  1040. "fail_page_alloc");
  1041. if (err)
  1042. return err;
  1043. dir = fail_page_alloc.attr.dentries.dir;
  1044. fail_page_alloc.ignore_gfp_wait_file =
  1045. debugfs_create_bool("ignore-gfp-wait", mode, dir,
  1046. &fail_page_alloc.ignore_gfp_wait);
  1047. fail_page_alloc.ignore_gfp_highmem_file =
  1048. debugfs_create_bool("ignore-gfp-highmem", mode, dir,
  1049. &fail_page_alloc.ignore_gfp_highmem);
  1050. fail_page_alloc.min_order_file =
  1051. debugfs_create_u32("min-order", mode, dir,
  1052. &fail_page_alloc.min_order);
  1053. if (!fail_page_alloc.ignore_gfp_wait_file ||
  1054. !fail_page_alloc.ignore_gfp_highmem_file ||
  1055. !fail_page_alloc.min_order_file) {
  1056. err = -ENOMEM;
  1057. debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
  1058. debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
  1059. debugfs_remove(fail_page_alloc.min_order_file);
  1060. cleanup_fault_attr_dentries(&fail_page_alloc.attr);
  1061. }
  1062. return err;
  1063. }
  1064. late_initcall(fail_page_alloc_debugfs);
  1065. #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
  1066. #else /* CONFIG_FAIL_PAGE_ALLOC */
  1067. static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
  1068. {
  1069. return 0;
  1070. }
  1071. #endif /* CONFIG_FAIL_PAGE_ALLOC */
  1072. /*
  1073. * Return 1 if free pages are above 'mark'. This takes into account the order
  1074. * of the allocation.
  1075. */
  1076. int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
  1077. int classzone_idx, int alloc_flags)
  1078. {
  1079. /* free_pages my go negative - that's OK */
  1080. long min = mark;
  1081. long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1;
  1082. int o;
  1083. if (alloc_flags & ALLOC_HIGH)
  1084. min -= min / 2;
  1085. if (alloc_flags & ALLOC_HARDER)
  1086. min -= min / 4;
  1087. if (free_pages <= min + z->lowmem_reserve[classzone_idx])
  1088. return 0;
  1089. for (o = 0; o < order; o++) {
  1090. /* At the next order, this order's pages become unavailable */
  1091. free_pages -= z->free_area[o].nr_free << o;
  1092. /* Require fewer higher order pages to be free */
  1093. min >>= 1;
  1094. if (free_pages <= min)
  1095. return 0;
  1096. }
  1097. return 1;
  1098. }
  1099. #ifdef CONFIG_NUMA
  1100. /*
  1101. * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
  1102. * skip over zones that are not allowed by the cpuset, or that have
  1103. * been recently (in last second) found to be nearly full. See further
  1104. * comments in mmzone.h. Reduces cache footprint of zonelist scans
  1105. * that have to skip over alot of full or unallowed zones.
  1106. *
  1107. * If the zonelist cache is present in the passed in zonelist, then
  1108. * returns a pointer to the allowed node mask (either the current
  1109. * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
  1110. *
  1111. * If the zonelist cache is not available for this zonelist, does
  1112. * nothing and returns NULL.
  1113. *
  1114. * If the fullzones BITMAP in the zonelist cache is stale (more than
  1115. * a second since last zap'd) then we zap it out (clear its bits.)
  1116. *
  1117. * We hold off even calling zlc_setup, until after we've checked the
  1118. * first zone in the zonelist, on the theory that most allocations will
  1119. * be satisfied from that first zone, so best to examine that zone as
  1120. * quickly as we can.
  1121. */
  1122. static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
  1123. {
  1124. struct zonelist_cache *zlc; /* cached zonelist speedup info */
  1125. nodemask_t *allowednodes; /* zonelist_cache approximation */
  1126. zlc = zonelist->zlcache_ptr;
  1127. if (!zlc)
  1128. return NULL;
  1129. if (jiffies - zlc->last_full_zap > 1 * HZ) {
  1130. bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
  1131. zlc->last_full_zap = jiffies;
  1132. }
  1133. allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
  1134. &cpuset_current_mems_allowed :
  1135. &node_states[N_HIGH_MEMORY];
  1136. return allowednodes;
  1137. }
  1138. /*
  1139. * Given 'z' scanning a zonelist, run a couple of quick checks to see
  1140. * if it is worth looking at further for free memory:
  1141. * 1) Check that the zone isn't thought to be full (doesn't have its
  1142. * bit set in the zonelist_cache fullzones BITMAP).
  1143. * 2) Check that the zones node (obtained from the zonelist_cache
  1144. * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
  1145. * Return true (non-zero) if zone is worth looking at further, or
  1146. * else return false (zero) if it is not.
  1147. *
  1148. * This check -ignores- the distinction between various watermarks,
  1149. * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
  1150. * found to be full for any variation of these watermarks, it will
  1151. * be considered full for up to one second by all requests, unless
  1152. * we are so low on memory on all allowed nodes that we are forced
  1153. * into the second scan of the zonelist.
  1154. *
  1155. * In the second scan we ignore this zonelist cache and exactly
  1156. * apply the watermarks to all zones, even it is slower to do so.
  1157. * We are low on memory in the second scan, and should leave no stone
  1158. * unturned looking for a free page.
  1159. */
  1160. static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
  1161. nodemask_t *allowednodes)
  1162. {
  1163. struct zonelist_cache *zlc; /* cached zonelist speedup info */
  1164. int i; /* index of *z in zonelist zones */
  1165. int n; /* node that zone *z is on */
  1166. zlc = zonelist->zlcache_ptr;
  1167. if (!zlc)
  1168. return 1;
  1169. i = z - zonelist->zones;
  1170. n = zlc->z_to_n[i];
  1171. /* This zone is worth trying if it is allowed but not full */
  1172. return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
  1173. }
  1174. /*
  1175. * Given 'z' scanning a zonelist, set the corresponding bit in
  1176. * zlc->fullzones, so that subsequent attempts to allocate a page
  1177. * from that zone don't waste time re-examining it.
  1178. */
  1179. static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
  1180. {
  1181. struct zonelist_cache *zlc; /* cached zonelist speedup info */
  1182. int i; /* index of *z in zonelist zones */
  1183. zlc = zonelist->zlcache_ptr;
  1184. if (!zlc)
  1185. return;
  1186. i = z - zonelist->zones;
  1187. set_bit(i, zlc->fullzones);
  1188. }
  1189. #else /* CONFIG_NUMA */
  1190. static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
  1191. {
  1192. return NULL;
  1193. }
  1194. static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
  1195. nodemask_t *allowednodes)
  1196. {
  1197. return 1;
  1198. }
  1199. static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
  1200. {
  1201. }
  1202. #endif /* CONFIG_NUMA */
  1203. /*
  1204. * get_page_from_freelist goes through the zonelist trying to allocate
  1205. * a page.
  1206. */
  1207. static struct page *
  1208. get_page_from_freelist(gfp_t gfp_mask, unsigned int order,
  1209. struct zonelist *zonelist, int alloc_flags)
  1210. {
  1211. struct zone **z;
  1212. struct page *page = NULL;
  1213. int classzone_idx = zone_idx(zonelist->zones[0]);
  1214. struct zone *zone;
  1215. nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
  1216. int zlc_active = 0; /* set if using zonelist_cache */
  1217. int did_zlc_setup = 0; /* just call zlc_setup() one time */
  1218. enum zone_type highest_zoneidx = -1; /* Gets set for policy zonelists */
  1219. zonelist_scan:
  1220. /*
  1221. * Scan zonelist, looking for a zone with enough free.
  1222. * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
  1223. */
  1224. z = zonelist->zones;
  1225. do {
  1226. /*
  1227. * In NUMA, this could be a policy zonelist which contains
  1228. * zones that may not be allowed by the current gfp_mask.
  1229. * Check the zone is allowed by the current flags
  1230. */
  1231. if (unlikely(alloc_should_filter_zonelist(zonelist))) {
  1232. if (highest_zoneidx == -1)
  1233. highest_zoneidx = gfp_zone(gfp_mask);
  1234. if (zone_idx(*z) > highest_zoneidx)
  1235. continue;
  1236. }
  1237. if (NUMA_BUILD && zlc_active &&
  1238. !zlc_zone_worth_trying(zonelist, z, allowednodes))
  1239. continue;
  1240. zone = *z;
  1241. if ((alloc_flags & ALLOC_CPUSET) &&
  1242. !cpuset_zone_allowed_softwall(zone, gfp_mask))
  1243. goto try_next_zone;
  1244. if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
  1245. unsigned long mark;
  1246. if (alloc_flags & ALLOC_WMARK_MIN)
  1247. mark = zone->pages_min;
  1248. else if (alloc_flags & ALLOC_WMARK_LOW)
  1249. mark = zone->pages_low;
  1250. else
  1251. mark = zone->pages_high;
  1252. if (!zone_watermark_ok(zone, order, mark,
  1253. classzone_idx, alloc_flags)) {
  1254. if (!zone_reclaim_mode ||
  1255. !zone_reclaim(zone, gfp_mask, order))
  1256. goto this_zone_full;
  1257. }
  1258. }
  1259. page = buffered_rmqueue(zonelist, zone, order, gfp_mask);
  1260. if (page)
  1261. break;
  1262. this_zone_full:
  1263. if (NUMA_BUILD)
  1264. zlc_mark_zone_full(zonelist, z);
  1265. try_next_zone:
  1266. if (NUMA_BUILD && !did_zlc_setup) {
  1267. /* we do zlc_setup after the first zone is tried */
  1268. allowednodes = zlc_setup(zonelist, alloc_flags);
  1269. zlc_active = 1;
  1270. did_zlc_setup = 1;
  1271. }
  1272. } while (*(++z) != NULL);
  1273. if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
  1274. /* Disable zlc cache for second zonelist scan */
  1275. zlc_active = 0;
  1276. goto zonelist_scan;
  1277. }
  1278. return page;
  1279. }
  1280. /*
  1281. * This is the 'heart' of the zoned buddy allocator.
  1282. */
  1283. struct page * fastcall
  1284. __alloc_pages(gfp_t gfp_mask, unsigned int order,
  1285. struct zonelist *zonelist)
  1286. {
  1287. const gfp_t wait = gfp_mask & __GFP_WAIT;
  1288. struct zone **z;
  1289. struct page *page;
  1290. struct reclaim_state reclaim_state;
  1291. struct task_struct *p = current;
  1292. int do_retry;
  1293. int alloc_flags;
  1294. int did_some_progress;
  1295. might_sleep_if(wait);
  1296. if (should_fail_alloc_page(gfp_mask, order))
  1297. return NULL;
  1298. restart:
  1299. z = zonelist->zones; /* the list of zones suitable for gfp_mask */
  1300. if (unlikely(*z == NULL)) {
  1301. /*
  1302. * Happens if we have an empty zonelist as a result of
  1303. * GFP_THISNODE being used on a memoryless node
  1304. */
  1305. return NULL;
  1306. }
  1307. page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
  1308. zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET);
  1309. if (page)
  1310. goto got_pg;
  1311. /*
  1312. * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
  1313. * __GFP_NOWARN set) should not cause reclaim since the subsystem
  1314. * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
  1315. * using a larger set of nodes after it has established that the
  1316. * allowed per node queues are empty and that nodes are
  1317. * over allocated.
  1318. */
  1319. if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
  1320. goto nopage;
  1321. for (z = zonelist->zones; *z; z++)
  1322. wakeup_kswapd(*z, order);
  1323. /*
  1324. * OK, we're below the kswapd watermark and have kicked background
  1325. * reclaim. Now things get more complex, so set up alloc_flags according
  1326. * to how we want to proceed.
  1327. *
  1328. * The caller may dip into page reserves a bit more if the caller
  1329. * cannot run direct reclaim, or if the caller has realtime scheduling
  1330. * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
  1331. * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
  1332. */
  1333. alloc_flags = ALLOC_WMARK_MIN;
  1334. if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
  1335. alloc_flags |= ALLOC_HARDER;
  1336. if (gfp_mask & __GFP_HIGH)
  1337. alloc_flags |= ALLOC_HIGH;
  1338. if (wait)
  1339. alloc_flags |= ALLOC_CPUSET;
  1340. /*
  1341. * Go through the zonelist again. Let __GFP_HIGH and allocations
  1342. * coming from realtime tasks go deeper into reserves.
  1343. *
  1344. * This is the last chance, in general, before the goto nopage.
  1345. * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
  1346. * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
  1347. */
  1348. page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
  1349. if (page)
  1350. goto got_pg;
  1351. /* This allocation should allow future memory freeing. */
  1352. rebalance:
  1353. if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
  1354. && !in_interrupt()) {
  1355. if (!(gfp_mask & __GFP_NOMEMALLOC)) {
  1356. nofail_alloc:
  1357. /* go through the zonelist yet again, ignoring mins */
  1358. page = get_page_from_freelist(gfp_mask, order,
  1359. zonelist, ALLOC_NO_WATERMARKS);
  1360. if (page)
  1361. goto got_pg;
  1362. if (gfp_mask & __GFP_NOFAIL) {
  1363. congestion_wait(WRITE, HZ/50);
  1364. goto nofail_alloc;
  1365. }
  1366. }
  1367. goto nopage;
  1368. }
  1369. /* Atomic allocations - we can't balance anything */
  1370. if (!wait)
  1371. goto nopage;
  1372. cond_resched();
  1373. /* We now go into synchronous reclaim */
  1374. cpuset_memory_pressure_bump();
  1375. p->flags |= PF_MEMALLOC;
  1376. reclaim_state.reclaimed_slab = 0;
  1377. p->reclaim_state = &reclaim_state;
  1378. did_some_progress = try_to_free_pages(zonelist->zones, order, gfp_mask);
  1379. p->reclaim_state = NULL;
  1380. p->flags &= ~PF_MEMALLOC;
  1381. cond_resched();
  1382. if (order != 0)
  1383. drain_all_local_pages();
  1384. if (likely(did_some_progress)) {
  1385. page = get_page_from_freelist(gfp_mask, order,
  1386. zonelist, alloc_flags);
  1387. if (page)
  1388. goto got_pg;
  1389. } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
  1390. /*
  1391. * Go through the zonelist yet one more time, keep
  1392. * very high watermark here, this is only to catch
  1393. * a parallel oom killing, we must fail if we're still
  1394. * under heavy pressure.
  1395. */
  1396. page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
  1397. zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET);
  1398. if (page)
  1399. goto got_pg;
  1400. /* The OOM killer will not help higher order allocs so fail */
  1401. if (order > PAGE_ALLOC_COSTLY_ORDER)
  1402. goto nopage;
  1403. out_of_memory(zonelist, gfp_mask, order);
  1404. goto restart;
  1405. }
  1406. /*
  1407. * Don't let big-order allocations loop unless the caller explicitly
  1408. * requests that. Wait for some write requests to complete then retry.
  1409. *
  1410. * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
  1411. * <= 3, but that may not be true in other implementations.
  1412. */
  1413. do_retry = 0;
  1414. if (!(gfp_mask & __GFP_NORETRY)) {
  1415. if ((order <= PAGE_ALLOC_COSTLY_ORDER) ||
  1416. (gfp_mask & __GFP_REPEAT))
  1417. do_retry = 1;
  1418. if (gfp_mask & __GFP_NOFAIL)
  1419. do_retry = 1;
  1420. }
  1421. if (do_retry) {
  1422. congestion_wait(WRITE, HZ/50);
  1423. goto rebalance;
  1424. }
  1425. nopage:
  1426. if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
  1427. printk(KERN_WARNING "%s: page allocation failure."
  1428. " order:%d, mode:0x%x\n",
  1429. p->comm, order, gfp_mask);
  1430. dump_stack();
  1431. show_mem();
  1432. }
  1433. got_pg:
  1434. return page;
  1435. }
  1436. EXPORT_SYMBOL(__alloc_pages);
  1437. /*
  1438. * Common helper functions.
  1439. */
  1440. fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
  1441. {
  1442. struct page * page;
  1443. page = alloc_pages(gfp_mask, order);
  1444. if (!page)
  1445. return 0;
  1446. return (unsigned long) page_address(page);
  1447. }
  1448. EXPORT_SYMBOL(__get_free_pages);
  1449. fastcall unsigned long get_zeroed_page(gfp_t gfp_mask)
  1450. {
  1451. struct page * page;
  1452. /*
  1453. * get_zeroed_page() returns a 32-bit address, which cannot represent
  1454. * a highmem page
  1455. */
  1456. VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
  1457. page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
  1458. if (page)
  1459. return (unsigned long) page_address(page);
  1460. return 0;
  1461. }
  1462. EXPORT_SYMBOL(get_zeroed_page);
  1463. void __pagevec_free(struct pagevec *pvec)
  1464. {
  1465. int i = pagevec_count(pvec);
  1466. while (--i >= 0)
  1467. free_hot_cold_page(pvec->pages[i], pvec->cold);
  1468. }
  1469. fastcall void __free_pages(struct page *page, unsigned int order)
  1470. {
  1471. if (put_page_testzero(page)) {
  1472. if (order == 0)
  1473. free_hot_page(page);
  1474. else
  1475. __free_pages_ok(page, order);
  1476. }
  1477. }
  1478. EXPORT_SYMBOL(__free_pages);
  1479. fastcall void free_pages(unsigned long addr, unsigned int order)
  1480. {
  1481. if (addr != 0) {
  1482. VM_BUG_ON(!virt_addr_valid((void *)addr));
  1483. __free_pages(virt_to_page((void *)addr), order);
  1484. }
  1485. }
  1486. EXPORT_SYMBOL(free_pages);
  1487. static unsigned int nr_free_zone_pages(int offset)
  1488. {
  1489. /* Just pick one node, since fallback list is circular */
  1490. pg_data_t *pgdat = NODE_DATA(numa_node_id());
  1491. unsigned int sum = 0;
  1492. struct zonelist *zonelist = pgdat->node_zonelists + offset;
  1493. struct zone **zonep = zonelist->zones;
  1494. struct zone *zone;
  1495. for (zone = *zonep++; zone; zone = *zonep++) {
  1496. unsigned long size = zone->present_pages;
  1497. unsigned long high = zone->pages_high;
  1498. if (size > high)
  1499. sum += size - high;
  1500. }
  1501. return sum;
  1502. }
  1503. /*
  1504. * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
  1505. */
  1506. unsigned int nr_free_buffer_pages(void)
  1507. {
  1508. return nr_free_zone_pages(gfp_zone(GFP_USER));
  1509. }
  1510. EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
  1511. /*
  1512. * Amount of free RAM allocatable within all zones
  1513. */
  1514. unsigned int nr_free_pagecache_pages(void)
  1515. {
  1516. return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
  1517. }
  1518. static inline void show_node(struct zone *zone)
  1519. {
  1520. if (NUMA_BUILD)
  1521. printk("Node %d ", zone_to_nid(zone));
  1522. }
  1523. void si_meminfo(struct sysinfo *val)
  1524. {
  1525. val->totalram = totalram_pages;
  1526. val->sharedram = 0;
  1527. val->freeram = global_page_state(NR_FREE_PAGES);
  1528. val->bufferram = nr_blockdev_pages();
  1529. val->totalhigh = totalhigh_pages;
  1530. val->freehigh = nr_free_highpages();
  1531. val->mem_unit = PAGE_SIZE;
  1532. }
  1533. EXPORT_SYMBOL(si_meminfo);
  1534. #ifdef CONFIG_NUMA
  1535. void si_meminfo_node(struct sysinfo *val, int nid)
  1536. {
  1537. pg_data_t *pgdat = NODE_DATA(nid);
  1538. val->totalram = pgdat->node_present_pages;
  1539. val->freeram = node_page_state(nid, NR_FREE_PAGES);
  1540. #ifdef CONFIG_HIGHMEM
  1541. val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
  1542. val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
  1543. NR_FREE_PAGES);
  1544. #else
  1545. val->totalhigh = 0;
  1546. val->freehigh = 0;
  1547. #endif
  1548. val->mem_unit = PAGE_SIZE;
  1549. }
  1550. #endif
  1551. #define K(x) ((x) << (PAGE_SHIFT-10))
  1552. /*
  1553. * Show free area list (used inside shift_scroll-lock stuff)
  1554. * We also calculate the percentage fragmentation. We do this by counting the
  1555. * memory on each free list with the exception of the first item on the list.
  1556. */
  1557. void show_free_areas(void)
  1558. {
  1559. int cpu;
  1560. struct zone *zone;
  1561. for_each_zone(zone) {
  1562. if (!populated_zone(zone))
  1563. continue;
  1564. show_node(zone);
  1565. printk("%s per-cpu:\n", zone->name);
  1566. for_each_online_cpu(cpu) {
  1567. struct per_cpu_pageset *pageset;
  1568. pageset = zone_pcp(zone, cpu);
  1569. printk("CPU %4d: Hot: hi:%5d, btch:%4d usd:%4d "
  1570. "Cold: hi:%5d, btch:%4d usd:%4d\n",
  1571. cpu, pageset->pcp[0].high,
  1572. pageset->pcp[0].batch, pageset->pcp[0].count,
  1573. pageset->pcp[1].high, pageset->pcp[1].batch,
  1574. pageset->pcp[1].count);
  1575. }
  1576. }
  1577. printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n"
  1578. " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
  1579. global_page_state(NR_ACTIVE),
  1580. global_page_state(NR_INACTIVE),
  1581. global_page_state(NR_FILE_DIRTY),
  1582. global_page_state(NR_WRITEBACK),
  1583. global_page_state(NR_UNSTABLE_NFS),
  1584. global_page_state(NR_FREE_PAGES),
  1585. global_page_state(NR_SLAB_RECLAIMABLE) +
  1586. global_page_state(NR_SLAB_UNRECLAIMABLE),
  1587. global_page_state(NR_FILE_MAPPED),
  1588. global_page_state(NR_PAGETABLE),
  1589. global_page_state(NR_BOUNCE));
  1590. for_each_zone(zone) {
  1591. int i;
  1592. if (!populated_zone(zone))
  1593. continue;
  1594. show_node(zone);
  1595. printk("%s"
  1596. " free:%lukB"
  1597. " min:%lukB"
  1598. " low:%lukB"
  1599. " high:%lukB"
  1600. " active:%lukB"
  1601. " inactive:%lukB"
  1602. " present:%lukB"
  1603. " pages_scanned:%lu"
  1604. " all_unreclaimable? %s"
  1605. "\n",
  1606. zone->name,
  1607. K(zone_page_state(zone, NR_FREE_PAGES)),
  1608. K(zone->pages_min),
  1609. K(zone->pages_low),
  1610. K(zone->pages_high),
  1611. K(zone_page_state(zone, NR_ACTIVE)),
  1612. K(zone_page_state(zone, NR_INACTIVE)),
  1613. K(zone->present_pages),
  1614. zone->pages_scanned,
  1615. (zone->all_unreclaimable ? "yes" : "no")
  1616. );
  1617. printk("lowmem_reserve[]:");
  1618. for (i = 0; i < MAX_NR_ZONES; i++)
  1619. printk(" %lu", zone->lowmem_reserve[i]);
  1620. printk("\n");
  1621. }
  1622. for_each_zone(zone) {
  1623. unsigned long nr[MAX_ORDER], flags, order, total = 0;
  1624. if (!populated_zone(zone))
  1625. continue;
  1626. show_node(zone);
  1627. printk("%s: ", zone->name);
  1628. spin_lock_irqsave(&zone->lock, flags);
  1629. for (order = 0; order < MAX_ORDER; order++) {
  1630. nr[order] = zone->free_area[order].nr_free;
  1631. total += nr[order] << order;
  1632. }
  1633. spin_unlock_irqrestore(&zone->lock, flags);
  1634. for (order = 0; order < MAX_ORDER; order++)
  1635. printk("%lu*%lukB ", nr[order], K(1UL) << order);
  1636. printk("= %lukB\n", K(total));
  1637. }
  1638. show_swap_cache_info();
  1639. }
  1640. /*
  1641. * Builds allocation fallback zone lists.
  1642. *
  1643. * Add all populated zones of a node to the zonelist.
  1644. */
  1645. static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
  1646. int nr_zones, enum zone_type zone_type)
  1647. {
  1648. struct zone *zone;
  1649. BUG_ON(zone_type >= MAX_NR_ZONES);
  1650. zone_type++;
  1651. do {
  1652. zone_type--;
  1653. zone = pgdat->node_zones + zone_type;
  1654. if (populated_zone(zone)) {
  1655. zonelist->zones[nr_zones++] = zone;
  1656. check_highest_zone(zone_type);
  1657. }
  1658. } while (zone_type);
  1659. return nr_zones;
  1660. }
  1661. /*
  1662. * zonelist_order:
  1663. * 0 = automatic detection of better ordering.
  1664. * 1 = order by ([node] distance, -zonetype)
  1665. * 2 = order by (-zonetype, [node] distance)
  1666. *
  1667. * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
  1668. * the same zonelist. So only NUMA can configure this param.
  1669. */
  1670. #define ZONELIST_ORDER_DEFAULT 0
  1671. #define ZONELIST_ORDER_NODE 1
  1672. #define ZONELIST_ORDER_ZONE 2
  1673. /* zonelist order in the kernel.
  1674. * set_zonelist_order() will set this to NODE or ZONE.
  1675. */
  1676. static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
  1677. static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
  1678. #ifdef CONFIG_NUMA
  1679. /* The value user specified ....changed by config */
  1680. static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
  1681. /* string for sysctl */
  1682. #define NUMA_ZONELIST_ORDER_LEN 16
  1683. char numa_zonelist_order[16] = "default";
  1684. /*
  1685. * interface for configure zonelist ordering.
  1686. * command line option "numa_zonelist_order"
  1687. * = "[dD]efault - default, automatic configuration.
  1688. * = "[nN]ode - order by node locality, then by zone within node
  1689. * = "[zZ]one - order by zone, then by locality within zone
  1690. */
  1691. static int __parse_numa_zonelist_order(char *s)
  1692. {
  1693. if (*s == 'd' || *s == 'D') {
  1694. user_zonelist_order = ZONELIST_ORDER_DEFAULT;
  1695. } else if (*s == 'n' || *s == 'N') {
  1696. user_zonelist_order = ZONELIST_ORDER_NODE;
  1697. } else if (*s == 'z' || *s == 'Z') {
  1698. user_zonelist_order = ZONELIST_ORDER_ZONE;
  1699. } else {
  1700. printk(KERN_WARNING
  1701. "Ignoring invalid numa_zonelist_order value: "
  1702. "%s\n", s);
  1703. return -EINVAL;
  1704. }
  1705. return 0;
  1706. }
  1707. static __init int setup_numa_zonelist_order(char *s)
  1708. {
  1709. if (s)
  1710. return __parse_numa_zonelist_order(s);
  1711. return 0;
  1712. }
  1713. early_param("numa_zonelist_order", setup_numa_zonelist_order);
  1714. /*
  1715. * sysctl handler for numa_zonelist_order
  1716. */
  1717. int numa_zonelist_order_handler(ctl_table *table, int write,
  1718. struct file *file, void __user *buffer, size_t *length,
  1719. loff_t *ppos)
  1720. {
  1721. char saved_string[NUMA_ZONELIST_ORDER_LEN];
  1722. int ret;
  1723. if (write)
  1724. strncpy(saved_string, (char*)table->data,
  1725. NUMA_ZONELIST_ORDER_LEN);
  1726. ret = proc_dostring(table, write, file, buffer, length, ppos);
  1727. if (ret)
  1728. return ret;
  1729. if (write) {
  1730. int oldval = user_zonelist_order;
  1731. if (__parse_numa_zonelist_order((char*)table->data)) {
  1732. /*
  1733. * bogus value. restore saved string
  1734. */
  1735. strncpy((char*)table->data, saved_string,
  1736. NUMA_ZONELIST_ORDER_LEN);
  1737. user_zonelist_order = oldval;
  1738. } else if (oldval != user_zonelist_order)
  1739. build_all_zonelists();
  1740. }
  1741. return 0;
  1742. }
  1743. #define MAX_NODE_LOAD (num_online_nodes())
  1744. static int node_load[MAX_NUMNODES];
  1745. /**
  1746. * find_next_best_node - find the next node that should appear in a given node's fallback list
  1747. * @node: node whose fallback list we're appending
  1748. * @used_node_mask: nodemask_t of already used nodes
  1749. *
  1750. * We use a number of factors to determine which is the next node that should
  1751. * appear on a given node's fallback list. The node should not have appeared
  1752. * already in @node's fallback list, and it should be the next closest node
  1753. * according to the distance array (which contains arbitrary distance values
  1754. * from each node to each node in the system), and should also prefer nodes
  1755. * with no CPUs, since presumably they'll have very little allocation pressure
  1756. * on them otherwise.
  1757. * It returns -1 if no node is found.
  1758. */
  1759. static int find_next_best_node(int node, nodemask_t *used_node_mask)
  1760. {
  1761. int n, val;
  1762. int min_val = INT_MAX;
  1763. int best_node = -1;
  1764. /* Use the local node if we haven't already */
  1765. if (!node_isset(node, *used_node_mask)) {
  1766. node_set(node, *used_node_mask);
  1767. return node;
  1768. }
  1769. for_each_node_state(n, N_HIGH_MEMORY) {
  1770. cpumask_t tmp;
  1771. /* Don't want a node to appear more than once */
  1772. if (node_isset(n, *used_node_mask))
  1773. continue;
  1774. /* Use the distance array to find the distance */
  1775. val = node_distance(node, n);
  1776. /* Penalize nodes under us ("prefer the next node") */
  1777. val += (n < node);
  1778. /* Give preference to headless and unused nodes */
  1779. tmp = node_to_cpumask(n);
  1780. if (!cpus_empty(tmp))
  1781. val += PENALTY_FOR_NODE_WITH_CPUS;
  1782. /* Slight preference for less loaded node */
  1783. val *= (MAX_NODE_LOAD*MAX_NUMNODES);
  1784. val += node_load[n];
  1785. if (val < min_val) {
  1786. min_val = val;
  1787. best_node = n;
  1788. }
  1789. }
  1790. if (best_node >= 0)
  1791. node_set(best_node, *used_node_mask);
  1792. return best_node;
  1793. }
  1794. /*
  1795. * Build zonelists ordered by node and zones within node.
  1796. * This results in maximum locality--normal zone overflows into local
  1797. * DMA zone, if any--but risks exhausting DMA zone.
  1798. */
  1799. static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
  1800. {
  1801. enum zone_type i;
  1802. int j;
  1803. struct zonelist *zonelist;
  1804. for (i = 0; i < MAX_NR_ZONES; i++) {
  1805. zonelist = pgdat->node_zonelists + i;
  1806. for (j = 0; zonelist->zones[j] != NULL; j++)
  1807. ;
  1808. j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
  1809. zonelist->zones[j] = NULL;
  1810. }
  1811. }
  1812. /*
  1813. * Build gfp_thisnode zonelists
  1814. */
  1815. static void build_thisnode_zonelists(pg_data_t *pgdat)
  1816. {
  1817. enum zone_type i;
  1818. int j;
  1819. struct zonelist *zonelist;
  1820. for (i = 0; i < MAX_NR_ZONES; i++) {
  1821. zonelist = pgdat->node_zonelists + MAX_NR_ZONES + i;
  1822. j = build_zonelists_node(pgdat, zonelist, 0, i);
  1823. zonelist->zones[j] = NULL;
  1824. }
  1825. }
  1826. /*
  1827. * Build zonelists ordered by zone and nodes within zones.
  1828. * This results in conserving DMA zone[s] until all Normal memory is
  1829. * exhausted, but results in overflowing to remote node while memory
  1830. * may still exist in local DMA zone.
  1831. */
  1832. static int node_order[MAX_NUMNODES];
  1833. static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
  1834. {
  1835. enum zone_type i;
  1836. int pos, j, node;
  1837. int zone_type; /* needs to be signed */
  1838. struct zone *z;
  1839. struct zonelist *zonelist;
  1840. for (i = 0; i < MAX_NR_ZONES; i++) {
  1841. zonelist = pgdat->node_zonelists + i;
  1842. pos = 0;
  1843. for (zone_type = i; zone_type >= 0; zone_type--) {
  1844. for (j = 0; j < nr_nodes; j++) {
  1845. node = node_order[j];
  1846. z = &NODE_DATA(node)->node_zones[zone_type];
  1847. if (populated_zone(z)) {
  1848. zonelist->zones[pos++] = z;
  1849. check_highest_zone(zone_type);
  1850. }
  1851. }
  1852. }
  1853. zonelist->zones[pos] = NULL;
  1854. }
  1855. }
  1856. static int default_zonelist_order(void)
  1857. {
  1858. int nid, zone_type;
  1859. unsigned long low_kmem_size,total_size;
  1860. struct zone *z;
  1861. int average_size;
  1862. /*
  1863. * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
  1864. * If they are really small and used heavily, the system can fall
  1865. * into OOM very easily.
  1866. * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
  1867. */
  1868. /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
  1869. low_kmem_size = 0;
  1870. total_size = 0;
  1871. for_each_online_node(nid) {
  1872. for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
  1873. z = &NODE_DATA(nid)->node_zones[zone_type];
  1874. if (populated_zone(z)) {
  1875. if (zone_type < ZONE_NORMAL)
  1876. low_kmem_size += z->present_pages;
  1877. total_size += z->present_pages;
  1878. }
  1879. }
  1880. }
  1881. if (!low_kmem_size || /* there are no DMA area. */
  1882. low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
  1883. return ZONELIST_ORDER_NODE;
  1884. /*
  1885. * look into each node's config.
  1886. * If there is a node whose DMA/DMA32 memory is very big area on
  1887. * local memory, NODE_ORDER may be suitable.
  1888. */
  1889. average_size = total_size /
  1890. (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
  1891. for_each_online_node(nid) {
  1892. low_kmem_size = 0;
  1893. total_size = 0;
  1894. for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
  1895. z = &NODE_DATA(nid)->node_zones[zone_type];
  1896. if (populated_zone(z)) {
  1897. if (zone_type < ZONE_NORMAL)
  1898. low_kmem_size += z->present_pages;
  1899. total_size += z->present_pages;
  1900. }
  1901. }
  1902. if (low_kmem_size &&
  1903. total_size > average_size && /* ignore small node */
  1904. low_kmem_size > total_size * 70/100)
  1905. return ZONELIST_ORDER_NODE;
  1906. }
  1907. return ZONELIST_ORDER_ZONE;
  1908. }
  1909. static void set_zonelist_order(void)
  1910. {
  1911. if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
  1912. current_zonelist_order = default_zonelist_order();
  1913. else
  1914. current_zonelist_order = user_zonelist_order;
  1915. }
  1916. static void build_zonelists(pg_data_t *pgdat)
  1917. {
  1918. int j, node, load;
  1919. enum zone_type i;
  1920. nodemask_t used_mask;
  1921. int local_node, prev_node;
  1922. struct zonelist *zonelist;
  1923. int order = current_zonelist_order;
  1924. /* initialize zonelists */
  1925. for (i = 0; i < MAX_ZONELISTS; i++) {
  1926. zonelist = pgdat->node_zonelists + i;
  1927. zonelist->zones[0] = NULL;
  1928. }
  1929. /* NUMA-aware ordering of nodes */
  1930. local_node = pgdat->node_id;
  1931. load = num_online_nodes();
  1932. prev_node = local_node;
  1933. nodes_clear(used_mask);
  1934. memset(node_load, 0, sizeof(node_load));
  1935. memset(node_order, 0, sizeof(node_order));
  1936. j = 0;
  1937. while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
  1938. int distance = node_distance(local_node, node);
  1939. /*
  1940. * If another node is sufficiently far away then it is better
  1941. * to reclaim pages in a zone before going off node.
  1942. */
  1943. if (distance > RECLAIM_DISTANCE)
  1944. zone_reclaim_mode = 1;
  1945. /*
  1946. * We don't want to pressure a particular node.
  1947. * So adding penalty to the first node in same
  1948. * distance group to make it round-robin.
  1949. */
  1950. if (distance != node_distance(local_node, prev_node))
  1951. node_load[node] = load;
  1952. prev_node = node;
  1953. load--;
  1954. if (order == ZONELIST_ORDER_NODE)
  1955. build_zonelists_in_node_order(pgdat, node);
  1956. else
  1957. node_order[j++] = node; /* remember order */
  1958. }
  1959. if (order == ZONELIST_ORDER_ZONE) {
  1960. /* calculate node order -- i.e., DMA last! */
  1961. build_zonelists_in_zone_order(pgdat, j);
  1962. }
  1963. build_thisnode_zonelists(pgdat);
  1964. }
  1965. /* Construct the zonelist performance cache - see further mmzone.h */
  1966. static void build_zonelist_cache(pg_data_t *pgdat)
  1967. {
  1968. int i;
  1969. for (i = 0; i < MAX_NR_ZONES; i++) {
  1970. struct zonelist *zonelist;
  1971. struct zonelist_cache *zlc;
  1972. struct zone **z;
  1973. zonelist = pgdat->node_zonelists + i;
  1974. zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
  1975. bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
  1976. for (z = zonelist->zones; *z; z++)
  1977. zlc->z_to_n[z - zonelist->zones] = zone_to_nid(*z);
  1978. }
  1979. }
  1980. #else /* CONFIG_NUMA */
  1981. static void set_zonelist_order(void)
  1982. {
  1983. current_zonelist_order = ZONELIST_ORDER_ZONE;
  1984. }
  1985. static void build_zonelists(pg_data_t *pgdat)
  1986. {
  1987. int node, local_node;
  1988. enum zone_type i,j;
  1989. local_node = pgdat->node_id;
  1990. for (i = 0; i < MAX_NR_ZONES; i++) {
  1991. struct zonelist *zonelist;
  1992. zonelist = pgdat->node_zonelists + i;
  1993. j = build_zonelists_node(pgdat, zonelist, 0, i);
  1994. /*
  1995. * Now we build the zonelist so that it contains the zones
  1996. * of all the other nodes.
  1997. * We don't want to pressure a particular node, so when
  1998. * building the zones for node N, we make sure that the
  1999. * zones coming right after the local ones are those from
  2000. * node N+1 (modulo N)
  2001. */
  2002. for (node = local_node + 1; node < MAX_NUMNODES; node++) {
  2003. if (!node_online(node))
  2004. continue;
  2005. j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
  2006. }
  2007. for (node = 0; node < local_node; node++) {
  2008. if (!node_online(node))
  2009. continue;
  2010. j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
  2011. }
  2012. zonelist->zones[j] = NULL;
  2013. }
  2014. }
  2015. /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
  2016. static void build_zonelist_cache(pg_data_t *pgdat)
  2017. {
  2018. int i;
  2019. for (i = 0; i < MAX_NR_ZONES; i++)
  2020. pgdat->node_zonelists[i].zlcache_ptr = NULL;
  2021. }
  2022. #endif /* CONFIG_NUMA */
  2023. /* return values int ....just for stop_machine_run() */
  2024. static int __build_all_zonelists(void *dummy)
  2025. {
  2026. int nid;
  2027. for_each_online_node(nid) {
  2028. pg_data_t *pgdat = NODE_DATA(nid);
  2029. build_zonelists(pgdat);
  2030. build_zonelist_cache(pgdat);
  2031. }
  2032. return 0;
  2033. }
  2034. void build_all_zonelists(void)
  2035. {
  2036. set_zonelist_order();
  2037. if (system_state == SYSTEM_BOOTING) {
  2038. __build_all_zonelists(NULL);
  2039. cpuset_init_current_mems_allowed();
  2040. } else {
  2041. /* we have to stop all cpus to guaranntee there is no user
  2042. of zonelist */
  2043. stop_machine_run(__build_all_zonelists, NULL, NR_CPUS);
  2044. /* cpuset refresh routine should be here */
  2045. }
  2046. vm_total_pages = nr_free_pagecache_pages();
  2047. /*
  2048. * Disable grouping by mobility if the number of pages in the
  2049. * system is too low to allow the mechanism to work. It would be
  2050. * more accurate, but expensive to check per-zone. This check is
  2051. * made on memory-hotadd so a system can start with mobility
  2052. * disabled and enable it later
  2053. */
  2054. if (vm_total_pages < (MAX_ORDER_NR_PAGES * MIGRATE_TYPES))
  2055. page_group_by_mobility_disabled = 1;
  2056. else
  2057. page_group_by_mobility_disabled = 0;
  2058. printk("Built %i zonelists in %s order, mobility grouping %s. "
  2059. "Total pages: %ld\n",
  2060. num_online_nodes(),
  2061. zonelist_order_name[current_zonelist_order],
  2062. page_group_by_mobility_disabled ? "off" : "on",
  2063. vm_total_pages);
  2064. #ifdef CONFIG_NUMA
  2065. printk("Policy zone: %s\n", zone_names[policy_zone]);
  2066. #endif
  2067. }
  2068. /*
  2069. * Helper functions to size the waitqueue hash table.
  2070. * Essentially these want to choose hash table sizes sufficiently
  2071. * large so that collisions trying to wait on pages are rare.
  2072. * But in fact, the number of active page waitqueues on typical
  2073. * systems is ridiculously low, less than 200. So this is even
  2074. * conservative, even though it seems large.
  2075. *
  2076. * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
  2077. * waitqueues, i.e. the size of the waitq table given the number of pages.
  2078. */
  2079. #define PAGES_PER_WAITQUEUE 256
  2080. #ifndef CONFIG_MEMORY_HOTPLUG
  2081. static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
  2082. {
  2083. unsigned long size = 1;
  2084. pages /= PAGES_PER_WAITQUEUE;
  2085. while (size < pages)
  2086. size <<= 1;
  2087. /*
  2088. * Once we have dozens or even hundreds of threads sleeping
  2089. * on IO we've got bigger problems than wait queue collision.
  2090. * Limit the size of the wait table to a reasonable size.
  2091. */
  2092. size = min(size, 4096UL);
  2093. return max(size, 4UL);
  2094. }
  2095. #else
  2096. /*
  2097. * A zone's size might be changed by hot-add, so it is not possible to determine
  2098. * a suitable size for its wait_table. So we use the maximum size now.
  2099. *
  2100. * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
  2101. *
  2102. * i386 (preemption config) : 4096 x 16 = 64Kbyte.
  2103. * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
  2104. * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
  2105. *
  2106. * The maximum entries are prepared when a zone's memory is (512K + 256) pages
  2107. * or more by the traditional way. (See above). It equals:
  2108. *
  2109. * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
  2110. * ia64(16K page size) : = ( 8G + 4M)byte.
  2111. * powerpc (64K page size) : = (32G +16M)byte.
  2112. */
  2113. static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
  2114. {
  2115. return 4096UL;
  2116. }
  2117. #endif
  2118. /*
  2119. * This is an integer logarithm so that shifts can be used later
  2120. * to extract the more random high bits from the multiplicative
  2121. * hash function before the remainder is taken.
  2122. */
  2123. static inline unsigned long wait_table_bits(unsigned long size)
  2124. {
  2125. return ffz(~size);
  2126. }
  2127. #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
  2128. /*
  2129. * Mark a number of MAX_ORDER_NR_PAGES blocks as MIGRATE_RESERVE. The number
  2130. * of blocks reserved is based on zone->pages_min. The memory within the
  2131. * reserve will tend to store contiguous free pages. Setting min_free_kbytes
  2132. * higher will lead to a bigger reserve which will get freed as contiguous
  2133. * blocks as reclaim kicks in
  2134. */
  2135. static void setup_zone_migrate_reserve(struct zone *zone)
  2136. {
  2137. unsigned long start_pfn, pfn, end_pfn;
  2138. struct page *page;
  2139. unsigned long reserve, block_migratetype;
  2140. /* Get the start pfn, end pfn and the number of blocks to reserve */
  2141. start_pfn = zone->zone_start_pfn;
  2142. end_pfn = start_pfn + zone->spanned_pages;
  2143. reserve = roundup(zone->pages_min, MAX_ORDER_NR_PAGES) >> (MAX_ORDER-1);
  2144. for (pfn = start_pfn; pfn < end_pfn; pfn += MAX_ORDER_NR_PAGES) {
  2145. if (!pfn_valid(pfn))
  2146. continue;
  2147. page = pfn_to_page(pfn);
  2148. /* Blocks with reserved pages will never free, skip them. */
  2149. if (PageReserved(page))
  2150. continue;
  2151. block_migratetype = get_pageblock_migratetype(page);
  2152. /* If this block is reserved, account for it */
  2153. if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
  2154. reserve--;
  2155. continue;
  2156. }
  2157. /* Suitable for reserving if this block is movable */
  2158. if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
  2159. set_pageblock_migratetype(page, MIGRATE_RESERVE);
  2160. move_freepages_block(zone, page, MIGRATE_RESERVE);
  2161. reserve--;
  2162. continue;
  2163. }
  2164. /*
  2165. * If the reserve is met and this is a previous reserved block,
  2166. * take it back
  2167. */
  2168. if (block_migratetype == MIGRATE_RESERVE) {
  2169. set_pageblock_migratetype(page, MIGRATE_MOVABLE);
  2170. move_freepages_block(zone, page, MIGRATE_MOVABLE);
  2171. }
  2172. }
  2173. }
  2174. /*
  2175. * Initially all pages are reserved - free ones are freed
  2176. * up by free_all_bootmem() once the early boot process is
  2177. * done. Non-atomic initialization, single-pass.
  2178. */
  2179. void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
  2180. unsigned long start_pfn, enum memmap_context context)
  2181. {
  2182. struct page *page;
  2183. unsigned long end_pfn = start_pfn + size;
  2184. unsigned long pfn;
  2185. for (pfn = start_pfn; pfn < end_pfn; pfn++) {
  2186. /*
  2187. * There can be holes in boot-time mem_map[]s
  2188. * handed to this function. They do not
  2189. * exist on hotplugged memory.
  2190. */
  2191. if (context == MEMMAP_EARLY) {
  2192. if (!early_pfn_valid(pfn))
  2193. continue;
  2194. if (!early_pfn_in_nid(pfn, nid))
  2195. continue;
  2196. }
  2197. page = pfn_to_page(pfn);
  2198. set_page_links(page, zone, nid, pfn);
  2199. init_page_count(page);
  2200. reset_page_mapcount(page);
  2201. SetPageReserved(page);
  2202. /*
  2203. * Mark the block movable so that blocks are reserved for
  2204. * movable at startup. This will force kernel allocations
  2205. * to reserve their blocks rather than leaking throughout
  2206. * the address space during boot when many long-lived
  2207. * kernel allocations are made. Later some blocks near
  2208. * the start are marked MIGRATE_RESERVE by
  2209. * setup_zone_migrate_reserve()
  2210. */
  2211. if ((pfn & (MAX_ORDER_NR_PAGES-1)))
  2212. set_pageblock_migratetype(page, MIGRATE_MOVABLE);
  2213. INIT_LIST_HEAD(&page->lru);
  2214. #ifdef WANT_PAGE_VIRTUAL
  2215. /* The shift won't overflow because ZONE_NORMAL is below 4G. */
  2216. if (!is_highmem_idx(zone))
  2217. set_page_address(page, __va(pfn << PAGE_SHIFT));
  2218. #endif
  2219. }
  2220. }
  2221. static void __meminit zone_init_free_lists(struct pglist_data *pgdat,
  2222. struct zone *zone, unsigned long size)
  2223. {
  2224. int order, t;
  2225. for_each_migratetype_order(order, t) {
  2226. INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
  2227. zone->free_area[order].nr_free = 0;
  2228. }
  2229. }
  2230. #ifndef __HAVE_ARCH_MEMMAP_INIT
  2231. #define memmap_init(size, nid, zone, start_pfn) \
  2232. memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
  2233. #endif
  2234. static int __devinit zone_batchsize(struct zone *zone)
  2235. {
  2236. int batch;
  2237. /*
  2238. * The per-cpu-pages pools are set to around 1000th of the
  2239. * size of the zone. But no more than 1/2 of a meg.
  2240. *
  2241. * OK, so we don't know how big the cache is. So guess.
  2242. */
  2243. batch = zone->present_pages / 1024;
  2244. if (batch * PAGE_SIZE > 512 * 1024)
  2245. batch = (512 * 1024) / PAGE_SIZE;
  2246. batch /= 4; /* We effectively *= 4 below */
  2247. if (batch < 1)
  2248. batch = 1;
  2249. /*
  2250. * Clamp the batch to a 2^n - 1 value. Having a power
  2251. * of 2 value was found to be more likely to have
  2252. * suboptimal cache aliasing properties in some cases.
  2253. *
  2254. * For example if 2 tasks are alternately allocating
  2255. * batches of pages, one task can end up with a lot
  2256. * of pages of one half of the possible page colors
  2257. * and the other with pages of the other colors.
  2258. */
  2259. batch = (1 << (fls(batch + batch/2)-1)) - 1;
  2260. return batch;
  2261. }
  2262. inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
  2263. {
  2264. struct per_cpu_pages *pcp;
  2265. memset(p, 0, sizeof(*p));
  2266. pcp = &p->pcp[0]; /* hot */
  2267. pcp->count = 0;
  2268. pcp->high = 6 * batch;
  2269. pcp->batch = max(1UL, 1 * batch);
  2270. INIT_LIST_HEAD(&pcp->list);
  2271. pcp = &p->pcp[1]; /* cold*/
  2272. pcp->count = 0;
  2273. pcp->high = 2 * batch;
  2274. pcp->batch = max(1UL, batch/2);
  2275. INIT_LIST_HEAD(&pcp->list);
  2276. }
  2277. /*
  2278. * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
  2279. * to the value high for the pageset p.
  2280. */
  2281. static void setup_pagelist_highmark(struct per_cpu_pageset *p,
  2282. unsigned long high)
  2283. {
  2284. struct per_cpu_pages *pcp;
  2285. pcp = &p->pcp[0]; /* hot list */
  2286. pcp->high = high;
  2287. pcp->batch = max(1UL, high/4);
  2288. if ((high/4) > (PAGE_SHIFT * 8))
  2289. pcp->batch = PAGE_SHIFT * 8;
  2290. }
  2291. #ifdef CONFIG_NUMA
  2292. /*
  2293. * Boot pageset table. One per cpu which is going to be used for all
  2294. * zones and all nodes. The parameters will be set in such a way
  2295. * that an item put on a list will immediately be handed over to
  2296. * the buddy list. This is safe since pageset manipulation is done
  2297. * with interrupts disabled.
  2298. *
  2299. * Some NUMA counter updates may also be caught by the boot pagesets.
  2300. *
  2301. * The boot_pagesets must be kept even after bootup is complete for
  2302. * unused processors and/or zones. They do play a role for bootstrapping
  2303. * hotplugged processors.
  2304. *
  2305. * zoneinfo_show() and maybe other functions do
  2306. * not check if the processor is online before following the pageset pointer.
  2307. * Other parts of the kernel may not check if the zone is available.
  2308. */
  2309. static struct per_cpu_pageset boot_pageset[NR_CPUS];
  2310. /*
  2311. * Dynamically allocate memory for the
  2312. * per cpu pageset array in struct zone.
  2313. */
  2314. static int __cpuinit process_zones(int cpu)
  2315. {
  2316. struct zone *zone, *dzone;
  2317. int node = cpu_to_node(cpu);
  2318. node_set_state(node, N_CPU); /* this node has a cpu */
  2319. for_each_zone(zone) {
  2320. if (!populated_zone(zone))
  2321. continue;
  2322. zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
  2323. GFP_KERNEL, node);
  2324. if (!zone_pcp(zone, cpu))
  2325. goto bad;
  2326. setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
  2327. if (percpu_pagelist_fraction)
  2328. setup_pagelist_highmark(zone_pcp(zone, cpu),
  2329. (zone->present_pages / percpu_pagelist_fraction));
  2330. }
  2331. return 0;
  2332. bad:
  2333. for_each_zone(dzone) {
  2334. if (!populated_zone(dzone))
  2335. continue;
  2336. if (dzone == zone)
  2337. break;
  2338. kfree(zone_pcp(dzone, cpu));
  2339. zone_pcp(dzone, cpu) = NULL;
  2340. }
  2341. return -ENOMEM;
  2342. }
  2343. static inline void free_zone_pagesets(int cpu)
  2344. {
  2345. struct zone *zone;
  2346. for_each_zone(zone) {
  2347. struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
  2348. /* Free per_cpu_pageset if it is slab allocated */
  2349. if (pset != &boot_pageset[cpu])
  2350. kfree(pset);
  2351. zone_pcp(zone, cpu) = NULL;
  2352. }
  2353. }
  2354. static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
  2355. unsigned long action,
  2356. void *hcpu)
  2357. {
  2358. int cpu = (long)hcpu;
  2359. int ret = NOTIFY_OK;
  2360. switch (action) {
  2361. case CPU_UP_PREPARE:
  2362. case CPU_UP_PREPARE_FROZEN:
  2363. if (process_zones(cpu))
  2364. ret = NOTIFY_BAD;
  2365. break;
  2366. case CPU_UP_CANCELED:
  2367. case CPU_UP_CANCELED_FROZEN:
  2368. case CPU_DEAD:
  2369. case CPU_DEAD_FROZEN:
  2370. free_zone_pagesets(cpu);
  2371. break;
  2372. default:
  2373. break;
  2374. }
  2375. return ret;
  2376. }
  2377. static struct notifier_block __cpuinitdata pageset_notifier =
  2378. { &pageset_cpuup_callback, NULL, 0 };
  2379. void __init setup_per_cpu_pageset(void)
  2380. {
  2381. int err;
  2382. /* Initialize per_cpu_pageset for cpu 0.
  2383. * A cpuup callback will do this for every cpu
  2384. * as it comes online
  2385. */
  2386. err = process_zones(smp_processor_id());
  2387. BUG_ON(err);
  2388. register_cpu_notifier(&pageset_notifier);
  2389. }
  2390. #endif
  2391. static noinline __init_refok
  2392. int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
  2393. {
  2394. int i;
  2395. struct pglist_data *pgdat = zone->zone_pgdat;
  2396. size_t alloc_size;
  2397. /*
  2398. * The per-page waitqueue mechanism uses hashed waitqueues
  2399. * per zone.
  2400. */
  2401. zone->wait_table_hash_nr_entries =
  2402. wait_table_hash_nr_entries(zone_size_pages);
  2403. zone->wait_table_bits =
  2404. wait_table_bits(zone->wait_table_hash_nr_entries);
  2405. alloc_size = zone->wait_table_hash_nr_entries
  2406. * sizeof(wait_queue_head_t);
  2407. if (system_state == SYSTEM_BOOTING) {
  2408. zone->wait_table = (wait_queue_head_t *)
  2409. alloc_bootmem_node(pgdat, alloc_size);
  2410. } else {
  2411. /*
  2412. * This case means that a zone whose size was 0 gets new memory
  2413. * via memory hot-add.
  2414. * But it may be the case that a new node was hot-added. In
  2415. * this case vmalloc() will not be able to use this new node's
  2416. * memory - this wait_table must be initialized to use this new
  2417. * node itself as well.
  2418. * To use this new node's memory, further consideration will be
  2419. * necessary.
  2420. */
  2421. zone->wait_table = vmalloc(alloc_size);
  2422. }
  2423. if (!zone->wait_table)
  2424. return -ENOMEM;
  2425. for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
  2426. init_waitqueue_head(zone->wait_table + i);
  2427. return 0;
  2428. }
  2429. static __meminit void zone_pcp_init(struct zone *zone)
  2430. {
  2431. int cpu;
  2432. unsigned long batch = zone_batchsize(zone);
  2433. for (cpu = 0; cpu < NR_CPUS; cpu++) {
  2434. #ifdef CONFIG_NUMA
  2435. /* Early boot. Slab allocator not functional yet */
  2436. zone_pcp(zone, cpu) = &boot_pageset[cpu];
  2437. setup_pageset(&boot_pageset[cpu],0);
  2438. #else
  2439. setup_pageset(zone_pcp(zone,cpu), batch);
  2440. #endif
  2441. }
  2442. if (zone->present_pages)
  2443. printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
  2444. zone->name, zone->present_pages, batch);
  2445. }
  2446. __meminit int init_currently_empty_zone(struct zone *zone,
  2447. unsigned long zone_start_pfn,
  2448. unsigned long size,
  2449. enum memmap_context context)
  2450. {
  2451. struct pglist_data *pgdat = zone->zone_pgdat;
  2452. int ret;
  2453. ret = zone_wait_table_init(zone, size);
  2454. if (ret)
  2455. return ret;
  2456. pgdat->nr_zones = zone_idx(zone) + 1;
  2457. zone->zone_start_pfn = zone_start_pfn;
  2458. memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
  2459. zone_init_free_lists(pgdat, zone, zone->spanned_pages);
  2460. return 0;
  2461. }
  2462. #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
  2463. /*
  2464. * Basic iterator support. Return the first range of PFNs for a node
  2465. * Note: nid == MAX_NUMNODES returns first region regardless of node
  2466. */
  2467. static int __meminit first_active_region_index_in_nid(int nid)
  2468. {
  2469. int i;
  2470. for (i = 0; i < nr_nodemap_entries; i++)
  2471. if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
  2472. return i;
  2473. return -1;
  2474. }
  2475. /*
  2476. * Basic iterator support. Return the next active range of PFNs for a node
  2477. * Note: nid == MAX_NUMNODES returns next region regardles of node
  2478. */
  2479. static int __meminit next_active_region_index_in_nid(int index, int nid)
  2480. {
  2481. for (index = index + 1; index < nr_nodemap_entries; index++)
  2482. if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
  2483. return index;
  2484. return -1;
  2485. }
  2486. #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
  2487. /*
  2488. * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
  2489. * Architectures may implement their own version but if add_active_range()
  2490. * was used and there are no special requirements, this is a convenient
  2491. * alternative
  2492. */
  2493. int __meminit early_pfn_to_nid(unsigned long pfn)
  2494. {
  2495. int i;
  2496. for (i = 0; i < nr_nodemap_entries; i++) {
  2497. unsigned long start_pfn = early_node_map[i].start_pfn;
  2498. unsigned long end_pfn = early_node_map[i].end_pfn;
  2499. if (start_pfn <= pfn && pfn < end_pfn)
  2500. return early_node_map[i].nid;
  2501. }
  2502. return 0;
  2503. }
  2504. #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
  2505. /* Basic iterator support to walk early_node_map[] */
  2506. #define for_each_active_range_index_in_nid(i, nid) \
  2507. for (i = first_active_region_index_in_nid(nid); i != -1; \
  2508. i = next_active_region_index_in_nid(i, nid))
  2509. /**
  2510. * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
  2511. * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
  2512. * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
  2513. *
  2514. * If an architecture guarantees that all ranges registered with
  2515. * add_active_ranges() contain no holes and may be freed, this
  2516. * this function may be used instead of calling free_bootmem() manually.
  2517. */
  2518. void __init free_bootmem_with_active_regions(int nid,
  2519. unsigned long max_low_pfn)
  2520. {
  2521. int i;
  2522. for_each_active_range_index_in_nid(i, nid) {
  2523. unsigned long size_pages = 0;
  2524. unsigned long end_pfn = early_node_map[i].end_pfn;
  2525. if (early_node_map[i].start_pfn >= max_low_pfn)
  2526. continue;
  2527. if (end_pfn > max_low_pfn)
  2528. end_pfn = max_low_pfn;
  2529. size_pages = end_pfn - early_node_map[i].start_pfn;
  2530. free_bootmem_node(NODE_DATA(early_node_map[i].nid),
  2531. PFN_PHYS(early_node_map[i].start_pfn),
  2532. size_pages << PAGE_SHIFT);
  2533. }
  2534. }
  2535. /**
  2536. * sparse_memory_present_with_active_regions - Call memory_present for each active range
  2537. * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
  2538. *
  2539. * If an architecture guarantees that all ranges registered with
  2540. * add_active_ranges() contain no holes and may be freed, this
  2541. * function may be used instead of calling memory_present() manually.
  2542. */
  2543. void __init sparse_memory_present_with_active_regions(int nid)
  2544. {
  2545. int i;
  2546. for_each_active_range_index_in_nid(i, nid)
  2547. memory_present(early_node_map[i].nid,
  2548. early_node_map[i].start_pfn,
  2549. early_node_map[i].end_pfn);
  2550. }
  2551. /**
  2552. * push_node_boundaries - Push node boundaries to at least the requested boundary
  2553. * @nid: The nid of the node to push the boundary for
  2554. * @start_pfn: The start pfn of the node
  2555. * @end_pfn: The end pfn of the node
  2556. *
  2557. * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
  2558. * time. Specifically, on x86_64, SRAT will report ranges that can potentially
  2559. * be hotplugged even though no physical memory exists. This function allows
  2560. * an arch to push out the node boundaries so mem_map is allocated that can
  2561. * be used later.
  2562. */
  2563. #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
  2564. void __init push_node_boundaries(unsigned int nid,
  2565. unsigned long start_pfn, unsigned long end_pfn)
  2566. {
  2567. printk(KERN_DEBUG "Entering push_node_boundaries(%u, %lu, %lu)\n",
  2568. nid, start_pfn, end_pfn);
  2569. /* Initialise the boundary for this node if necessary */
  2570. if (node_boundary_end_pfn[nid] == 0)
  2571. node_boundary_start_pfn[nid] = -1UL;
  2572. /* Update the boundaries */
  2573. if (node_boundary_start_pfn[nid] > start_pfn)
  2574. node_boundary_start_pfn[nid] = start_pfn;
  2575. if (node_boundary_end_pfn[nid] < end_pfn)
  2576. node_boundary_end_pfn[nid] = end_pfn;
  2577. }
  2578. /* If necessary, push the node boundary out for reserve hotadd */
  2579. static void __meminit account_node_boundary(unsigned int nid,
  2580. unsigned long *start_pfn, unsigned long *end_pfn)
  2581. {
  2582. printk(KERN_DEBUG "Entering account_node_boundary(%u, %lu, %lu)\n",
  2583. nid, *start_pfn, *end_pfn);
  2584. /* Return if boundary information has not been provided */
  2585. if (node_boundary_end_pfn[nid] == 0)
  2586. return;
  2587. /* Check the boundaries and update if necessary */
  2588. if (node_boundary_start_pfn[nid] < *start_pfn)
  2589. *start_pfn = node_boundary_start_pfn[nid];
  2590. if (node_boundary_end_pfn[nid] > *end_pfn)
  2591. *end_pfn = node_boundary_end_pfn[nid];
  2592. }
  2593. #else
  2594. void __init push_node_boundaries(unsigned int nid,
  2595. unsigned long start_pfn, unsigned long end_pfn) {}
  2596. static void __meminit account_node_boundary(unsigned int nid,
  2597. unsigned long *start_pfn, unsigned long *end_pfn) {}
  2598. #endif
  2599. /**
  2600. * get_pfn_range_for_nid - Return the start and end page frames for a node
  2601. * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
  2602. * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
  2603. * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
  2604. *
  2605. * It returns the start and end page frame of a node based on information
  2606. * provided by an arch calling add_active_range(). If called for a node
  2607. * with no available memory, a warning is printed and the start and end
  2608. * PFNs will be 0.
  2609. */
  2610. void __meminit get_pfn_range_for_nid(unsigned int nid,
  2611. unsigned long *start_pfn, unsigned long *end_pfn)
  2612. {
  2613. int i;
  2614. *start_pfn = -1UL;
  2615. *end_pfn = 0;
  2616. for_each_active_range_index_in_nid(i, nid) {
  2617. *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
  2618. *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
  2619. }
  2620. if (*start_pfn == -1UL)
  2621. *start_pfn = 0;
  2622. /* Push the node boundaries out if requested */
  2623. account_node_boundary(nid, start_pfn, end_pfn);
  2624. }
  2625. /*
  2626. * This finds a zone that can be used for ZONE_MOVABLE pages. The
  2627. * assumption is made that zones within a node are ordered in monotonic
  2628. * increasing memory addresses so that the "highest" populated zone is used
  2629. */
  2630. void __init find_usable_zone_for_movable(void)
  2631. {
  2632. int zone_index;
  2633. for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
  2634. if (zone_index == ZONE_MOVABLE)
  2635. continue;
  2636. if (arch_zone_highest_possible_pfn[zone_index] >
  2637. arch_zone_lowest_possible_pfn[zone_index])
  2638. break;
  2639. }
  2640. VM_BUG_ON(zone_index == -1);
  2641. movable_zone = zone_index;
  2642. }
  2643. /*
  2644. * The zone ranges provided by the architecture do not include ZONE_MOVABLE
  2645. * because it is sized independant of architecture. Unlike the other zones,
  2646. * the starting point for ZONE_MOVABLE is not fixed. It may be different
  2647. * in each node depending on the size of each node and how evenly kernelcore
  2648. * is distributed. This helper function adjusts the zone ranges
  2649. * provided by the architecture for a given node by using the end of the
  2650. * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
  2651. * zones within a node are in order of monotonic increases memory addresses
  2652. */
  2653. void __meminit adjust_zone_range_for_zone_movable(int nid,
  2654. unsigned long zone_type,
  2655. unsigned long node_start_pfn,
  2656. unsigned long node_end_pfn,
  2657. unsigned long *zone_start_pfn,
  2658. unsigned long *zone_end_pfn)
  2659. {
  2660. /* Only adjust if ZONE_MOVABLE is on this node */
  2661. if (zone_movable_pfn[nid]) {
  2662. /* Size ZONE_MOVABLE */
  2663. if (zone_type == ZONE_MOVABLE) {
  2664. *zone_start_pfn = zone_movable_pfn[nid];
  2665. *zone_end_pfn = min(node_end_pfn,
  2666. arch_zone_highest_possible_pfn[movable_zone]);
  2667. /* Adjust for ZONE_MOVABLE starting within this range */
  2668. } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
  2669. *zone_end_pfn > zone_movable_pfn[nid]) {
  2670. *zone_end_pfn = zone_movable_pfn[nid];
  2671. /* Check if this whole range is within ZONE_MOVABLE */
  2672. } else if (*zone_start_pfn >= zone_movable_pfn[nid])
  2673. *zone_start_pfn = *zone_end_pfn;
  2674. }
  2675. }
  2676. /*
  2677. * Return the number of pages a zone spans in a node, including holes
  2678. * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
  2679. */
  2680. static unsigned long __meminit zone_spanned_pages_in_node(int nid,
  2681. unsigned long zone_type,
  2682. unsigned long *ignored)
  2683. {
  2684. unsigned long node_start_pfn, node_end_pfn;
  2685. unsigned long zone_start_pfn, zone_end_pfn;
  2686. /* Get the start and end of the node and zone */
  2687. get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
  2688. zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
  2689. zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
  2690. adjust_zone_range_for_zone_movable(nid, zone_type,
  2691. node_start_pfn, node_end_pfn,
  2692. &zone_start_pfn, &zone_end_pfn);
  2693. /* Check that this node has pages within the zone's required range */
  2694. if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
  2695. return 0;
  2696. /* Move the zone boundaries inside the node if necessary */
  2697. zone_end_pfn = min(zone_end_pfn, node_end_pfn);
  2698. zone_start_pfn = max(zone_start_pfn, node_start_pfn);
  2699. /* Return the spanned pages */
  2700. return zone_end_pfn - zone_start_pfn;
  2701. }
  2702. /*
  2703. * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
  2704. * then all holes in the requested range will be accounted for.
  2705. */
  2706. unsigned long __meminit __absent_pages_in_range(int nid,
  2707. unsigned long range_start_pfn,
  2708. unsigned long range_end_pfn)
  2709. {
  2710. int i = 0;
  2711. unsigned long prev_end_pfn = 0, hole_pages = 0;
  2712. unsigned long start_pfn;
  2713. /* Find the end_pfn of the first active range of pfns in the node */
  2714. i = first_active_region_index_in_nid(nid);
  2715. if (i == -1)
  2716. return 0;
  2717. prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
  2718. /* Account for ranges before physical memory on this node */
  2719. if (early_node_map[i].start_pfn > range_start_pfn)
  2720. hole_pages = prev_end_pfn - range_start_pfn;
  2721. /* Find all holes for the zone within the node */
  2722. for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
  2723. /* No need to continue if prev_end_pfn is outside the zone */
  2724. if (prev_end_pfn >= range_end_pfn)
  2725. break;
  2726. /* Make sure the end of the zone is not within the hole */
  2727. start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
  2728. prev_end_pfn = max(prev_end_pfn, range_start_pfn);
  2729. /* Update the hole size cound and move on */
  2730. if (start_pfn > range_start_pfn) {
  2731. BUG_ON(prev_end_pfn > start_pfn);
  2732. hole_pages += start_pfn - prev_end_pfn;
  2733. }
  2734. prev_end_pfn = early_node_map[i].end_pfn;
  2735. }
  2736. /* Account for ranges past physical memory on this node */
  2737. if (range_end_pfn > prev_end_pfn)
  2738. hole_pages += range_end_pfn -
  2739. max(range_start_pfn, prev_end_pfn);
  2740. return hole_pages;
  2741. }
  2742. /**
  2743. * absent_pages_in_range - Return number of page frames in holes within a range
  2744. * @start_pfn: The start PFN to start searching for holes
  2745. * @end_pfn: The end PFN to stop searching for holes
  2746. *
  2747. * It returns the number of pages frames in memory holes within a range.
  2748. */
  2749. unsigned long __init absent_pages_in_range(unsigned long start_pfn,
  2750. unsigned long end_pfn)
  2751. {
  2752. return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
  2753. }
  2754. /* Return the number of page frames in holes in a zone on a node */
  2755. static unsigned long __meminit zone_absent_pages_in_node(int nid,
  2756. unsigned long zone_type,
  2757. unsigned long *ignored)
  2758. {
  2759. unsigned long node_start_pfn, node_end_pfn;
  2760. unsigned long zone_start_pfn, zone_end_pfn;
  2761. get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
  2762. zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
  2763. node_start_pfn);
  2764. zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
  2765. node_end_pfn);
  2766. adjust_zone_range_for_zone_movable(nid, zone_type,
  2767. node_start_pfn, node_end_pfn,
  2768. &zone_start_pfn, &zone_end_pfn);
  2769. return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
  2770. }
  2771. #else
  2772. static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
  2773. unsigned long zone_type,
  2774. unsigned long *zones_size)
  2775. {
  2776. return zones_size[zone_type];
  2777. }
  2778. static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
  2779. unsigned long zone_type,
  2780. unsigned long *zholes_size)
  2781. {
  2782. if (!zholes_size)
  2783. return 0;
  2784. return zholes_size[zone_type];
  2785. }
  2786. #endif
  2787. static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
  2788. unsigned long *zones_size, unsigned long *zholes_size)
  2789. {
  2790. unsigned long realtotalpages, totalpages = 0;
  2791. enum zone_type i;
  2792. for (i = 0; i < MAX_NR_ZONES; i++)
  2793. totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
  2794. zones_size);
  2795. pgdat->node_spanned_pages = totalpages;
  2796. realtotalpages = totalpages;
  2797. for (i = 0; i < MAX_NR_ZONES; i++)
  2798. realtotalpages -=
  2799. zone_absent_pages_in_node(pgdat->node_id, i,
  2800. zholes_size);
  2801. pgdat->node_present_pages = realtotalpages;
  2802. printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
  2803. realtotalpages);
  2804. }
  2805. #ifndef CONFIG_SPARSEMEM
  2806. /*
  2807. * Calculate the size of the zone->blockflags rounded to an unsigned long
  2808. * Start by making sure zonesize is a multiple of MAX_ORDER-1 by rounding up
  2809. * Then figure 1 NR_PAGEBLOCK_BITS worth of bits per MAX_ORDER-1, finally
  2810. * round what is now in bits to nearest long in bits, then return it in
  2811. * bytes.
  2812. */
  2813. static unsigned long __init usemap_size(unsigned long zonesize)
  2814. {
  2815. unsigned long usemapsize;
  2816. usemapsize = roundup(zonesize, MAX_ORDER_NR_PAGES);
  2817. usemapsize = usemapsize >> (MAX_ORDER-1);
  2818. usemapsize *= NR_PAGEBLOCK_BITS;
  2819. usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
  2820. return usemapsize / 8;
  2821. }
  2822. static void __init setup_usemap(struct pglist_data *pgdat,
  2823. struct zone *zone, unsigned long zonesize)
  2824. {
  2825. unsigned long usemapsize = usemap_size(zonesize);
  2826. zone->pageblock_flags = NULL;
  2827. if (usemapsize) {
  2828. zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
  2829. memset(zone->pageblock_flags, 0, usemapsize);
  2830. }
  2831. }
  2832. #else
  2833. static void inline setup_usemap(struct pglist_data *pgdat,
  2834. struct zone *zone, unsigned long zonesize) {}
  2835. #endif /* CONFIG_SPARSEMEM */
  2836. /*
  2837. * Set up the zone data structures:
  2838. * - mark all pages reserved
  2839. * - mark all memory queues empty
  2840. * - clear the memory bitmaps
  2841. */
  2842. static void __meminit free_area_init_core(struct pglist_data *pgdat,
  2843. unsigned long *zones_size, unsigned long *zholes_size)
  2844. {
  2845. enum zone_type j;
  2846. int nid = pgdat->node_id;
  2847. unsigned long zone_start_pfn = pgdat->node_start_pfn;
  2848. int ret;
  2849. pgdat_resize_init(pgdat);
  2850. pgdat->nr_zones = 0;
  2851. init_waitqueue_head(&pgdat->kswapd_wait);
  2852. pgdat->kswapd_max_order = 0;
  2853. for (j = 0; j < MAX_NR_ZONES; j++) {
  2854. struct zone *zone = pgdat->node_zones + j;
  2855. unsigned long size, realsize, memmap_pages;
  2856. size = zone_spanned_pages_in_node(nid, j, zones_size);
  2857. realsize = size - zone_absent_pages_in_node(nid, j,
  2858. zholes_size);
  2859. /*
  2860. * Adjust realsize so that it accounts for how much memory
  2861. * is used by this zone for memmap. This affects the watermark
  2862. * and per-cpu initialisations
  2863. */
  2864. memmap_pages = (size * sizeof(struct page)) >> PAGE_SHIFT;
  2865. if (realsize >= memmap_pages) {
  2866. realsize -= memmap_pages;
  2867. printk(KERN_DEBUG
  2868. " %s zone: %lu pages used for memmap\n",
  2869. zone_names[j], memmap_pages);
  2870. } else
  2871. printk(KERN_WARNING
  2872. " %s zone: %lu pages exceeds realsize %lu\n",
  2873. zone_names[j], memmap_pages, realsize);
  2874. /* Account for reserved pages */
  2875. if (j == 0 && realsize > dma_reserve) {
  2876. realsize -= dma_reserve;
  2877. printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
  2878. zone_names[0], dma_reserve);
  2879. }
  2880. if (!is_highmem_idx(j))
  2881. nr_kernel_pages += realsize;
  2882. nr_all_pages += realsize;
  2883. zone->spanned_pages = size;
  2884. zone->present_pages = realsize;
  2885. #ifdef CONFIG_NUMA
  2886. zone->node = nid;
  2887. zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
  2888. / 100;
  2889. zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
  2890. #endif
  2891. zone->name = zone_names[j];
  2892. spin_lock_init(&zone->lock);
  2893. spin_lock_init(&zone->lru_lock);
  2894. zone_seqlock_init(zone);
  2895. zone->zone_pgdat = pgdat;
  2896. zone->prev_priority = DEF_PRIORITY;
  2897. zone_pcp_init(zone);
  2898. INIT_LIST_HEAD(&zone->active_list);
  2899. INIT_LIST_HEAD(&zone->inactive_list);
  2900. zone->nr_scan_active = 0;
  2901. zone->nr_scan_inactive = 0;
  2902. zap_zone_vm_stats(zone);
  2903. atomic_set(&zone->reclaim_in_progress, 0);
  2904. if (!size)
  2905. continue;
  2906. setup_usemap(pgdat, zone, size);
  2907. ret = init_currently_empty_zone(zone, zone_start_pfn,
  2908. size, MEMMAP_EARLY);
  2909. BUG_ON(ret);
  2910. zone_start_pfn += size;
  2911. }
  2912. }
  2913. static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
  2914. {
  2915. /* Skip empty nodes */
  2916. if (!pgdat->node_spanned_pages)
  2917. return;
  2918. #ifdef CONFIG_FLAT_NODE_MEM_MAP
  2919. /* ia64 gets its own node_mem_map, before this, without bootmem */
  2920. if (!pgdat->node_mem_map) {
  2921. unsigned long size, start, end;
  2922. struct page *map;
  2923. /*
  2924. * The zone's endpoints aren't required to be MAX_ORDER
  2925. * aligned but the node_mem_map endpoints must be in order
  2926. * for the buddy allocator to function correctly.
  2927. */
  2928. start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
  2929. end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
  2930. end = ALIGN(end, MAX_ORDER_NR_PAGES);
  2931. size = (end - start) * sizeof(struct page);
  2932. map = alloc_remap(pgdat->node_id, size);
  2933. if (!map)
  2934. map = alloc_bootmem_node(pgdat, size);
  2935. pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
  2936. }
  2937. #ifndef CONFIG_NEED_MULTIPLE_NODES
  2938. /*
  2939. * With no DISCONTIG, the global mem_map is just set as node 0's
  2940. */
  2941. if (pgdat == NODE_DATA(0)) {
  2942. mem_map = NODE_DATA(0)->node_mem_map;
  2943. #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
  2944. if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
  2945. mem_map -= pgdat->node_start_pfn;
  2946. #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
  2947. }
  2948. #endif
  2949. #endif /* CONFIG_FLAT_NODE_MEM_MAP */
  2950. }
  2951. void __meminit free_area_init_node(int nid, struct pglist_data *pgdat,
  2952. unsigned long *zones_size, unsigned long node_start_pfn,
  2953. unsigned long *zholes_size)
  2954. {
  2955. pgdat->node_id = nid;
  2956. pgdat->node_start_pfn = node_start_pfn;
  2957. calculate_node_totalpages(pgdat, zones_size, zholes_size);
  2958. alloc_node_mem_map(pgdat);
  2959. free_area_init_core(pgdat, zones_size, zholes_size);
  2960. }
  2961. #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
  2962. #if MAX_NUMNODES > 1
  2963. /*
  2964. * Figure out the number of possible node ids.
  2965. */
  2966. static void __init setup_nr_node_ids(void)
  2967. {
  2968. unsigned int node;
  2969. unsigned int highest = 0;
  2970. for_each_node_mask(node, node_possible_map)
  2971. highest = node;
  2972. nr_node_ids = highest + 1;
  2973. }
  2974. #else
  2975. static inline void setup_nr_node_ids(void)
  2976. {
  2977. }
  2978. #endif
  2979. /**
  2980. * add_active_range - Register a range of PFNs backed by physical memory
  2981. * @nid: The node ID the range resides on
  2982. * @start_pfn: The start PFN of the available physical memory
  2983. * @end_pfn: The end PFN of the available physical memory
  2984. *
  2985. * These ranges are stored in an early_node_map[] and later used by
  2986. * free_area_init_nodes() to calculate zone sizes and holes. If the
  2987. * range spans a memory hole, it is up to the architecture to ensure
  2988. * the memory is not freed by the bootmem allocator. If possible
  2989. * the range being registered will be merged with existing ranges.
  2990. */
  2991. void __init add_active_range(unsigned int nid, unsigned long start_pfn,
  2992. unsigned long end_pfn)
  2993. {
  2994. int i;
  2995. printk(KERN_DEBUG "Entering add_active_range(%d, %lu, %lu) "
  2996. "%d entries of %d used\n",
  2997. nid, start_pfn, end_pfn,
  2998. nr_nodemap_entries, MAX_ACTIVE_REGIONS);
  2999. /* Merge with existing active regions if possible */
  3000. for (i = 0; i < nr_nodemap_entries; i++) {
  3001. if (early_node_map[i].nid != nid)
  3002. continue;
  3003. /* Skip if an existing region covers this new one */
  3004. if (start_pfn >= early_node_map[i].start_pfn &&
  3005. end_pfn <= early_node_map[i].end_pfn)
  3006. return;
  3007. /* Merge forward if suitable */
  3008. if (start_pfn <= early_node_map[i].end_pfn &&
  3009. end_pfn > early_node_map[i].end_pfn) {
  3010. early_node_map[i].end_pfn = end_pfn;
  3011. return;
  3012. }
  3013. /* Merge backward if suitable */
  3014. if (start_pfn < early_node_map[i].end_pfn &&
  3015. end_pfn >= early_node_map[i].start_pfn) {
  3016. early_node_map[i].start_pfn = start_pfn;
  3017. return;
  3018. }
  3019. }
  3020. /* Check that early_node_map is large enough */
  3021. if (i >= MAX_ACTIVE_REGIONS) {
  3022. printk(KERN_CRIT "More than %d memory regions, truncating\n",
  3023. MAX_ACTIVE_REGIONS);
  3024. return;
  3025. }
  3026. early_node_map[i].nid = nid;
  3027. early_node_map[i].start_pfn = start_pfn;
  3028. early_node_map[i].end_pfn = end_pfn;
  3029. nr_nodemap_entries = i + 1;
  3030. }
  3031. /**
  3032. * shrink_active_range - Shrink an existing registered range of PFNs
  3033. * @nid: The node id the range is on that should be shrunk
  3034. * @old_end_pfn: The old end PFN of the range
  3035. * @new_end_pfn: The new PFN of the range
  3036. *
  3037. * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
  3038. * The map is kept at the end physical page range that has already been
  3039. * registered with add_active_range(). This function allows an arch to shrink
  3040. * an existing registered range.
  3041. */
  3042. void __init shrink_active_range(unsigned int nid, unsigned long old_end_pfn,
  3043. unsigned long new_end_pfn)
  3044. {
  3045. int i;
  3046. /* Find the old active region end and shrink */
  3047. for_each_active_range_index_in_nid(i, nid)
  3048. if (early_node_map[i].end_pfn == old_end_pfn) {
  3049. early_node_map[i].end_pfn = new_end_pfn;
  3050. break;
  3051. }
  3052. }
  3053. /**
  3054. * remove_all_active_ranges - Remove all currently registered regions
  3055. *
  3056. * During discovery, it may be found that a table like SRAT is invalid
  3057. * and an alternative discovery method must be used. This function removes
  3058. * all currently registered regions.
  3059. */
  3060. void __init remove_all_active_ranges(void)
  3061. {
  3062. memset(early_node_map, 0, sizeof(early_node_map));
  3063. nr_nodemap_entries = 0;
  3064. #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
  3065. memset(node_boundary_start_pfn, 0, sizeof(node_boundary_start_pfn));
  3066. memset(node_boundary_end_pfn, 0, sizeof(node_boundary_end_pfn));
  3067. #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
  3068. }
  3069. /* Compare two active node_active_regions */
  3070. static int __init cmp_node_active_region(const void *a, const void *b)
  3071. {
  3072. struct node_active_region *arange = (struct node_active_region *)a;
  3073. struct node_active_region *brange = (struct node_active_region *)b;
  3074. /* Done this way to avoid overflows */
  3075. if (arange->start_pfn > brange->start_pfn)
  3076. return 1;
  3077. if (arange->start_pfn < brange->start_pfn)
  3078. return -1;
  3079. return 0;
  3080. }
  3081. /* sort the node_map by start_pfn */
  3082. static void __init sort_node_map(void)
  3083. {
  3084. sort(early_node_map, (size_t)nr_nodemap_entries,
  3085. sizeof(struct node_active_region),
  3086. cmp_node_active_region, NULL);
  3087. }
  3088. /* Find the lowest pfn for a node */
  3089. unsigned long __init find_min_pfn_for_node(unsigned long nid)
  3090. {
  3091. int i;
  3092. unsigned long min_pfn = ULONG_MAX;
  3093. /* Assuming a sorted map, the first range found has the starting pfn */
  3094. for_each_active_range_index_in_nid(i, nid)
  3095. min_pfn = min(min_pfn, early_node_map[i].start_pfn);
  3096. if (min_pfn == ULONG_MAX) {
  3097. printk(KERN_WARNING
  3098. "Could not find start_pfn for node %lu\n", nid);
  3099. return 0;
  3100. }
  3101. return min_pfn;
  3102. }
  3103. /**
  3104. * find_min_pfn_with_active_regions - Find the minimum PFN registered
  3105. *
  3106. * It returns the minimum PFN based on information provided via
  3107. * add_active_range().
  3108. */
  3109. unsigned long __init find_min_pfn_with_active_regions(void)
  3110. {
  3111. return find_min_pfn_for_node(MAX_NUMNODES);
  3112. }
  3113. /**
  3114. * find_max_pfn_with_active_regions - Find the maximum PFN registered
  3115. *
  3116. * It returns the maximum PFN based on information provided via
  3117. * add_active_range().
  3118. */
  3119. unsigned long __init find_max_pfn_with_active_regions(void)
  3120. {
  3121. int i;
  3122. unsigned long max_pfn = 0;
  3123. for (i = 0; i < nr_nodemap_entries; i++)
  3124. max_pfn = max(max_pfn, early_node_map[i].end_pfn);
  3125. return max_pfn;
  3126. }
  3127. /*
  3128. * early_calculate_totalpages()
  3129. * Sum pages in active regions for movable zone.
  3130. * Populate N_HIGH_MEMORY for calculating usable_nodes.
  3131. */
  3132. unsigned long __init early_calculate_totalpages(void)
  3133. {
  3134. int i;
  3135. unsigned long totalpages = 0;
  3136. for (i = 0; i < nr_nodemap_entries; i++) {
  3137. unsigned long pages = early_node_map[i].end_pfn -
  3138. early_node_map[i].start_pfn;
  3139. totalpages += pages;
  3140. if (pages)
  3141. node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
  3142. }
  3143. return totalpages;
  3144. }
  3145. /*
  3146. * Find the PFN the Movable zone begins in each node. Kernel memory
  3147. * is spread evenly between nodes as long as the nodes have enough
  3148. * memory. When they don't, some nodes will have more kernelcore than
  3149. * others
  3150. */
  3151. void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
  3152. {
  3153. int i, nid;
  3154. unsigned long usable_startpfn;
  3155. unsigned long kernelcore_node, kernelcore_remaining;
  3156. unsigned long totalpages = early_calculate_totalpages();
  3157. int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
  3158. /*
  3159. * If movablecore was specified, calculate what size of
  3160. * kernelcore that corresponds so that memory usable for
  3161. * any allocation type is evenly spread. If both kernelcore
  3162. * and movablecore are specified, then the value of kernelcore
  3163. * will be used for required_kernelcore if it's greater than
  3164. * what movablecore would have allowed.
  3165. */
  3166. if (required_movablecore) {
  3167. unsigned long corepages;
  3168. /*
  3169. * Round-up so that ZONE_MOVABLE is at least as large as what
  3170. * was requested by the user
  3171. */
  3172. required_movablecore =
  3173. roundup(required_movablecore, MAX_ORDER_NR_PAGES);
  3174. corepages = totalpages - required_movablecore;
  3175. required_kernelcore = max(required_kernelcore, corepages);
  3176. }
  3177. /* If kernelcore was not specified, there is no ZONE_MOVABLE */
  3178. if (!required_kernelcore)
  3179. return;
  3180. /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
  3181. find_usable_zone_for_movable();
  3182. usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
  3183. restart:
  3184. /* Spread kernelcore memory as evenly as possible throughout nodes */
  3185. kernelcore_node = required_kernelcore / usable_nodes;
  3186. for_each_node_state(nid, N_HIGH_MEMORY) {
  3187. /*
  3188. * Recalculate kernelcore_node if the division per node
  3189. * now exceeds what is necessary to satisfy the requested
  3190. * amount of memory for the kernel
  3191. */
  3192. if (required_kernelcore < kernelcore_node)
  3193. kernelcore_node = required_kernelcore / usable_nodes;
  3194. /*
  3195. * As the map is walked, we track how much memory is usable
  3196. * by the kernel using kernelcore_remaining. When it is
  3197. * 0, the rest of the node is usable by ZONE_MOVABLE
  3198. */
  3199. kernelcore_remaining = kernelcore_node;
  3200. /* Go through each range of PFNs within this node */
  3201. for_each_active_range_index_in_nid(i, nid) {
  3202. unsigned long start_pfn, end_pfn;
  3203. unsigned long size_pages;
  3204. start_pfn = max(early_node_map[i].start_pfn,
  3205. zone_movable_pfn[nid]);
  3206. end_pfn = early_node_map[i].end_pfn;
  3207. if (start_pfn >= end_pfn)
  3208. continue;
  3209. /* Account for what is only usable for kernelcore */
  3210. if (start_pfn < usable_startpfn) {
  3211. unsigned long kernel_pages;
  3212. kernel_pages = min(end_pfn, usable_startpfn)
  3213. - start_pfn;
  3214. kernelcore_remaining -= min(kernel_pages,
  3215. kernelcore_remaining);
  3216. required_kernelcore -= min(kernel_pages,
  3217. required_kernelcore);
  3218. /* Continue if range is now fully accounted */
  3219. if (end_pfn <= usable_startpfn) {
  3220. /*
  3221. * Push zone_movable_pfn to the end so
  3222. * that if we have to rebalance
  3223. * kernelcore across nodes, we will
  3224. * not double account here
  3225. */
  3226. zone_movable_pfn[nid] = end_pfn;
  3227. continue;
  3228. }
  3229. start_pfn = usable_startpfn;
  3230. }
  3231. /*
  3232. * The usable PFN range for ZONE_MOVABLE is from
  3233. * start_pfn->end_pfn. Calculate size_pages as the
  3234. * number of pages used as kernelcore
  3235. */
  3236. size_pages = end_pfn - start_pfn;
  3237. if (size_pages > kernelcore_remaining)
  3238. size_pages = kernelcore_remaining;
  3239. zone_movable_pfn[nid] = start_pfn + size_pages;
  3240. /*
  3241. * Some kernelcore has been met, update counts and
  3242. * break if the kernelcore for this node has been
  3243. * satisified
  3244. */
  3245. required_kernelcore -= min(required_kernelcore,
  3246. size_pages);
  3247. kernelcore_remaining -= size_pages;
  3248. if (!kernelcore_remaining)
  3249. break;
  3250. }
  3251. }
  3252. /*
  3253. * If there is still required_kernelcore, we do another pass with one
  3254. * less node in the count. This will push zone_movable_pfn[nid] further
  3255. * along on the nodes that still have memory until kernelcore is
  3256. * satisified
  3257. */
  3258. usable_nodes--;
  3259. if (usable_nodes && required_kernelcore > usable_nodes)
  3260. goto restart;
  3261. /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
  3262. for (nid = 0; nid < MAX_NUMNODES; nid++)
  3263. zone_movable_pfn[nid] =
  3264. roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
  3265. }
  3266. /* Any regular memory on that node ? */
  3267. static void check_for_regular_memory(pg_data_t *pgdat)
  3268. {
  3269. #ifdef CONFIG_HIGHMEM
  3270. enum zone_type zone_type;
  3271. for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
  3272. struct zone *zone = &pgdat->node_zones[zone_type];
  3273. if (zone->present_pages)
  3274. node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
  3275. }
  3276. #endif
  3277. }
  3278. /**
  3279. * free_area_init_nodes - Initialise all pg_data_t and zone data
  3280. * @max_zone_pfn: an array of max PFNs for each zone
  3281. *
  3282. * This will call free_area_init_node() for each active node in the system.
  3283. * Using the page ranges provided by add_active_range(), the size of each
  3284. * zone in each node and their holes is calculated. If the maximum PFN
  3285. * between two adjacent zones match, it is assumed that the zone is empty.
  3286. * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
  3287. * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
  3288. * starts where the previous one ended. For example, ZONE_DMA32 starts
  3289. * at arch_max_dma_pfn.
  3290. */
  3291. void __init free_area_init_nodes(unsigned long *max_zone_pfn)
  3292. {
  3293. unsigned long nid;
  3294. enum zone_type i;
  3295. /* Sort early_node_map as initialisation assumes it is sorted */
  3296. sort_node_map();
  3297. /* Record where the zone boundaries are */
  3298. memset(arch_zone_lowest_possible_pfn, 0,
  3299. sizeof(arch_zone_lowest_possible_pfn));
  3300. memset(arch_zone_highest_possible_pfn, 0,
  3301. sizeof(arch_zone_highest_possible_pfn));
  3302. arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
  3303. arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
  3304. for (i = 1; i < MAX_NR_ZONES; i++) {
  3305. if (i == ZONE_MOVABLE)
  3306. continue;
  3307. arch_zone_lowest_possible_pfn[i] =
  3308. arch_zone_highest_possible_pfn[i-1];
  3309. arch_zone_highest_possible_pfn[i] =
  3310. max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
  3311. }
  3312. arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
  3313. arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
  3314. /* Find the PFNs that ZONE_MOVABLE begins at in each node */
  3315. memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
  3316. find_zone_movable_pfns_for_nodes(zone_movable_pfn);
  3317. /* Print out the zone ranges */
  3318. printk("Zone PFN ranges:\n");
  3319. for (i = 0; i < MAX_NR_ZONES; i++) {
  3320. if (i == ZONE_MOVABLE)
  3321. continue;
  3322. printk(" %-8s %8lu -> %8lu\n",
  3323. zone_names[i],
  3324. arch_zone_lowest_possible_pfn[i],
  3325. arch_zone_highest_possible_pfn[i]);
  3326. }
  3327. /* Print out the PFNs ZONE_MOVABLE begins at in each node */
  3328. printk("Movable zone start PFN for each node\n");
  3329. for (i = 0; i < MAX_NUMNODES; i++) {
  3330. if (zone_movable_pfn[i])
  3331. printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
  3332. }
  3333. /* Print out the early_node_map[] */
  3334. printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
  3335. for (i = 0; i < nr_nodemap_entries; i++)
  3336. printk(" %3d: %8lu -> %8lu\n", early_node_map[i].nid,
  3337. early_node_map[i].start_pfn,
  3338. early_node_map[i].end_pfn);
  3339. /* Initialise every node */
  3340. setup_nr_node_ids();
  3341. for_each_online_node(nid) {
  3342. pg_data_t *pgdat = NODE_DATA(nid);
  3343. free_area_init_node(nid, pgdat, NULL,
  3344. find_min_pfn_for_node(nid), NULL);
  3345. /* Any memory on that node */
  3346. if (pgdat->node_present_pages)
  3347. node_set_state(nid, N_HIGH_MEMORY);
  3348. check_for_regular_memory(pgdat);
  3349. }
  3350. }
  3351. static int __init cmdline_parse_core(char *p, unsigned long *core)
  3352. {
  3353. unsigned long long coremem;
  3354. if (!p)
  3355. return -EINVAL;
  3356. coremem = memparse(p, &p);
  3357. *core = coremem >> PAGE_SHIFT;
  3358. /* Paranoid check that UL is enough for the coremem value */
  3359. WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
  3360. return 0;
  3361. }
  3362. /*
  3363. * kernelcore=size sets the amount of memory for use for allocations that
  3364. * cannot be reclaimed or migrated.
  3365. */
  3366. static int __init cmdline_parse_kernelcore(char *p)
  3367. {
  3368. return cmdline_parse_core(p, &required_kernelcore);
  3369. }
  3370. /*
  3371. * movablecore=size sets the amount of memory for use for allocations that
  3372. * can be reclaimed or migrated.
  3373. */
  3374. static int __init cmdline_parse_movablecore(char *p)
  3375. {
  3376. return cmdline_parse_core(p, &required_movablecore);
  3377. }
  3378. early_param("kernelcore", cmdline_parse_kernelcore);
  3379. early_param("movablecore", cmdline_parse_movablecore);
  3380. #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
  3381. /**
  3382. * set_dma_reserve - set the specified number of pages reserved in the first zone
  3383. * @new_dma_reserve: The number of pages to mark reserved
  3384. *
  3385. * The per-cpu batchsize and zone watermarks are determined by present_pages.
  3386. * In the DMA zone, a significant percentage may be consumed by kernel image
  3387. * and other unfreeable allocations which can skew the watermarks badly. This
  3388. * function may optionally be used to account for unfreeable pages in the
  3389. * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
  3390. * smaller per-cpu batchsize.
  3391. */
  3392. void __init set_dma_reserve(unsigned long new_dma_reserve)
  3393. {
  3394. dma_reserve = new_dma_reserve;
  3395. }
  3396. #ifndef CONFIG_NEED_MULTIPLE_NODES
  3397. static bootmem_data_t contig_bootmem_data;
  3398. struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
  3399. EXPORT_SYMBOL(contig_page_data);
  3400. #endif
  3401. void __init free_area_init(unsigned long *zones_size)
  3402. {
  3403. free_area_init_node(0, NODE_DATA(0), zones_size,
  3404. __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
  3405. }
  3406. static int page_alloc_cpu_notify(struct notifier_block *self,
  3407. unsigned long action, void *hcpu)
  3408. {
  3409. int cpu = (unsigned long)hcpu;
  3410. if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
  3411. local_irq_disable();
  3412. __drain_pages(cpu);
  3413. vm_events_fold_cpu(cpu);
  3414. local_irq_enable();
  3415. refresh_cpu_vm_stats(cpu);
  3416. }
  3417. return NOTIFY_OK;
  3418. }
  3419. void __init page_alloc_init(void)
  3420. {
  3421. hotcpu_notifier(page_alloc_cpu_notify, 0);
  3422. }
  3423. /*
  3424. * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
  3425. * or min_free_kbytes changes.
  3426. */
  3427. static void calculate_totalreserve_pages(void)
  3428. {
  3429. struct pglist_data *pgdat;
  3430. unsigned long reserve_pages = 0;
  3431. enum zone_type i, j;
  3432. for_each_online_pgdat(pgdat) {
  3433. for (i = 0; i < MAX_NR_ZONES; i++) {
  3434. struct zone *zone = pgdat->node_zones + i;
  3435. unsigned long max = 0;
  3436. /* Find valid and maximum lowmem_reserve in the zone */
  3437. for (j = i; j < MAX_NR_ZONES; j++) {
  3438. if (zone->lowmem_reserve[j] > max)
  3439. max = zone->lowmem_reserve[j];
  3440. }
  3441. /* we treat pages_high as reserved pages. */
  3442. max += zone->pages_high;
  3443. if (max > zone->present_pages)
  3444. max = zone->present_pages;
  3445. reserve_pages += max;
  3446. }
  3447. }
  3448. totalreserve_pages = reserve_pages;
  3449. }
  3450. /*
  3451. * setup_per_zone_lowmem_reserve - called whenever
  3452. * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
  3453. * has a correct pages reserved value, so an adequate number of
  3454. * pages are left in the zone after a successful __alloc_pages().
  3455. */
  3456. static void setup_per_zone_lowmem_reserve(void)
  3457. {
  3458. struct pglist_data *pgdat;
  3459. enum zone_type j, idx;
  3460. for_each_online_pgdat(pgdat) {
  3461. for (j = 0; j < MAX_NR_ZONES; j++) {
  3462. struct zone *zone = pgdat->node_zones + j;
  3463. unsigned long present_pages = zone->present_pages;
  3464. zone->lowmem_reserve[j] = 0;
  3465. idx = j;
  3466. while (idx) {
  3467. struct zone *lower_zone;
  3468. idx--;
  3469. if (sysctl_lowmem_reserve_ratio[idx] < 1)
  3470. sysctl_lowmem_reserve_ratio[idx] = 1;
  3471. lower_zone = pgdat->node_zones + idx;
  3472. lower_zone->lowmem_reserve[j] = present_pages /
  3473. sysctl_lowmem_reserve_ratio[idx];
  3474. present_pages += lower_zone->present_pages;
  3475. }
  3476. }
  3477. }
  3478. /* update totalreserve_pages */
  3479. calculate_totalreserve_pages();
  3480. }
  3481. /**
  3482. * setup_per_zone_pages_min - called when min_free_kbytes changes.
  3483. *
  3484. * Ensures that the pages_{min,low,high} values for each zone are set correctly
  3485. * with respect to min_free_kbytes.
  3486. */
  3487. void setup_per_zone_pages_min(void)
  3488. {
  3489. unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
  3490. unsigned long lowmem_pages = 0;
  3491. struct zone *zone;
  3492. unsigned long flags;
  3493. /* Calculate total number of !ZONE_HIGHMEM pages */
  3494. for_each_zone(zone) {
  3495. if (!is_highmem(zone))
  3496. lowmem_pages += zone->present_pages;
  3497. }
  3498. for_each_zone(zone) {
  3499. u64 tmp;
  3500. spin_lock_irqsave(&zone->lru_lock, flags);
  3501. tmp = (u64)pages_min * zone->present_pages;
  3502. do_div(tmp, lowmem_pages);
  3503. if (is_highmem(zone)) {
  3504. /*
  3505. * __GFP_HIGH and PF_MEMALLOC allocations usually don't
  3506. * need highmem pages, so cap pages_min to a small
  3507. * value here.
  3508. *
  3509. * The (pages_high-pages_low) and (pages_low-pages_min)
  3510. * deltas controls asynch page reclaim, and so should
  3511. * not be capped for highmem.
  3512. */
  3513. int min_pages;
  3514. min_pages = zone->present_pages / 1024;
  3515. if (min_pages < SWAP_CLUSTER_MAX)
  3516. min_pages = SWAP_CLUSTER_MAX;
  3517. if (min_pages > 128)
  3518. min_pages = 128;
  3519. zone->pages_min = min_pages;
  3520. } else {
  3521. /*
  3522. * If it's a lowmem zone, reserve a number of pages
  3523. * proportionate to the zone's size.
  3524. */
  3525. zone->pages_min = tmp;
  3526. }
  3527. zone->pages_low = zone->pages_min + (tmp >> 2);
  3528. zone->pages_high = zone->pages_min + (tmp >> 1);
  3529. setup_zone_migrate_reserve(zone);
  3530. spin_unlock_irqrestore(&zone->lru_lock, flags);
  3531. }
  3532. /* update totalreserve_pages */
  3533. calculate_totalreserve_pages();
  3534. }
  3535. /*
  3536. * Initialise min_free_kbytes.
  3537. *
  3538. * For small machines we want it small (128k min). For large machines
  3539. * we want it large (64MB max). But it is not linear, because network
  3540. * bandwidth does not increase linearly with machine size. We use
  3541. *
  3542. * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
  3543. * min_free_kbytes = sqrt(lowmem_kbytes * 16)
  3544. *
  3545. * which yields
  3546. *
  3547. * 16MB: 512k
  3548. * 32MB: 724k
  3549. * 64MB: 1024k
  3550. * 128MB: 1448k
  3551. * 256MB: 2048k
  3552. * 512MB: 2896k
  3553. * 1024MB: 4096k
  3554. * 2048MB: 5792k
  3555. * 4096MB: 8192k
  3556. * 8192MB: 11584k
  3557. * 16384MB: 16384k
  3558. */
  3559. static int __init init_per_zone_pages_min(void)
  3560. {
  3561. unsigned long lowmem_kbytes;
  3562. lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
  3563. min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
  3564. if (min_free_kbytes < 128)
  3565. min_free_kbytes = 128;
  3566. if (min_free_kbytes > 65536)
  3567. min_free_kbytes = 65536;
  3568. setup_per_zone_pages_min();
  3569. setup_per_zone_lowmem_reserve();
  3570. return 0;
  3571. }
  3572. module_init(init_per_zone_pages_min)
  3573. /*
  3574. * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
  3575. * that we can call two helper functions whenever min_free_kbytes
  3576. * changes.
  3577. */
  3578. int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
  3579. struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
  3580. {
  3581. proc_dointvec(table, write, file, buffer, length, ppos);
  3582. if (write)
  3583. setup_per_zone_pages_min();
  3584. return 0;
  3585. }
  3586. #ifdef CONFIG_NUMA
  3587. int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
  3588. struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
  3589. {
  3590. struct zone *zone;
  3591. int rc;
  3592. rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
  3593. if (rc)
  3594. return rc;
  3595. for_each_zone(zone)
  3596. zone->min_unmapped_pages = (zone->present_pages *
  3597. sysctl_min_unmapped_ratio) / 100;
  3598. return 0;
  3599. }
  3600. int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
  3601. struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
  3602. {
  3603. struct zone *zone;
  3604. int rc;
  3605. rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
  3606. if (rc)
  3607. return rc;
  3608. for_each_zone(zone)
  3609. zone->min_slab_pages = (zone->present_pages *
  3610. sysctl_min_slab_ratio) / 100;
  3611. return 0;
  3612. }
  3613. #endif
  3614. /*
  3615. * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
  3616. * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
  3617. * whenever sysctl_lowmem_reserve_ratio changes.
  3618. *
  3619. * The reserve ratio obviously has absolutely no relation with the
  3620. * pages_min watermarks. The lowmem reserve ratio can only make sense
  3621. * if in function of the boot time zone sizes.
  3622. */
  3623. int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
  3624. struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
  3625. {
  3626. proc_dointvec_minmax(table, write, file, buffer, length, ppos);
  3627. setup_per_zone_lowmem_reserve();
  3628. return 0;
  3629. }
  3630. /*
  3631. * percpu_pagelist_fraction - changes the pcp->high for each zone on each
  3632. * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
  3633. * can have before it gets flushed back to buddy allocator.
  3634. */
  3635. int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
  3636. struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
  3637. {
  3638. struct zone *zone;
  3639. unsigned int cpu;
  3640. int ret;
  3641. ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
  3642. if (!write || (ret == -EINVAL))
  3643. return ret;
  3644. for_each_zone(zone) {
  3645. for_each_online_cpu(cpu) {
  3646. unsigned long high;
  3647. high = zone->present_pages / percpu_pagelist_fraction;
  3648. setup_pagelist_highmark(zone_pcp(zone, cpu), high);
  3649. }
  3650. }
  3651. return 0;
  3652. }
  3653. int hashdist = HASHDIST_DEFAULT;
  3654. #ifdef CONFIG_NUMA
  3655. static int __init set_hashdist(char *str)
  3656. {
  3657. if (!str)
  3658. return 0;
  3659. hashdist = simple_strtoul(str, &str, 0);
  3660. return 1;
  3661. }
  3662. __setup("hashdist=", set_hashdist);
  3663. #endif
  3664. /*
  3665. * allocate a large system hash table from bootmem
  3666. * - it is assumed that the hash table must contain an exact power-of-2
  3667. * quantity of entries
  3668. * - limit is the number of hash buckets, not the total allocation size
  3669. */
  3670. void *__init alloc_large_system_hash(const char *tablename,
  3671. unsigned long bucketsize,
  3672. unsigned long numentries,
  3673. int scale,
  3674. int flags,
  3675. unsigned int *_hash_shift,
  3676. unsigned int *_hash_mask,
  3677. unsigned long limit)
  3678. {
  3679. unsigned long long max = limit;
  3680. unsigned long log2qty, size;
  3681. void *table = NULL;
  3682. /* allow the kernel cmdline to have a say */
  3683. if (!numentries) {
  3684. /* round applicable memory size up to nearest megabyte */
  3685. numentries = nr_kernel_pages;
  3686. numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
  3687. numentries >>= 20 - PAGE_SHIFT;
  3688. numentries <<= 20 - PAGE_SHIFT;
  3689. /* limit to 1 bucket per 2^scale bytes of low memory */
  3690. if (scale > PAGE_SHIFT)
  3691. numentries >>= (scale - PAGE_SHIFT);
  3692. else
  3693. numentries <<= (PAGE_SHIFT - scale);
  3694. /* Make sure we've got at least a 0-order allocation.. */
  3695. if (unlikely((numentries * bucketsize) < PAGE_SIZE))
  3696. numentries = PAGE_SIZE / bucketsize;
  3697. }
  3698. numentries = roundup_pow_of_two(numentries);
  3699. /* limit allocation size to 1/16 total memory by default */
  3700. if (max == 0) {
  3701. max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
  3702. do_div(max, bucketsize);
  3703. }
  3704. if (numentries > max)
  3705. numentries = max;
  3706. log2qty = ilog2(numentries);
  3707. do {
  3708. size = bucketsize << log2qty;
  3709. if (flags & HASH_EARLY)
  3710. table = alloc_bootmem(size);
  3711. else if (hashdist)
  3712. table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
  3713. else {
  3714. unsigned long order;
  3715. for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
  3716. ;
  3717. table = (void*) __get_free_pages(GFP_ATOMIC, order);
  3718. /*
  3719. * If bucketsize is not a power-of-two, we may free
  3720. * some pages at the end of hash table.
  3721. */
  3722. if (table) {
  3723. unsigned long alloc_end = (unsigned long)table +
  3724. (PAGE_SIZE << order);
  3725. unsigned long used = (unsigned long)table +
  3726. PAGE_ALIGN(size);
  3727. split_page(virt_to_page(table), order);
  3728. while (used < alloc_end) {
  3729. free_page(used);
  3730. used += PAGE_SIZE;
  3731. }
  3732. }
  3733. }
  3734. } while (!table && size > PAGE_SIZE && --log2qty);
  3735. if (!table)
  3736. panic("Failed to allocate %s hash table\n", tablename);
  3737. printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n",
  3738. tablename,
  3739. (1U << log2qty),
  3740. ilog2(size) - PAGE_SHIFT,
  3741. size);
  3742. if (_hash_shift)
  3743. *_hash_shift = log2qty;
  3744. if (_hash_mask)
  3745. *_hash_mask = (1 << log2qty) - 1;
  3746. return table;
  3747. }
  3748. #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
  3749. struct page *pfn_to_page(unsigned long pfn)
  3750. {
  3751. return __pfn_to_page(pfn);
  3752. }
  3753. unsigned long page_to_pfn(struct page *page)
  3754. {
  3755. return __page_to_pfn(page);
  3756. }
  3757. EXPORT_SYMBOL(pfn_to_page);
  3758. EXPORT_SYMBOL(page_to_pfn);
  3759. #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
  3760. /* Return a pointer to the bitmap storing bits affecting a block of pages */
  3761. static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
  3762. unsigned long pfn)
  3763. {
  3764. #ifdef CONFIG_SPARSEMEM
  3765. return __pfn_to_section(pfn)->pageblock_flags;
  3766. #else
  3767. return zone->pageblock_flags;
  3768. #endif /* CONFIG_SPARSEMEM */
  3769. }
  3770. static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
  3771. {
  3772. #ifdef CONFIG_SPARSEMEM
  3773. pfn &= (PAGES_PER_SECTION-1);
  3774. return (pfn >> (MAX_ORDER-1)) * NR_PAGEBLOCK_BITS;
  3775. #else
  3776. pfn = pfn - zone->zone_start_pfn;
  3777. return (pfn >> (MAX_ORDER-1)) * NR_PAGEBLOCK_BITS;
  3778. #endif /* CONFIG_SPARSEMEM */
  3779. }
  3780. /**
  3781. * get_pageblock_flags_group - Return the requested group of flags for the MAX_ORDER_NR_PAGES block of pages
  3782. * @page: The page within the block of interest
  3783. * @start_bitidx: The first bit of interest to retrieve
  3784. * @end_bitidx: The last bit of interest
  3785. * returns pageblock_bits flags
  3786. */
  3787. unsigned long get_pageblock_flags_group(struct page *page,
  3788. int start_bitidx, int end_bitidx)
  3789. {
  3790. struct zone *zone;
  3791. unsigned long *bitmap;
  3792. unsigned long pfn, bitidx;
  3793. unsigned long flags = 0;
  3794. unsigned long value = 1;
  3795. zone = page_zone(page);
  3796. pfn = page_to_pfn(page);
  3797. bitmap = get_pageblock_bitmap(zone, pfn);
  3798. bitidx = pfn_to_bitidx(zone, pfn);
  3799. for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
  3800. if (test_bit(bitidx + start_bitidx, bitmap))
  3801. flags |= value;
  3802. return flags;
  3803. }
  3804. /**
  3805. * set_pageblock_flags_group - Set the requested group of flags for a MAX_ORDER_NR_PAGES block of pages
  3806. * @page: The page within the block of interest
  3807. * @start_bitidx: The first bit of interest
  3808. * @end_bitidx: The last bit of interest
  3809. * @flags: The flags to set
  3810. */
  3811. void set_pageblock_flags_group(struct page *page, unsigned long flags,
  3812. int start_bitidx, int end_bitidx)
  3813. {
  3814. struct zone *zone;
  3815. unsigned long *bitmap;
  3816. unsigned long pfn, bitidx;
  3817. unsigned long value = 1;
  3818. zone = page_zone(page);
  3819. pfn = page_to_pfn(page);
  3820. bitmap = get_pageblock_bitmap(zone, pfn);
  3821. bitidx = pfn_to_bitidx(zone, pfn);
  3822. for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
  3823. if (flags & value)
  3824. __set_bit(bitidx + start_bitidx, bitmap);
  3825. else
  3826. __clear_bit(bitidx + start_bitidx, bitmap);
  3827. }