page_alloc.c 129 KB

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385138613871388138913901391139213931394139513961397139813991400140114021403140414051406140714081409141014111412141314141415141614171418141914201421142214231424142514261427142814291430143114321433143414351436143714381439144014411442144314441445144614471448144914501451145214531454145514561457145814591460146114621463146414651466146714681469147014711472147314741475147614771478147914801481148214831484148514861487148814891490149114921493149414951496149714981499150015011502150315041505150615071508150915101511151215131514151515161517151815191520152115221523152415251526152715281529153015311532153315341535153615371538153915401541154215431544154515461547154815491550155115521553155415551556155715581559156015611562156315641565156615671568156915701571157215731574157515761577157815791580158115821583158415851586158715881589159015911592159315941595159615971598159916001601160216031604160516061607160816091610161116121613161416151616161716181619162016211622162316241625162616271628162916301631163216331634163516361637163816391640164116421643164416451646164716481649165016511652165316541655165616571658165916601661166216631664166516661667166816691670167116721673167416751676167716781679168016811682168316841685168616871688168916901691169216931694169516961697169816991700170117021703170417051706170717081709171017111712171317141715171617171718171917201721172217231724172517261727172817291730173117321733173417351736173717381739174017411742174317441745174617471748174917501751175217531754175517561757175817591760176117621763176417651766176717681769177017711772177317741775177617771778177917801781178217831784178517861787178817891790179117921793179417951796179717981799180018011802180318041805180618071808180918101811181218131814181518161817181818191820182118221823182418251826182718281829183018311832183318341835183618371838183918401841184218431844184518461847184818491850185118521853185418551856185718581859186018611862186318641865186618671868186918701871187218731874187518761877187818791880188118821883188418851886188718881889189018911892189318941895189618971898189919001901190219031904190519061907190819091910191119121913191419151916191719181919192019211922192319241925192619271928192919301931193219331934193519361937193819391940194119421943194419451946194719481949195019511952195319541955195619571958195919601961196219631964196519661967196819691970197119721973197419751976197719781979198019811982198319841985198619871988198919901991199219931994199519961997199819992000200120022003200420052006200720082009201020112012201320142015201620172018201920202021202220232024202520262027202820292030203120322033203420352036203720382039204020412042204320442045204620472048204920502051205220532054205520562057205820592060206120622063206420652066206720682069207020712072207320742075207620772078207920802081208220832084208520862087208820892090209120922093209420952096209720982099210021012102210321042105210621072108210921102111211221132114211521162117211821192120212121222123212421252126212721282129213021312132213321342135213621372138213921402141214221432144214521462147214821492150215121522153215421552156215721582159216021612162216321642165216621672168216921702171217221732174217521762177217821792180218121822183218421852186218721882189219021912192219321942195219621972198219922002201220222032204220522062207220822092210221122122213221422152216221722182219222022212222222322242225222622272228222922302231223222332234223522362237223822392240224122422243224422452246224722482249225022512252225322542255225622572258225922602261226222632264226522662267226822692270227122722273227422752276227722782279228022812282228322842285228622872288228922902291229222932294229522962297229822992300230123022303230423052306230723082309231023112312231323142315231623172318231923202321232223232324232523262327232823292330233123322333233423352336233723382339234023412342234323442345234623472348234923502351235223532354235523562357235823592360236123622363236423652366236723682369237023712372237323742375237623772378237923802381238223832384238523862387238823892390239123922393239423952396239723982399240024012402240324042405240624072408240924102411241224132414241524162417241824192420242124222423242424252426242724282429243024312432243324342435243624372438243924402441244224432444244524462447244824492450245124522453245424552456245724582459246024612462246324642465246624672468246924702471247224732474247524762477247824792480248124822483248424852486248724882489249024912492249324942495249624972498249925002501250225032504250525062507250825092510251125122513251425152516251725182519252025212522252325242525252625272528252925302531253225332534253525362537253825392540254125422543254425452546254725482549255025512552255325542555255625572558255925602561256225632564256525662567256825692570257125722573257425752576257725782579258025812582258325842585258625872588258925902591259225932594259525962597259825992600260126022603260426052606260726082609261026112612261326142615261626172618261926202621262226232624262526262627262826292630263126322633263426352636263726382639264026412642264326442645264626472648264926502651265226532654265526562657265826592660266126622663266426652666266726682669267026712672267326742675267626772678267926802681268226832684268526862687268826892690269126922693269426952696269726982699270027012702270327042705270627072708270927102711271227132714271527162717271827192720272127222723272427252726272727282729273027312732273327342735273627372738273927402741274227432744274527462747274827492750275127522753275427552756275727582759276027612762276327642765276627672768276927702771277227732774277527762777277827792780278127822783278427852786278727882789279027912792279327942795279627972798279928002801280228032804280528062807280828092810281128122813281428152816281728182819282028212822282328242825282628272828282928302831283228332834283528362837283828392840284128422843284428452846284728482849285028512852285328542855285628572858285928602861286228632864286528662867286828692870287128722873287428752876287728782879288028812882288328842885288628872888288928902891289228932894289528962897289828992900290129022903290429052906290729082909291029112912291329142915291629172918291929202921292229232924292529262927292829292930293129322933293429352936293729382939294029412942294329442945294629472948294929502951295229532954295529562957295829592960296129622963296429652966296729682969297029712972297329742975297629772978297929802981298229832984298529862987298829892990299129922993299429952996299729982999300030013002300330043005300630073008300930103011301230133014301530163017301830193020302130223023302430253026302730283029303030313032303330343035303630373038303930403041304230433044304530463047304830493050305130523053305430553056305730583059306030613062306330643065306630673068306930703071307230733074307530763077307830793080308130823083308430853086308730883089309030913092309330943095309630973098309931003101310231033104310531063107310831093110311131123113311431153116311731183119312031213122312331243125312631273128312931303131313231333134313531363137313831393140314131423143314431453146314731483149315031513152315331543155315631573158315931603161316231633164316531663167316831693170317131723173317431753176317731783179318031813182318331843185318631873188318931903191319231933194319531963197319831993200320132023203320432053206320732083209321032113212321332143215321632173218321932203221322232233224322532263227322832293230323132323233323432353236323732383239324032413242324332443245324632473248324932503251325232533254325532563257325832593260326132623263326432653266326732683269327032713272327332743275327632773278327932803281328232833284328532863287328832893290329132923293329432953296329732983299330033013302330333043305330633073308330933103311331233133314331533163317331833193320332133223323332433253326332733283329333033313332333333343335333633373338333933403341334233433344334533463347334833493350335133523353335433553356335733583359336033613362336333643365336633673368336933703371337233733374337533763377337833793380338133823383338433853386338733883389339033913392339333943395339633973398339934003401340234033404340534063407340834093410341134123413341434153416341734183419342034213422342334243425342634273428342934303431343234333434343534363437343834393440344134423443344434453446344734483449345034513452345334543455345634573458345934603461346234633464346534663467346834693470347134723473347434753476347734783479348034813482348334843485348634873488348934903491349234933494349534963497349834993500350135023503350435053506350735083509351035113512351335143515351635173518351935203521352235233524352535263527352835293530353135323533353435353536353735383539354035413542354335443545354635473548354935503551355235533554355535563557355835593560356135623563356435653566356735683569357035713572357335743575357635773578357935803581358235833584358535863587358835893590359135923593359435953596359735983599360036013602360336043605360636073608360936103611361236133614361536163617361836193620362136223623362436253626362736283629363036313632363336343635363636373638363936403641364236433644364536463647364836493650365136523653365436553656365736583659366036613662366336643665366636673668366936703671367236733674367536763677367836793680368136823683368436853686368736883689369036913692369336943695369636973698369937003701370237033704370537063707370837093710371137123713371437153716371737183719372037213722372337243725372637273728372937303731373237333734373537363737373837393740374137423743374437453746374737483749375037513752375337543755375637573758375937603761376237633764376537663767376837693770377137723773377437753776377737783779378037813782378337843785378637873788378937903791379237933794379537963797379837993800380138023803380438053806380738083809381038113812381338143815381638173818381938203821382238233824382538263827382838293830383138323833383438353836383738383839384038413842384338443845384638473848384938503851385238533854385538563857385838593860386138623863386438653866386738683869387038713872387338743875387638773878387938803881388238833884388538863887388838893890389138923893389438953896389738983899390039013902390339043905390639073908390939103911391239133914391539163917391839193920392139223923392439253926392739283929393039313932393339343935393639373938393939403941394239433944394539463947394839493950395139523953395439553956395739583959396039613962396339643965396639673968396939703971397239733974397539763977397839793980398139823983398439853986398739883989399039913992399339943995399639973998399940004001400240034004400540064007400840094010401140124013401440154016401740184019402040214022402340244025402640274028402940304031403240334034403540364037403840394040404140424043404440454046404740484049405040514052405340544055405640574058405940604061406240634064406540664067406840694070407140724073407440754076407740784079408040814082408340844085408640874088408940904091409240934094409540964097409840994100410141024103410441054106410741084109411041114112411341144115411641174118411941204121412241234124412541264127412841294130413141324133413441354136413741384139414041414142414341444145414641474148414941504151415241534154415541564157415841594160416141624163416441654166416741684169417041714172417341744175417641774178417941804181418241834184418541864187418841894190419141924193419441954196419741984199420042014202420342044205420642074208420942104211421242134214421542164217421842194220422142224223422442254226422742284229423042314232423342344235423642374238423942404241424242434244424542464247424842494250425142524253425442554256425742584259426042614262426342644265426642674268426942704271427242734274427542764277427842794280428142824283428442854286428742884289429042914292429342944295429642974298429943004301430243034304430543064307430843094310431143124313431443154316431743184319432043214322432343244325432643274328432943304331433243334334433543364337433843394340434143424343434443454346434743484349435043514352435343544355435643574358435943604361436243634364436543664367436843694370437143724373437443754376437743784379438043814382438343844385438643874388438943904391439243934394439543964397439843994400440144024403440444054406440744084409441044114412441344144415441644174418441944204421442244234424442544264427442844294430443144324433443444354436443744384439444044414442444344444445444644474448444944504451445244534454445544564457445844594460446144624463446444654466446744684469447044714472447344744475447644774478447944804481448244834484448544864487448844894490449144924493449444954496449744984499450045014502450345044505450645074508450945104511451245134514451545164517451845194520452145224523452445254526452745284529453045314532453345344535453645374538453945404541454245434544454545464547454845494550455145524553455445554556455745584559456045614562456345644565456645674568456945704571457245734574457545764577457845794580458145824583458445854586458745884589459045914592459345944595459645974598459946004601460246034604460546064607460846094610461146124613461446154616461746184619462046214622462346244625462646274628462946304631463246334634463546364637463846394640464146424643464446454646464746484649465046514652465346544655465646574658465946604661466246634664466546664667466846694670467146724673467446754676467746784679468046814682468346844685
  1. /*
  2. * linux/mm/page_alloc.c
  3. *
  4. * Manages the free list, the system allocates free pages here.
  5. * Note that kmalloc() lives in slab.c
  6. *
  7. * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
  8. * Swap reorganised 29.12.95, Stephen Tweedie
  9. * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
  10. * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
  11. * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
  12. * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
  13. * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
  14. * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
  15. */
  16. #include <linux/stddef.h>
  17. #include <linux/mm.h>
  18. #include <linux/swap.h>
  19. #include <linux/interrupt.h>
  20. #include <linux/pagemap.h>
  21. #include <linux/jiffies.h>
  22. #include <linux/bootmem.h>
  23. #include <linux/compiler.h>
  24. #include <linux/kernel.h>
  25. #include <linux/module.h>
  26. #include <linux/suspend.h>
  27. #include <linux/pagevec.h>
  28. #include <linux/blkdev.h>
  29. #include <linux/slab.h>
  30. #include <linux/oom.h>
  31. #include <linux/notifier.h>
  32. #include <linux/topology.h>
  33. #include <linux/sysctl.h>
  34. #include <linux/cpu.h>
  35. #include <linux/cpuset.h>
  36. #include <linux/memory_hotplug.h>
  37. #include <linux/nodemask.h>
  38. #include <linux/vmalloc.h>
  39. #include <linux/mempolicy.h>
  40. #include <linux/stop_machine.h>
  41. #include <linux/sort.h>
  42. #include <linux/pfn.h>
  43. #include <linux/backing-dev.h>
  44. #include <linux/fault-inject.h>
  45. #include <linux/page-isolation.h>
  46. #include <linux/memcontrol.h>
  47. #include <linux/debugobjects.h>
  48. #include <asm/tlbflush.h>
  49. #include <asm/div64.h>
  50. #include "internal.h"
  51. /*
  52. * Array of node states.
  53. */
  54. nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
  55. [N_POSSIBLE] = NODE_MASK_ALL,
  56. [N_ONLINE] = { { [0] = 1UL } },
  57. #ifndef CONFIG_NUMA
  58. [N_NORMAL_MEMORY] = { { [0] = 1UL } },
  59. #ifdef CONFIG_HIGHMEM
  60. [N_HIGH_MEMORY] = { { [0] = 1UL } },
  61. #endif
  62. [N_CPU] = { { [0] = 1UL } },
  63. #endif /* NUMA */
  64. };
  65. EXPORT_SYMBOL(node_states);
  66. unsigned long totalram_pages __read_mostly;
  67. unsigned long totalreserve_pages __read_mostly;
  68. long nr_swap_pages;
  69. int percpu_pagelist_fraction;
  70. #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
  71. int pageblock_order __read_mostly;
  72. #endif
  73. static void __free_pages_ok(struct page *page, unsigned int order);
  74. /*
  75. * results with 256, 32 in the lowmem_reserve sysctl:
  76. * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
  77. * 1G machine -> (16M dma, 784M normal, 224M high)
  78. * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
  79. * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
  80. * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
  81. *
  82. * TBD: should special case ZONE_DMA32 machines here - in those we normally
  83. * don't need any ZONE_NORMAL reservation
  84. */
  85. int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
  86. #ifdef CONFIG_ZONE_DMA
  87. 256,
  88. #endif
  89. #ifdef CONFIG_ZONE_DMA32
  90. 256,
  91. #endif
  92. #ifdef CONFIG_HIGHMEM
  93. 32,
  94. #endif
  95. 32,
  96. };
  97. EXPORT_SYMBOL(totalram_pages);
  98. static char * const zone_names[MAX_NR_ZONES] = {
  99. #ifdef CONFIG_ZONE_DMA
  100. "DMA",
  101. #endif
  102. #ifdef CONFIG_ZONE_DMA32
  103. "DMA32",
  104. #endif
  105. "Normal",
  106. #ifdef CONFIG_HIGHMEM
  107. "HighMem",
  108. #endif
  109. "Movable",
  110. };
  111. int min_free_kbytes = 1024;
  112. unsigned long __meminitdata nr_kernel_pages;
  113. unsigned long __meminitdata nr_all_pages;
  114. static unsigned long __meminitdata dma_reserve;
  115. #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
  116. /*
  117. * MAX_ACTIVE_REGIONS determines the maximum number of distinct
  118. * ranges of memory (RAM) that may be registered with add_active_range().
  119. * Ranges passed to add_active_range() will be merged if possible
  120. * so the number of times add_active_range() can be called is
  121. * related to the number of nodes and the number of holes
  122. */
  123. #ifdef CONFIG_MAX_ACTIVE_REGIONS
  124. /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
  125. #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
  126. #else
  127. #if MAX_NUMNODES >= 32
  128. /* If there can be many nodes, allow up to 50 holes per node */
  129. #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
  130. #else
  131. /* By default, allow up to 256 distinct regions */
  132. #define MAX_ACTIVE_REGIONS 256
  133. #endif
  134. #endif
  135. static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS];
  136. static int __meminitdata nr_nodemap_entries;
  137. static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
  138. static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
  139. #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
  140. static unsigned long __meminitdata node_boundary_start_pfn[MAX_NUMNODES];
  141. static unsigned long __meminitdata node_boundary_end_pfn[MAX_NUMNODES];
  142. #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
  143. static unsigned long __initdata required_kernelcore;
  144. static unsigned long __initdata required_movablecore;
  145. static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
  146. /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
  147. int movable_zone;
  148. EXPORT_SYMBOL(movable_zone);
  149. #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
  150. #if MAX_NUMNODES > 1
  151. int nr_node_ids __read_mostly = MAX_NUMNODES;
  152. EXPORT_SYMBOL(nr_node_ids);
  153. #endif
  154. int page_group_by_mobility_disabled __read_mostly;
  155. static void set_pageblock_migratetype(struct page *page, int migratetype)
  156. {
  157. set_pageblock_flags_group(page, (unsigned long)migratetype,
  158. PB_migrate, PB_migrate_end);
  159. }
  160. #ifdef CONFIG_DEBUG_VM
  161. static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
  162. {
  163. int ret = 0;
  164. unsigned seq;
  165. unsigned long pfn = page_to_pfn(page);
  166. do {
  167. seq = zone_span_seqbegin(zone);
  168. if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
  169. ret = 1;
  170. else if (pfn < zone->zone_start_pfn)
  171. ret = 1;
  172. } while (zone_span_seqretry(zone, seq));
  173. return ret;
  174. }
  175. static int page_is_consistent(struct zone *zone, struct page *page)
  176. {
  177. if (!pfn_valid_within(page_to_pfn(page)))
  178. return 0;
  179. if (zone != page_zone(page))
  180. return 0;
  181. return 1;
  182. }
  183. /*
  184. * Temporary debugging check for pages not lying within a given zone.
  185. */
  186. static int bad_range(struct zone *zone, struct page *page)
  187. {
  188. if (page_outside_zone_boundaries(zone, page))
  189. return 1;
  190. if (!page_is_consistent(zone, page))
  191. return 1;
  192. return 0;
  193. }
  194. #else
  195. static inline int bad_range(struct zone *zone, struct page *page)
  196. {
  197. return 0;
  198. }
  199. #endif
  200. static void bad_page(struct page *page)
  201. {
  202. void *pc = page_get_page_cgroup(page);
  203. printk(KERN_EMERG "Bad page state in process '%s'\n" KERN_EMERG
  204. "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n",
  205. current->comm, page, (int)(2*sizeof(unsigned long)),
  206. (unsigned long)page->flags, page->mapping,
  207. page_mapcount(page), page_count(page));
  208. if (pc) {
  209. printk(KERN_EMERG "cgroup:%p\n", pc);
  210. page_reset_bad_cgroup(page);
  211. }
  212. printk(KERN_EMERG "Trying to fix it up, but a reboot is needed\n"
  213. KERN_EMERG "Backtrace:\n");
  214. dump_stack();
  215. page->flags &= ~PAGE_FLAGS_CLEAR_WHEN_BAD;
  216. set_page_count(page, 0);
  217. reset_page_mapcount(page);
  218. page->mapping = NULL;
  219. add_taint(TAINT_BAD_PAGE);
  220. }
  221. /*
  222. * Higher-order pages are called "compound pages". They are structured thusly:
  223. *
  224. * The first PAGE_SIZE page is called the "head page".
  225. *
  226. * The remaining PAGE_SIZE pages are called "tail pages".
  227. *
  228. * All pages have PG_compound set. All pages have their ->private pointing at
  229. * the head page (even the head page has this).
  230. *
  231. * The first tail page's ->lru.next holds the address of the compound page's
  232. * put_page() function. Its ->lru.prev holds the order of allocation.
  233. * This usage means that zero-order pages may not be compound.
  234. */
  235. static void free_compound_page(struct page *page)
  236. {
  237. __free_pages_ok(page, compound_order(page));
  238. }
  239. void prep_compound_page(struct page *page, unsigned long order)
  240. {
  241. int i;
  242. int nr_pages = 1 << order;
  243. struct page *p = page + 1;
  244. set_compound_page_dtor(page, free_compound_page);
  245. set_compound_order(page, order);
  246. __SetPageHead(page);
  247. for (i = 1; i < nr_pages; i++, p++) {
  248. if (unlikely((i & (MAX_ORDER_NR_PAGES - 1)) == 0))
  249. p = pfn_to_page(page_to_pfn(page) + i);
  250. __SetPageTail(p);
  251. p->first_page = page;
  252. }
  253. }
  254. static void destroy_compound_page(struct page *page, unsigned long order)
  255. {
  256. int i;
  257. int nr_pages = 1 << order;
  258. struct page *p = page + 1;
  259. if (unlikely(compound_order(page) != order))
  260. bad_page(page);
  261. if (unlikely(!PageHead(page)))
  262. bad_page(page);
  263. __ClearPageHead(page);
  264. for (i = 1; i < nr_pages; i++, p++) {
  265. if (unlikely((i & (MAX_ORDER_NR_PAGES - 1)) == 0))
  266. p = pfn_to_page(page_to_pfn(page) + i);
  267. if (unlikely(!PageTail(p) |
  268. (p->first_page != page)))
  269. bad_page(page);
  270. __ClearPageTail(p);
  271. }
  272. }
  273. static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
  274. {
  275. int i;
  276. /*
  277. * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
  278. * and __GFP_HIGHMEM from hard or soft interrupt context.
  279. */
  280. VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
  281. for (i = 0; i < (1 << order); i++)
  282. clear_highpage(page + i);
  283. }
  284. static inline void set_page_order(struct page *page, int order)
  285. {
  286. set_page_private(page, order);
  287. __SetPageBuddy(page);
  288. }
  289. static inline void rmv_page_order(struct page *page)
  290. {
  291. __ClearPageBuddy(page);
  292. set_page_private(page, 0);
  293. }
  294. /*
  295. * Locate the struct page for both the matching buddy in our
  296. * pair (buddy1) and the combined O(n+1) page they form (page).
  297. *
  298. * 1) Any buddy B1 will have an order O twin B2 which satisfies
  299. * the following equation:
  300. * B2 = B1 ^ (1 << O)
  301. * For example, if the starting buddy (buddy2) is #8 its order
  302. * 1 buddy is #10:
  303. * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
  304. *
  305. * 2) Any buddy B will have an order O+1 parent P which
  306. * satisfies the following equation:
  307. * P = B & ~(1 << O)
  308. *
  309. * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
  310. */
  311. static inline struct page *
  312. __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
  313. {
  314. unsigned long buddy_idx = page_idx ^ (1 << order);
  315. return page + (buddy_idx - page_idx);
  316. }
  317. static inline unsigned long
  318. __find_combined_index(unsigned long page_idx, unsigned int order)
  319. {
  320. return (page_idx & ~(1 << order));
  321. }
  322. /*
  323. * This function checks whether a page is free && is the buddy
  324. * we can do coalesce a page and its buddy if
  325. * (a) the buddy is not in a hole &&
  326. * (b) the buddy is in the buddy system &&
  327. * (c) a page and its buddy have the same order &&
  328. * (d) a page and its buddy are in the same zone.
  329. *
  330. * For recording whether a page is in the buddy system, we use PG_buddy.
  331. * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
  332. *
  333. * For recording page's order, we use page_private(page).
  334. */
  335. static inline int page_is_buddy(struct page *page, struct page *buddy,
  336. int order)
  337. {
  338. if (!pfn_valid_within(page_to_pfn(buddy)))
  339. return 0;
  340. if (page_zone_id(page) != page_zone_id(buddy))
  341. return 0;
  342. if (PageBuddy(buddy) && page_order(buddy) == order) {
  343. BUG_ON(page_count(buddy) != 0);
  344. return 1;
  345. }
  346. return 0;
  347. }
  348. /*
  349. * Freeing function for a buddy system allocator.
  350. *
  351. * The concept of a buddy system is to maintain direct-mapped table
  352. * (containing bit values) for memory blocks of various "orders".
  353. * The bottom level table contains the map for the smallest allocatable
  354. * units of memory (here, pages), and each level above it describes
  355. * pairs of units from the levels below, hence, "buddies".
  356. * At a high level, all that happens here is marking the table entry
  357. * at the bottom level available, and propagating the changes upward
  358. * as necessary, plus some accounting needed to play nicely with other
  359. * parts of the VM system.
  360. * At each level, we keep a list of pages, which are heads of continuous
  361. * free pages of length of (1 << order) and marked with PG_buddy. Page's
  362. * order is recorded in page_private(page) field.
  363. * So when we are allocating or freeing one, we can derive the state of the
  364. * other. That is, if we allocate a small block, and both were
  365. * free, the remainder of the region must be split into blocks.
  366. * If a block is freed, and its buddy is also free, then this
  367. * triggers coalescing into a block of larger size.
  368. *
  369. * -- wli
  370. */
  371. static inline void __free_one_page(struct page *page,
  372. struct zone *zone, unsigned int order)
  373. {
  374. unsigned long page_idx;
  375. int order_size = 1 << order;
  376. int migratetype = get_pageblock_migratetype(page);
  377. if (unlikely(PageCompound(page)))
  378. destroy_compound_page(page, order);
  379. page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
  380. VM_BUG_ON(page_idx & (order_size - 1));
  381. VM_BUG_ON(bad_range(zone, page));
  382. __mod_zone_page_state(zone, NR_FREE_PAGES, order_size);
  383. while (order < MAX_ORDER-1) {
  384. unsigned long combined_idx;
  385. struct page *buddy;
  386. buddy = __page_find_buddy(page, page_idx, order);
  387. if (!page_is_buddy(page, buddy, order))
  388. break;
  389. /* Our buddy is free, merge with it and move up one order. */
  390. list_del(&buddy->lru);
  391. zone->free_area[order].nr_free--;
  392. rmv_page_order(buddy);
  393. combined_idx = __find_combined_index(page_idx, order);
  394. page = page + (combined_idx - page_idx);
  395. page_idx = combined_idx;
  396. order++;
  397. }
  398. set_page_order(page, order);
  399. list_add(&page->lru,
  400. &zone->free_area[order].free_list[migratetype]);
  401. zone->free_area[order].nr_free++;
  402. }
  403. static inline int free_pages_check(struct page *page)
  404. {
  405. if (unlikely(page_mapcount(page) |
  406. (page->mapping != NULL) |
  407. (page_get_page_cgroup(page) != NULL) |
  408. (page_count(page) != 0) |
  409. (page->flags & PAGE_FLAGS_CHECK_AT_FREE)))
  410. bad_page(page);
  411. if (PageDirty(page))
  412. __ClearPageDirty(page);
  413. if (PageSwapBacked(page))
  414. __ClearPageSwapBacked(page);
  415. /*
  416. * For now, we report if PG_reserved was found set, but do not
  417. * clear it, and do not free the page. But we shall soon need
  418. * to do more, for when the ZERO_PAGE count wraps negative.
  419. */
  420. return PageReserved(page);
  421. }
  422. /*
  423. * Frees a list of pages.
  424. * Assumes all pages on list are in same zone, and of same order.
  425. * count is the number of pages to free.
  426. *
  427. * If the zone was previously in an "all pages pinned" state then look to
  428. * see if this freeing clears that state.
  429. *
  430. * And clear the zone's pages_scanned counter, to hold off the "all pages are
  431. * pinned" detection logic.
  432. */
  433. static void free_pages_bulk(struct zone *zone, int count,
  434. struct list_head *list, int order)
  435. {
  436. spin_lock(&zone->lock);
  437. zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
  438. zone->pages_scanned = 0;
  439. while (count--) {
  440. struct page *page;
  441. VM_BUG_ON(list_empty(list));
  442. page = list_entry(list->prev, struct page, lru);
  443. /* have to delete it as __free_one_page list manipulates */
  444. list_del(&page->lru);
  445. __free_one_page(page, zone, order);
  446. }
  447. spin_unlock(&zone->lock);
  448. }
  449. static void free_one_page(struct zone *zone, struct page *page, int order)
  450. {
  451. spin_lock(&zone->lock);
  452. zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
  453. zone->pages_scanned = 0;
  454. __free_one_page(page, zone, order);
  455. spin_unlock(&zone->lock);
  456. }
  457. static void __free_pages_ok(struct page *page, unsigned int order)
  458. {
  459. unsigned long flags;
  460. int i;
  461. int reserved = 0;
  462. for (i = 0 ; i < (1 << order) ; ++i)
  463. reserved += free_pages_check(page + i);
  464. if (reserved)
  465. return;
  466. if (!PageHighMem(page)) {
  467. debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
  468. debug_check_no_obj_freed(page_address(page),
  469. PAGE_SIZE << order);
  470. }
  471. arch_free_page(page, order);
  472. kernel_map_pages(page, 1 << order, 0);
  473. local_irq_save(flags);
  474. __count_vm_events(PGFREE, 1 << order);
  475. free_one_page(page_zone(page), page, order);
  476. local_irq_restore(flags);
  477. }
  478. /*
  479. * permit the bootmem allocator to evade page validation on high-order frees
  480. */
  481. void __meminit __free_pages_bootmem(struct page *page, unsigned int order)
  482. {
  483. if (order == 0) {
  484. __ClearPageReserved(page);
  485. set_page_count(page, 0);
  486. set_page_refcounted(page);
  487. __free_page(page);
  488. } else {
  489. int loop;
  490. prefetchw(page);
  491. for (loop = 0; loop < BITS_PER_LONG; loop++) {
  492. struct page *p = &page[loop];
  493. if (loop + 1 < BITS_PER_LONG)
  494. prefetchw(p + 1);
  495. __ClearPageReserved(p);
  496. set_page_count(p, 0);
  497. }
  498. set_page_refcounted(page);
  499. __free_pages(page, order);
  500. }
  501. }
  502. /*
  503. * The order of subdivision here is critical for the IO subsystem.
  504. * Please do not alter this order without good reasons and regression
  505. * testing. Specifically, as large blocks of memory are subdivided,
  506. * the order in which smaller blocks are delivered depends on the order
  507. * they're subdivided in this function. This is the primary factor
  508. * influencing the order in which pages are delivered to the IO
  509. * subsystem according to empirical testing, and this is also justified
  510. * by considering the behavior of a buddy system containing a single
  511. * large block of memory acted on by a series of small allocations.
  512. * This behavior is a critical factor in sglist merging's success.
  513. *
  514. * -- wli
  515. */
  516. static inline void expand(struct zone *zone, struct page *page,
  517. int low, int high, struct free_area *area,
  518. int migratetype)
  519. {
  520. unsigned long size = 1 << high;
  521. while (high > low) {
  522. area--;
  523. high--;
  524. size >>= 1;
  525. VM_BUG_ON(bad_range(zone, &page[size]));
  526. list_add(&page[size].lru, &area->free_list[migratetype]);
  527. area->nr_free++;
  528. set_page_order(&page[size], high);
  529. }
  530. }
  531. /*
  532. * This page is about to be returned from the page allocator
  533. */
  534. static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
  535. {
  536. if (unlikely(page_mapcount(page) |
  537. (page->mapping != NULL) |
  538. (page_get_page_cgroup(page) != NULL) |
  539. (page_count(page) != 0) |
  540. (page->flags & PAGE_FLAGS_CHECK_AT_PREP)))
  541. bad_page(page);
  542. /*
  543. * For now, we report if PG_reserved was found set, but do not
  544. * clear it, and do not allocate the page: as a safety net.
  545. */
  546. if (PageReserved(page))
  547. return 1;
  548. page->flags &= ~(1 << PG_uptodate | 1 << PG_error | 1 << PG_reclaim |
  549. 1 << PG_referenced | 1 << PG_arch_1 |
  550. 1 << PG_owner_priv_1 | 1 << PG_mappedtodisk);
  551. set_page_private(page, 0);
  552. set_page_refcounted(page);
  553. arch_alloc_page(page, order);
  554. kernel_map_pages(page, 1 << order, 1);
  555. if (gfp_flags & __GFP_ZERO)
  556. prep_zero_page(page, order, gfp_flags);
  557. if (order && (gfp_flags & __GFP_COMP))
  558. prep_compound_page(page, order);
  559. return 0;
  560. }
  561. /*
  562. * Go through the free lists for the given migratetype and remove
  563. * the smallest available page from the freelists
  564. */
  565. static struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
  566. int migratetype)
  567. {
  568. unsigned int current_order;
  569. struct free_area * area;
  570. struct page *page;
  571. /* Find a page of the appropriate size in the preferred list */
  572. for (current_order = order; current_order < MAX_ORDER; ++current_order) {
  573. area = &(zone->free_area[current_order]);
  574. if (list_empty(&area->free_list[migratetype]))
  575. continue;
  576. page = list_entry(area->free_list[migratetype].next,
  577. struct page, lru);
  578. list_del(&page->lru);
  579. rmv_page_order(page);
  580. area->nr_free--;
  581. __mod_zone_page_state(zone, NR_FREE_PAGES, - (1UL << order));
  582. expand(zone, page, order, current_order, area, migratetype);
  583. return page;
  584. }
  585. return NULL;
  586. }
  587. /*
  588. * This array describes the order lists are fallen back to when
  589. * the free lists for the desirable migrate type are depleted
  590. */
  591. static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
  592. [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
  593. [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
  594. [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
  595. [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
  596. };
  597. /*
  598. * Move the free pages in a range to the free lists of the requested type.
  599. * Note that start_page and end_pages are not aligned on a pageblock
  600. * boundary. If alignment is required, use move_freepages_block()
  601. */
  602. static int move_freepages(struct zone *zone,
  603. struct page *start_page, struct page *end_page,
  604. int migratetype)
  605. {
  606. struct page *page;
  607. unsigned long order;
  608. int pages_moved = 0;
  609. #ifndef CONFIG_HOLES_IN_ZONE
  610. /*
  611. * page_zone is not safe to call in this context when
  612. * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
  613. * anyway as we check zone boundaries in move_freepages_block().
  614. * Remove at a later date when no bug reports exist related to
  615. * grouping pages by mobility
  616. */
  617. BUG_ON(page_zone(start_page) != page_zone(end_page));
  618. #endif
  619. for (page = start_page; page <= end_page;) {
  620. /* Make sure we are not inadvertently changing nodes */
  621. VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone));
  622. if (!pfn_valid_within(page_to_pfn(page))) {
  623. page++;
  624. continue;
  625. }
  626. if (!PageBuddy(page)) {
  627. page++;
  628. continue;
  629. }
  630. order = page_order(page);
  631. list_del(&page->lru);
  632. list_add(&page->lru,
  633. &zone->free_area[order].free_list[migratetype]);
  634. page += 1 << order;
  635. pages_moved += 1 << order;
  636. }
  637. return pages_moved;
  638. }
  639. static int move_freepages_block(struct zone *zone, struct page *page,
  640. int migratetype)
  641. {
  642. unsigned long start_pfn, end_pfn;
  643. struct page *start_page, *end_page;
  644. start_pfn = page_to_pfn(page);
  645. start_pfn = start_pfn & ~(pageblock_nr_pages-1);
  646. start_page = pfn_to_page(start_pfn);
  647. end_page = start_page + pageblock_nr_pages - 1;
  648. end_pfn = start_pfn + pageblock_nr_pages - 1;
  649. /* Do not cross zone boundaries */
  650. if (start_pfn < zone->zone_start_pfn)
  651. start_page = page;
  652. if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
  653. return 0;
  654. return move_freepages(zone, start_page, end_page, migratetype);
  655. }
  656. /* Remove an element from the buddy allocator from the fallback list */
  657. static struct page *__rmqueue_fallback(struct zone *zone, int order,
  658. int start_migratetype)
  659. {
  660. struct free_area * area;
  661. int current_order;
  662. struct page *page;
  663. int migratetype, i;
  664. /* Find the largest possible block of pages in the other list */
  665. for (current_order = MAX_ORDER-1; current_order >= order;
  666. --current_order) {
  667. for (i = 0; i < MIGRATE_TYPES - 1; i++) {
  668. migratetype = fallbacks[start_migratetype][i];
  669. /* MIGRATE_RESERVE handled later if necessary */
  670. if (migratetype == MIGRATE_RESERVE)
  671. continue;
  672. area = &(zone->free_area[current_order]);
  673. if (list_empty(&area->free_list[migratetype]))
  674. continue;
  675. page = list_entry(area->free_list[migratetype].next,
  676. struct page, lru);
  677. area->nr_free--;
  678. /*
  679. * If breaking a large block of pages, move all free
  680. * pages to the preferred allocation list. If falling
  681. * back for a reclaimable kernel allocation, be more
  682. * agressive about taking ownership of free pages
  683. */
  684. if (unlikely(current_order >= (pageblock_order >> 1)) ||
  685. start_migratetype == MIGRATE_RECLAIMABLE) {
  686. unsigned long pages;
  687. pages = move_freepages_block(zone, page,
  688. start_migratetype);
  689. /* Claim the whole block if over half of it is free */
  690. if (pages >= (1 << (pageblock_order-1)))
  691. set_pageblock_migratetype(page,
  692. start_migratetype);
  693. migratetype = start_migratetype;
  694. }
  695. /* Remove the page from the freelists */
  696. list_del(&page->lru);
  697. rmv_page_order(page);
  698. __mod_zone_page_state(zone, NR_FREE_PAGES,
  699. -(1UL << order));
  700. if (current_order == pageblock_order)
  701. set_pageblock_migratetype(page,
  702. start_migratetype);
  703. expand(zone, page, order, current_order, area, migratetype);
  704. return page;
  705. }
  706. }
  707. /* Use MIGRATE_RESERVE rather than fail an allocation */
  708. return __rmqueue_smallest(zone, order, MIGRATE_RESERVE);
  709. }
  710. /*
  711. * Do the hard work of removing an element from the buddy allocator.
  712. * Call me with the zone->lock already held.
  713. */
  714. static struct page *__rmqueue(struct zone *zone, unsigned int order,
  715. int migratetype)
  716. {
  717. struct page *page;
  718. page = __rmqueue_smallest(zone, order, migratetype);
  719. if (unlikely(!page))
  720. page = __rmqueue_fallback(zone, order, migratetype);
  721. return page;
  722. }
  723. /*
  724. * Obtain a specified number of elements from the buddy allocator, all under
  725. * a single hold of the lock, for efficiency. Add them to the supplied list.
  726. * Returns the number of new pages which were placed at *list.
  727. */
  728. static int rmqueue_bulk(struct zone *zone, unsigned int order,
  729. unsigned long count, struct list_head *list,
  730. int migratetype)
  731. {
  732. int i;
  733. spin_lock(&zone->lock);
  734. for (i = 0; i < count; ++i) {
  735. struct page *page = __rmqueue(zone, order, migratetype);
  736. if (unlikely(page == NULL))
  737. break;
  738. /*
  739. * Split buddy pages returned by expand() are received here
  740. * in physical page order. The page is added to the callers and
  741. * list and the list head then moves forward. From the callers
  742. * perspective, the linked list is ordered by page number in
  743. * some conditions. This is useful for IO devices that can
  744. * merge IO requests if the physical pages are ordered
  745. * properly.
  746. */
  747. list_add(&page->lru, list);
  748. set_page_private(page, migratetype);
  749. list = &page->lru;
  750. }
  751. spin_unlock(&zone->lock);
  752. return i;
  753. }
  754. #ifdef CONFIG_NUMA
  755. /*
  756. * Called from the vmstat counter updater to drain pagesets of this
  757. * currently executing processor on remote nodes after they have
  758. * expired.
  759. *
  760. * Note that this function must be called with the thread pinned to
  761. * a single processor.
  762. */
  763. void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
  764. {
  765. unsigned long flags;
  766. int to_drain;
  767. local_irq_save(flags);
  768. if (pcp->count >= pcp->batch)
  769. to_drain = pcp->batch;
  770. else
  771. to_drain = pcp->count;
  772. free_pages_bulk(zone, to_drain, &pcp->list, 0);
  773. pcp->count -= to_drain;
  774. local_irq_restore(flags);
  775. }
  776. #endif
  777. /*
  778. * Drain pages of the indicated processor.
  779. *
  780. * The processor must either be the current processor and the
  781. * thread pinned to the current processor or a processor that
  782. * is not online.
  783. */
  784. static void drain_pages(unsigned int cpu)
  785. {
  786. unsigned long flags;
  787. struct zone *zone;
  788. for_each_zone(zone) {
  789. struct per_cpu_pageset *pset;
  790. struct per_cpu_pages *pcp;
  791. if (!populated_zone(zone))
  792. continue;
  793. pset = zone_pcp(zone, cpu);
  794. pcp = &pset->pcp;
  795. local_irq_save(flags);
  796. free_pages_bulk(zone, pcp->count, &pcp->list, 0);
  797. pcp->count = 0;
  798. local_irq_restore(flags);
  799. }
  800. }
  801. /*
  802. * Spill all of this CPU's per-cpu pages back into the buddy allocator.
  803. */
  804. void drain_local_pages(void *arg)
  805. {
  806. drain_pages(smp_processor_id());
  807. }
  808. /*
  809. * Spill all the per-cpu pages from all CPUs back into the buddy allocator
  810. */
  811. void drain_all_pages(void)
  812. {
  813. on_each_cpu(drain_local_pages, NULL, 1);
  814. }
  815. #ifdef CONFIG_HIBERNATION
  816. void mark_free_pages(struct zone *zone)
  817. {
  818. unsigned long pfn, max_zone_pfn;
  819. unsigned long flags;
  820. int order, t;
  821. struct list_head *curr;
  822. if (!zone->spanned_pages)
  823. return;
  824. spin_lock_irqsave(&zone->lock, flags);
  825. max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
  826. for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
  827. if (pfn_valid(pfn)) {
  828. struct page *page = pfn_to_page(pfn);
  829. if (!swsusp_page_is_forbidden(page))
  830. swsusp_unset_page_free(page);
  831. }
  832. for_each_migratetype_order(order, t) {
  833. list_for_each(curr, &zone->free_area[order].free_list[t]) {
  834. unsigned long i;
  835. pfn = page_to_pfn(list_entry(curr, struct page, lru));
  836. for (i = 0; i < (1UL << order); i++)
  837. swsusp_set_page_free(pfn_to_page(pfn + i));
  838. }
  839. }
  840. spin_unlock_irqrestore(&zone->lock, flags);
  841. }
  842. #endif /* CONFIG_PM */
  843. /*
  844. * Free a 0-order page
  845. */
  846. static void free_hot_cold_page(struct page *page, int cold)
  847. {
  848. struct zone *zone = page_zone(page);
  849. struct per_cpu_pages *pcp;
  850. unsigned long flags;
  851. if (PageAnon(page))
  852. page->mapping = NULL;
  853. if (free_pages_check(page))
  854. return;
  855. if (!PageHighMem(page)) {
  856. debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
  857. debug_check_no_obj_freed(page_address(page), PAGE_SIZE);
  858. }
  859. arch_free_page(page, 0);
  860. kernel_map_pages(page, 1, 0);
  861. pcp = &zone_pcp(zone, get_cpu())->pcp;
  862. local_irq_save(flags);
  863. __count_vm_event(PGFREE);
  864. if (cold)
  865. list_add_tail(&page->lru, &pcp->list);
  866. else
  867. list_add(&page->lru, &pcp->list);
  868. set_page_private(page, get_pageblock_migratetype(page));
  869. pcp->count++;
  870. if (pcp->count >= pcp->high) {
  871. free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
  872. pcp->count -= pcp->batch;
  873. }
  874. local_irq_restore(flags);
  875. put_cpu();
  876. }
  877. void free_hot_page(struct page *page)
  878. {
  879. free_hot_cold_page(page, 0);
  880. }
  881. void free_cold_page(struct page *page)
  882. {
  883. free_hot_cold_page(page, 1);
  884. }
  885. /*
  886. * split_page takes a non-compound higher-order page, and splits it into
  887. * n (1<<order) sub-pages: page[0..n]
  888. * Each sub-page must be freed individually.
  889. *
  890. * Note: this is probably too low level an operation for use in drivers.
  891. * Please consult with lkml before using this in your driver.
  892. */
  893. void split_page(struct page *page, unsigned int order)
  894. {
  895. int i;
  896. VM_BUG_ON(PageCompound(page));
  897. VM_BUG_ON(!page_count(page));
  898. for (i = 1; i < (1 << order); i++)
  899. set_page_refcounted(page + i);
  900. }
  901. /*
  902. * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
  903. * we cheat by calling it from here, in the order > 0 path. Saves a branch
  904. * or two.
  905. */
  906. static struct page *buffered_rmqueue(struct zone *preferred_zone,
  907. struct zone *zone, int order, gfp_t gfp_flags)
  908. {
  909. unsigned long flags;
  910. struct page *page;
  911. int cold = !!(gfp_flags & __GFP_COLD);
  912. int cpu;
  913. int migratetype = allocflags_to_migratetype(gfp_flags);
  914. again:
  915. cpu = get_cpu();
  916. if (likely(order == 0)) {
  917. struct per_cpu_pages *pcp;
  918. pcp = &zone_pcp(zone, cpu)->pcp;
  919. local_irq_save(flags);
  920. if (!pcp->count) {
  921. pcp->count = rmqueue_bulk(zone, 0,
  922. pcp->batch, &pcp->list, migratetype);
  923. if (unlikely(!pcp->count))
  924. goto failed;
  925. }
  926. /* Find a page of the appropriate migrate type */
  927. if (cold) {
  928. list_for_each_entry_reverse(page, &pcp->list, lru)
  929. if (page_private(page) == migratetype)
  930. break;
  931. } else {
  932. list_for_each_entry(page, &pcp->list, lru)
  933. if (page_private(page) == migratetype)
  934. break;
  935. }
  936. /* Allocate more to the pcp list if necessary */
  937. if (unlikely(&page->lru == &pcp->list)) {
  938. pcp->count += rmqueue_bulk(zone, 0,
  939. pcp->batch, &pcp->list, migratetype);
  940. page = list_entry(pcp->list.next, struct page, lru);
  941. }
  942. list_del(&page->lru);
  943. pcp->count--;
  944. } else {
  945. spin_lock_irqsave(&zone->lock, flags);
  946. page = __rmqueue(zone, order, migratetype);
  947. spin_unlock(&zone->lock);
  948. if (!page)
  949. goto failed;
  950. }
  951. __count_zone_vm_events(PGALLOC, zone, 1 << order);
  952. zone_statistics(preferred_zone, zone);
  953. local_irq_restore(flags);
  954. put_cpu();
  955. VM_BUG_ON(bad_range(zone, page));
  956. if (prep_new_page(page, order, gfp_flags))
  957. goto again;
  958. return page;
  959. failed:
  960. local_irq_restore(flags);
  961. put_cpu();
  962. return NULL;
  963. }
  964. #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
  965. #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
  966. #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
  967. #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
  968. #define ALLOC_HARDER 0x10 /* try to alloc harder */
  969. #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
  970. #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
  971. #ifdef CONFIG_FAIL_PAGE_ALLOC
  972. static struct fail_page_alloc_attr {
  973. struct fault_attr attr;
  974. u32 ignore_gfp_highmem;
  975. u32 ignore_gfp_wait;
  976. u32 min_order;
  977. #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
  978. struct dentry *ignore_gfp_highmem_file;
  979. struct dentry *ignore_gfp_wait_file;
  980. struct dentry *min_order_file;
  981. #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
  982. } fail_page_alloc = {
  983. .attr = FAULT_ATTR_INITIALIZER,
  984. .ignore_gfp_wait = 1,
  985. .ignore_gfp_highmem = 1,
  986. .min_order = 1,
  987. };
  988. static int __init setup_fail_page_alloc(char *str)
  989. {
  990. return setup_fault_attr(&fail_page_alloc.attr, str);
  991. }
  992. __setup("fail_page_alloc=", setup_fail_page_alloc);
  993. static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
  994. {
  995. if (order < fail_page_alloc.min_order)
  996. return 0;
  997. if (gfp_mask & __GFP_NOFAIL)
  998. return 0;
  999. if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
  1000. return 0;
  1001. if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
  1002. return 0;
  1003. return should_fail(&fail_page_alloc.attr, 1 << order);
  1004. }
  1005. #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
  1006. static int __init fail_page_alloc_debugfs(void)
  1007. {
  1008. mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
  1009. struct dentry *dir;
  1010. int err;
  1011. err = init_fault_attr_dentries(&fail_page_alloc.attr,
  1012. "fail_page_alloc");
  1013. if (err)
  1014. return err;
  1015. dir = fail_page_alloc.attr.dentries.dir;
  1016. fail_page_alloc.ignore_gfp_wait_file =
  1017. debugfs_create_bool("ignore-gfp-wait", mode, dir,
  1018. &fail_page_alloc.ignore_gfp_wait);
  1019. fail_page_alloc.ignore_gfp_highmem_file =
  1020. debugfs_create_bool("ignore-gfp-highmem", mode, dir,
  1021. &fail_page_alloc.ignore_gfp_highmem);
  1022. fail_page_alloc.min_order_file =
  1023. debugfs_create_u32("min-order", mode, dir,
  1024. &fail_page_alloc.min_order);
  1025. if (!fail_page_alloc.ignore_gfp_wait_file ||
  1026. !fail_page_alloc.ignore_gfp_highmem_file ||
  1027. !fail_page_alloc.min_order_file) {
  1028. err = -ENOMEM;
  1029. debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
  1030. debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
  1031. debugfs_remove(fail_page_alloc.min_order_file);
  1032. cleanup_fault_attr_dentries(&fail_page_alloc.attr);
  1033. }
  1034. return err;
  1035. }
  1036. late_initcall(fail_page_alloc_debugfs);
  1037. #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
  1038. #else /* CONFIG_FAIL_PAGE_ALLOC */
  1039. static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
  1040. {
  1041. return 0;
  1042. }
  1043. #endif /* CONFIG_FAIL_PAGE_ALLOC */
  1044. /*
  1045. * Return 1 if free pages are above 'mark'. This takes into account the order
  1046. * of the allocation.
  1047. */
  1048. int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
  1049. int classzone_idx, int alloc_flags)
  1050. {
  1051. /* free_pages my go negative - that's OK */
  1052. long min = mark;
  1053. long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1;
  1054. int o;
  1055. if (alloc_flags & ALLOC_HIGH)
  1056. min -= min / 2;
  1057. if (alloc_flags & ALLOC_HARDER)
  1058. min -= min / 4;
  1059. if (free_pages <= min + z->lowmem_reserve[classzone_idx])
  1060. return 0;
  1061. for (o = 0; o < order; o++) {
  1062. /* At the next order, this order's pages become unavailable */
  1063. free_pages -= z->free_area[o].nr_free << o;
  1064. /* Require fewer higher order pages to be free */
  1065. min >>= 1;
  1066. if (free_pages <= min)
  1067. return 0;
  1068. }
  1069. return 1;
  1070. }
  1071. #ifdef CONFIG_NUMA
  1072. /*
  1073. * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
  1074. * skip over zones that are not allowed by the cpuset, or that have
  1075. * been recently (in last second) found to be nearly full. See further
  1076. * comments in mmzone.h. Reduces cache footprint of zonelist scans
  1077. * that have to skip over a lot of full or unallowed zones.
  1078. *
  1079. * If the zonelist cache is present in the passed in zonelist, then
  1080. * returns a pointer to the allowed node mask (either the current
  1081. * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
  1082. *
  1083. * If the zonelist cache is not available for this zonelist, does
  1084. * nothing and returns NULL.
  1085. *
  1086. * If the fullzones BITMAP in the zonelist cache is stale (more than
  1087. * a second since last zap'd) then we zap it out (clear its bits.)
  1088. *
  1089. * We hold off even calling zlc_setup, until after we've checked the
  1090. * first zone in the zonelist, on the theory that most allocations will
  1091. * be satisfied from that first zone, so best to examine that zone as
  1092. * quickly as we can.
  1093. */
  1094. static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
  1095. {
  1096. struct zonelist_cache *zlc; /* cached zonelist speedup info */
  1097. nodemask_t *allowednodes; /* zonelist_cache approximation */
  1098. zlc = zonelist->zlcache_ptr;
  1099. if (!zlc)
  1100. return NULL;
  1101. if (time_after(jiffies, zlc->last_full_zap + HZ)) {
  1102. bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
  1103. zlc->last_full_zap = jiffies;
  1104. }
  1105. allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
  1106. &cpuset_current_mems_allowed :
  1107. &node_states[N_HIGH_MEMORY];
  1108. return allowednodes;
  1109. }
  1110. /*
  1111. * Given 'z' scanning a zonelist, run a couple of quick checks to see
  1112. * if it is worth looking at further for free memory:
  1113. * 1) Check that the zone isn't thought to be full (doesn't have its
  1114. * bit set in the zonelist_cache fullzones BITMAP).
  1115. * 2) Check that the zones node (obtained from the zonelist_cache
  1116. * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
  1117. * Return true (non-zero) if zone is worth looking at further, or
  1118. * else return false (zero) if it is not.
  1119. *
  1120. * This check -ignores- the distinction between various watermarks,
  1121. * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
  1122. * found to be full for any variation of these watermarks, it will
  1123. * be considered full for up to one second by all requests, unless
  1124. * we are so low on memory on all allowed nodes that we are forced
  1125. * into the second scan of the zonelist.
  1126. *
  1127. * In the second scan we ignore this zonelist cache and exactly
  1128. * apply the watermarks to all zones, even it is slower to do so.
  1129. * We are low on memory in the second scan, and should leave no stone
  1130. * unturned looking for a free page.
  1131. */
  1132. static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
  1133. nodemask_t *allowednodes)
  1134. {
  1135. struct zonelist_cache *zlc; /* cached zonelist speedup info */
  1136. int i; /* index of *z in zonelist zones */
  1137. int n; /* node that zone *z is on */
  1138. zlc = zonelist->zlcache_ptr;
  1139. if (!zlc)
  1140. return 1;
  1141. i = z - zonelist->_zonerefs;
  1142. n = zlc->z_to_n[i];
  1143. /* This zone is worth trying if it is allowed but not full */
  1144. return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
  1145. }
  1146. /*
  1147. * Given 'z' scanning a zonelist, set the corresponding bit in
  1148. * zlc->fullzones, so that subsequent attempts to allocate a page
  1149. * from that zone don't waste time re-examining it.
  1150. */
  1151. static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
  1152. {
  1153. struct zonelist_cache *zlc; /* cached zonelist speedup info */
  1154. int i; /* index of *z in zonelist zones */
  1155. zlc = zonelist->zlcache_ptr;
  1156. if (!zlc)
  1157. return;
  1158. i = z - zonelist->_zonerefs;
  1159. set_bit(i, zlc->fullzones);
  1160. }
  1161. #else /* CONFIG_NUMA */
  1162. static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
  1163. {
  1164. return NULL;
  1165. }
  1166. static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
  1167. nodemask_t *allowednodes)
  1168. {
  1169. return 1;
  1170. }
  1171. static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
  1172. {
  1173. }
  1174. #endif /* CONFIG_NUMA */
  1175. /*
  1176. * get_page_from_freelist goes through the zonelist trying to allocate
  1177. * a page.
  1178. */
  1179. static struct page *
  1180. get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
  1181. struct zonelist *zonelist, int high_zoneidx, int alloc_flags)
  1182. {
  1183. struct zoneref *z;
  1184. struct page *page = NULL;
  1185. int classzone_idx;
  1186. struct zone *zone, *preferred_zone;
  1187. nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
  1188. int zlc_active = 0; /* set if using zonelist_cache */
  1189. int did_zlc_setup = 0; /* just call zlc_setup() one time */
  1190. (void)first_zones_zonelist(zonelist, high_zoneidx, nodemask,
  1191. &preferred_zone);
  1192. if (!preferred_zone)
  1193. return NULL;
  1194. classzone_idx = zone_idx(preferred_zone);
  1195. zonelist_scan:
  1196. /*
  1197. * Scan zonelist, looking for a zone with enough free.
  1198. * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
  1199. */
  1200. for_each_zone_zonelist_nodemask(zone, z, zonelist,
  1201. high_zoneidx, nodemask) {
  1202. if (NUMA_BUILD && zlc_active &&
  1203. !zlc_zone_worth_trying(zonelist, z, allowednodes))
  1204. continue;
  1205. if ((alloc_flags & ALLOC_CPUSET) &&
  1206. !cpuset_zone_allowed_softwall(zone, gfp_mask))
  1207. goto try_next_zone;
  1208. if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
  1209. unsigned long mark;
  1210. if (alloc_flags & ALLOC_WMARK_MIN)
  1211. mark = zone->pages_min;
  1212. else if (alloc_flags & ALLOC_WMARK_LOW)
  1213. mark = zone->pages_low;
  1214. else
  1215. mark = zone->pages_high;
  1216. if (!zone_watermark_ok(zone, order, mark,
  1217. classzone_idx, alloc_flags)) {
  1218. if (!zone_reclaim_mode ||
  1219. !zone_reclaim(zone, gfp_mask, order))
  1220. goto this_zone_full;
  1221. }
  1222. }
  1223. page = buffered_rmqueue(preferred_zone, zone, order, gfp_mask);
  1224. if (page)
  1225. break;
  1226. this_zone_full:
  1227. if (NUMA_BUILD)
  1228. zlc_mark_zone_full(zonelist, z);
  1229. try_next_zone:
  1230. if (NUMA_BUILD && !did_zlc_setup) {
  1231. /* we do zlc_setup after the first zone is tried */
  1232. allowednodes = zlc_setup(zonelist, alloc_flags);
  1233. zlc_active = 1;
  1234. did_zlc_setup = 1;
  1235. }
  1236. }
  1237. if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
  1238. /* Disable zlc cache for second zonelist scan */
  1239. zlc_active = 0;
  1240. goto zonelist_scan;
  1241. }
  1242. return page;
  1243. }
  1244. /*
  1245. * This is the 'heart' of the zoned buddy allocator.
  1246. */
  1247. struct page *
  1248. __alloc_pages_internal(gfp_t gfp_mask, unsigned int order,
  1249. struct zonelist *zonelist, nodemask_t *nodemask)
  1250. {
  1251. const gfp_t wait = gfp_mask & __GFP_WAIT;
  1252. enum zone_type high_zoneidx = gfp_zone(gfp_mask);
  1253. struct zoneref *z;
  1254. struct zone *zone;
  1255. struct page *page;
  1256. struct reclaim_state reclaim_state;
  1257. struct task_struct *p = current;
  1258. int do_retry;
  1259. int alloc_flags;
  1260. unsigned long did_some_progress;
  1261. unsigned long pages_reclaimed = 0;
  1262. might_sleep_if(wait);
  1263. if (should_fail_alloc_page(gfp_mask, order))
  1264. return NULL;
  1265. restart:
  1266. z = zonelist->_zonerefs; /* the list of zones suitable for gfp_mask */
  1267. if (unlikely(!z->zone)) {
  1268. /*
  1269. * Happens if we have an empty zonelist as a result of
  1270. * GFP_THISNODE being used on a memoryless node
  1271. */
  1272. return NULL;
  1273. }
  1274. page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
  1275. zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET);
  1276. if (page)
  1277. goto got_pg;
  1278. /*
  1279. * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
  1280. * __GFP_NOWARN set) should not cause reclaim since the subsystem
  1281. * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
  1282. * using a larger set of nodes after it has established that the
  1283. * allowed per node queues are empty and that nodes are
  1284. * over allocated.
  1285. */
  1286. if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
  1287. goto nopage;
  1288. for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
  1289. wakeup_kswapd(zone, order);
  1290. /*
  1291. * OK, we're below the kswapd watermark and have kicked background
  1292. * reclaim. Now things get more complex, so set up alloc_flags according
  1293. * to how we want to proceed.
  1294. *
  1295. * The caller may dip into page reserves a bit more if the caller
  1296. * cannot run direct reclaim, or if the caller has realtime scheduling
  1297. * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
  1298. * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
  1299. */
  1300. alloc_flags = ALLOC_WMARK_MIN;
  1301. if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
  1302. alloc_flags |= ALLOC_HARDER;
  1303. if (gfp_mask & __GFP_HIGH)
  1304. alloc_flags |= ALLOC_HIGH;
  1305. if (wait)
  1306. alloc_flags |= ALLOC_CPUSET;
  1307. /*
  1308. * Go through the zonelist again. Let __GFP_HIGH and allocations
  1309. * coming from realtime tasks go deeper into reserves.
  1310. *
  1311. * This is the last chance, in general, before the goto nopage.
  1312. * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
  1313. * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
  1314. */
  1315. page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
  1316. high_zoneidx, alloc_flags);
  1317. if (page)
  1318. goto got_pg;
  1319. /* This allocation should allow future memory freeing. */
  1320. rebalance:
  1321. if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
  1322. && !in_interrupt()) {
  1323. if (!(gfp_mask & __GFP_NOMEMALLOC)) {
  1324. nofail_alloc:
  1325. /* go through the zonelist yet again, ignoring mins */
  1326. page = get_page_from_freelist(gfp_mask, nodemask, order,
  1327. zonelist, high_zoneidx, ALLOC_NO_WATERMARKS);
  1328. if (page)
  1329. goto got_pg;
  1330. if (gfp_mask & __GFP_NOFAIL) {
  1331. congestion_wait(WRITE, HZ/50);
  1332. goto nofail_alloc;
  1333. }
  1334. }
  1335. goto nopage;
  1336. }
  1337. /* Atomic allocations - we can't balance anything */
  1338. if (!wait)
  1339. goto nopage;
  1340. cond_resched();
  1341. /* We now go into synchronous reclaim */
  1342. cpuset_memory_pressure_bump();
  1343. p->flags |= PF_MEMALLOC;
  1344. reclaim_state.reclaimed_slab = 0;
  1345. p->reclaim_state = &reclaim_state;
  1346. did_some_progress = try_to_free_pages(zonelist, order, gfp_mask);
  1347. p->reclaim_state = NULL;
  1348. p->flags &= ~PF_MEMALLOC;
  1349. cond_resched();
  1350. if (order != 0)
  1351. drain_all_pages();
  1352. if (likely(did_some_progress)) {
  1353. page = get_page_from_freelist(gfp_mask, nodemask, order,
  1354. zonelist, high_zoneidx, alloc_flags);
  1355. if (page)
  1356. goto got_pg;
  1357. } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
  1358. if (!try_set_zone_oom(zonelist, gfp_mask)) {
  1359. schedule_timeout_uninterruptible(1);
  1360. goto restart;
  1361. }
  1362. /*
  1363. * Go through the zonelist yet one more time, keep
  1364. * very high watermark here, this is only to catch
  1365. * a parallel oom killing, we must fail if we're still
  1366. * under heavy pressure.
  1367. */
  1368. page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
  1369. order, zonelist, high_zoneidx,
  1370. ALLOC_WMARK_HIGH|ALLOC_CPUSET);
  1371. if (page) {
  1372. clear_zonelist_oom(zonelist, gfp_mask);
  1373. goto got_pg;
  1374. }
  1375. /* The OOM killer will not help higher order allocs so fail */
  1376. if (order > PAGE_ALLOC_COSTLY_ORDER) {
  1377. clear_zonelist_oom(zonelist, gfp_mask);
  1378. goto nopage;
  1379. }
  1380. out_of_memory(zonelist, gfp_mask, order);
  1381. clear_zonelist_oom(zonelist, gfp_mask);
  1382. goto restart;
  1383. }
  1384. /*
  1385. * Don't let big-order allocations loop unless the caller explicitly
  1386. * requests that. Wait for some write requests to complete then retry.
  1387. *
  1388. * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
  1389. * means __GFP_NOFAIL, but that may not be true in other
  1390. * implementations.
  1391. *
  1392. * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
  1393. * specified, then we retry until we no longer reclaim any pages
  1394. * (above), or we've reclaimed an order of pages at least as
  1395. * large as the allocation's order. In both cases, if the
  1396. * allocation still fails, we stop retrying.
  1397. */
  1398. pages_reclaimed += did_some_progress;
  1399. do_retry = 0;
  1400. if (!(gfp_mask & __GFP_NORETRY)) {
  1401. if (order <= PAGE_ALLOC_COSTLY_ORDER) {
  1402. do_retry = 1;
  1403. } else {
  1404. if (gfp_mask & __GFP_REPEAT &&
  1405. pages_reclaimed < (1 << order))
  1406. do_retry = 1;
  1407. }
  1408. if (gfp_mask & __GFP_NOFAIL)
  1409. do_retry = 1;
  1410. }
  1411. if (do_retry) {
  1412. congestion_wait(WRITE, HZ/50);
  1413. goto rebalance;
  1414. }
  1415. nopage:
  1416. if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
  1417. printk(KERN_WARNING "%s: page allocation failure."
  1418. " order:%d, mode:0x%x\n",
  1419. p->comm, order, gfp_mask);
  1420. dump_stack();
  1421. show_mem();
  1422. }
  1423. got_pg:
  1424. return page;
  1425. }
  1426. EXPORT_SYMBOL(__alloc_pages_internal);
  1427. /*
  1428. * Common helper functions.
  1429. */
  1430. unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
  1431. {
  1432. struct page * page;
  1433. page = alloc_pages(gfp_mask, order);
  1434. if (!page)
  1435. return 0;
  1436. return (unsigned long) page_address(page);
  1437. }
  1438. EXPORT_SYMBOL(__get_free_pages);
  1439. unsigned long get_zeroed_page(gfp_t gfp_mask)
  1440. {
  1441. struct page * page;
  1442. /*
  1443. * get_zeroed_page() returns a 32-bit address, which cannot represent
  1444. * a highmem page
  1445. */
  1446. VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
  1447. page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
  1448. if (page)
  1449. return (unsigned long) page_address(page);
  1450. return 0;
  1451. }
  1452. EXPORT_SYMBOL(get_zeroed_page);
  1453. void __pagevec_free(struct pagevec *pvec)
  1454. {
  1455. int i = pagevec_count(pvec);
  1456. while (--i >= 0)
  1457. free_hot_cold_page(pvec->pages[i], pvec->cold);
  1458. }
  1459. void __free_pages(struct page *page, unsigned int order)
  1460. {
  1461. if (put_page_testzero(page)) {
  1462. if (order == 0)
  1463. free_hot_page(page);
  1464. else
  1465. __free_pages_ok(page, order);
  1466. }
  1467. }
  1468. EXPORT_SYMBOL(__free_pages);
  1469. void free_pages(unsigned long addr, unsigned int order)
  1470. {
  1471. if (addr != 0) {
  1472. VM_BUG_ON(!virt_addr_valid((void *)addr));
  1473. __free_pages(virt_to_page((void *)addr), order);
  1474. }
  1475. }
  1476. EXPORT_SYMBOL(free_pages);
  1477. /**
  1478. * alloc_pages_exact - allocate an exact number physically-contiguous pages.
  1479. * @size: the number of bytes to allocate
  1480. * @gfp_mask: GFP flags for the allocation
  1481. *
  1482. * This function is similar to alloc_pages(), except that it allocates the
  1483. * minimum number of pages to satisfy the request. alloc_pages() can only
  1484. * allocate memory in power-of-two pages.
  1485. *
  1486. * This function is also limited by MAX_ORDER.
  1487. *
  1488. * Memory allocated by this function must be released by free_pages_exact().
  1489. */
  1490. void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
  1491. {
  1492. unsigned int order = get_order(size);
  1493. unsigned long addr;
  1494. addr = __get_free_pages(gfp_mask, order);
  1495. if (addr) {
  1496. unsigned long alloc_end = addr + (PAGE_SIZE << order);
  1497. unsigned long used = addr + PAGE_ALIGN(size);
  1498. split_page(virt_to_page(addr), order);
  1499. while (used < alloc_end) {
  1500. free_page(used);
  1501. used += PAGE_SIZE;
  1502. }
  1503. }
  1504. return (void *)addr;
  1505. }
  1506. EXPORT_SYMBOL(alloc_pages_exact);
  1507. /**
  1508. * free_pages_exact - release memory allocated via alloc_pages_exact()
  1509. * @virt: the value returned by alloc_pages_exact.
  1510. * @size: size of allocation, same value as passed to alloc_pages_exact().
  1511. *
  1512. * Release the memory allocated by a previous call to alloc_pages_exact.
  1513. */
  1514. void free_pages_exact(void *virt, size_t size)
  1515. {
  1516. unsigned long addr = (unsigned long)virt;
  1517. unsigned long end = addr + PAGE_ALIGN(size);
  1518. while (addr < end) {
  1519. free_page(addr);
  1520. addr += PAGE_SIZE;
  1521. }
  1522. }
  1523. EXPORT_SYMBOL(free_pages_exact);
  1524. static unsigned int nr_free_zone_pages(int offset)
  1525. {
  1526. struct zoneref *z;
  1527. struct zone *zone;
  1528. /* Just pick one node, since fallback list is circular */
  1529. unsigned int sum = 0;
  1530. struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
  1531. for_each_zone_zonelist(zone, z, zonelist, offset) {
  1532. unsigned long size = zone->present_pages;
  1533. unsigned long high = zone->pages_high;
  1534. if (size > high)
  1535. sum += size - high;
  1536. }
  1537. return sum;
  1538. }
  1539. /*
  1540. * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
  1541. */
  1542. unsigned int nr_free_buffer_pages(void)
  1543. {
  1544. return nr_free_zone_pages(gfp_zone(GFP_USER));
  1545. }
  1546. EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
  1547. /*
  1548. * Amount of free RAM allocatable within all zones
  1549. */
  1550. unsigned int nr_free_pagecache_pages(void)
  1551. {
  1552. return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
  1553. }
  1554. static inline void show_node(struct zone *zone)
  1555. {
  1556. if (NUMA_BUILD)
  1557. printk("Node %d ", zone_to_nid(zone));
  1558. }
  1559. void si_meminfo(struct sysinfo *val)
  1560. {
  1561. val->totalram = totalram_pages;
  1562. val->sharedram = 0;
  1563. val->freeram = global_page_state(NR_FREE_PAGES);
  1564. val->bufferram = nr_blockdev_pages();
  1565. val->totalhigh = totalhigh_pages;
  1566. val->freehigh = nr_free_highpages();
  1567. val->mem_unit = PAGE_SIZE;
  1568. }
  1569. EXPORT_SYMBOL(si_meminfo);
  1570. #ifdef CONFIG_NUMA
  1571. void si_meminfo_node(struct sysinfo *val, int nid)
  1572. {
  1573. pg_data_t *pgdat = NODE_DATA(nid);
  1574. val->totalram = pgdat->node_present_pages;
  1575. val->freeram = node_page_state(nid, NR_FREE_PAGES);
  1576. #ifdef CONFIG_HIGHMEM
  1577. val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
  1578. val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
  1579. NR_FREE_PAGES);
  1580. #else
  1581. val->totalhigh = 0;
  1582. val->freehigh = 0;
  1583. #endif
  1584. val->mem_unit = PAGE_SIZE;
  1585. }
  1586. #endif
  1587. #define K(x) ((x) << (PAGE_SHIFT-10))
  1588. /*
  1589. * Show free area list (used inside shift_scroll-lock stuff)
  1590. * We also calculate the percentage fragmentation. We do this by counting the
  1591. * memory on each free list with the exception of the first item on the list.
  1592. */
  1593. void show_free_areas(void)
  1594. {
  1595. int cpu;
  1596. struct zone *zone;
  1597. for_each_zone(zone) {
  1598. if (!populated_zone(zone))
  1599. continue;
  1600. show_node(zone);
  1601. printk("%s per-cpu:\n", zone->name);
  1602. for_each_online_cpu(cpu) {
  1603. struct per_cpu_pageset *pageset;
  1604. pageset = zone_pcp(zone, cpu);
  1605. printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
  1606. cpu, pageset->pcp.high,
  1607. pageset->pcp.batch, pageset->pcp.count);
  1608. }
  1609. }
  1610. printk("Active_anon:%lu active_file:%lu inactive_anon%lu\n"
  1611. " inactive_file:%lu dirty:%lu writeback:%lu unstable:%lu\n"
  1612. " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
  1613. global_page_state(NR_ACTIVE_ANON),
  1614. global_page_state(NR_ACTIVE_FILE),
  1615. global_page_state(NR_INACTIVE_ANON),
  1616. global_page_state(NR_INACTIVE_FILE),
  1617. global_page_state(NR_FILE_DIRTY),
  1618. global_page_state(NR_WRITEBACK),
  1619. global_page_state(NR_UNSTABLE_NFS),
  1620. global_page_state(NR_FREE_PAGES),
  1621. global_page_state(NR_SLAB_RECLAIMABLE) +
  1622. global_page_state(NR_SLAB_UNRECLAIMABLE),
  1623. global_page_state(NR_FILE_MAPPED),
  1624. global_page_state(NR_PAGETABLE),
  1625. global_page_state(NR_BOUNCE));
  1626. for_each_zone(zone) {
  1627. int i;
  1628. if (!populated_zone(zone))
  1629. continue;
  1630. show_node(zone);
  1631. printk("%s"
  1632. " free:%lukB"
  1633. " min:%lukB"
  1634. " low:%lukB"
  1635. " high:%lukB"
  1636. " active_anon:%lukB"
  1637. " inactive_anon:%lukB"
  1638. " active_file:%lukB"
  1639. " inactive_file:%lukB"
  1640. " present:%lukB"
  1641. " pages_scanned:%lu"
  1642. " all_unreclaimable? %s"
  1643. "\n",
  1644. zone->name,
  1645. K(zone_page_state(zone, NR_FREE_PAGES)),
  1646. K(zone->pages_min),
  1647. K(zone->pages_low),
  1648. K(zone->pages_high),
  1649. K(zone_page_state(zone, NR_ACTIVE_ANON)),
  1650. K(zone_page_state(zone, NR_INACTIVE_ANON)),
  1651. K(zone_page_state(zone, NR_ACTIVE_FILE)),
  1652. K(zone_page_state(zone, NR_INACTIVE_FILE)),
  1653. K(zone->present_pages),
  1654. zone->pages_scanned,
  1655. (zone_is_all_unreclaimable(zone) ? "yes" : "no")
  1656. );
  1657. printk("lowmem_reserve[]:");
  1658. for (i = 0; i < MAX_NR_ZONES; i++)
  1659. printk(" %lu", zone->lowmem_reserve[i]);
  1660. printk("\n");
  1661. }
  1662. for_each_zone(zone) {
  1663. unsigned long nr[MAX_ORDER], flags, order, total = 0;
  1664. if (!populated_zone(zone))
  1665. continue;
  1666. show_node(zone);
  1667. printk("%s: ", zone->name);
  1668. spin_lock_irqsave(&zone->lock, flags);
  1669. for (order = 0; order < MAX_ORDER; order++) {
  1670. nr[order] = zone->free_area[order].nr_free;
  1671. total += nr[order] << order;
  1672. }
  1673. spin_unlock_irqrestore(&zone->lock, flags);
  1674. for (order = 0; order < MAX_ORDER; order++)
  1675. printk("%lu*%lukB ", nr[order], K(1UL) << order);
  1676. printk("= %lukB\n", K(total));
  1677. }
  1678. printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
  1679. show_swap_cache_info();
  1680. }
  1681. static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
  1682. {
  1683. zoneref->zone = zone;
  1684. zoneref->zone_idx = zone_idx(zone);
  1685. }
  1686. /*
  1687. * Builds allocation fallback zone lists.
  1688. *
  1689. * Add all populated zones of a node to the zonelist.
  1690. */
  1691. static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
  1692. int nr_zones, enum zone_type zone_type)
  1693. {
  1694. struct zone *zone;
  1695. BUG_ON(zone_type >= MAX_NR_ZONES);
  1696. zone_type++;
  1697. do {
  1698. zone_type--;
  1699. zone = pgdat->node_zones + zone_type;
  1700. if (populated_zone(zone)) {
  1701. zoneref_set_zone(zone,
  1702. &zonelist->_zonerefs[nr_zones++]);
  1703. check_highest_zone(zone_type);
  1704. }
  1705. } while (zone_type);
  1706. return nr_zones;
  1707. }
  1708. /*
  1709. * zonelist_order:
  1710. * 0 = automatic detection of better ordering.
  1711. * 1 = order by ([node] distance, -zonetype)
  1712. * 2 = order by (-zonetype, [node] distance)
  1713. *
  1714. * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
  1715. * the same zonelist. So only NUMA can configure this param.
  1716. */
  1717. #define ZONELIST_ORDER_DEFAULT 0
  1718. #define ZONELIST_ORDER_NODE 1
  1719. #define ZONELIST_ORDER_ZONE 2
  1720. /* zonelist order in the kernel.
  1721. * set_zonelist_order() will set this to NODE or ZONE.
  1722. */
  1723. static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
  1724. static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
  1725. #ifdef CONFIG_NUMA
  1726. /* The value user specified ....changed by config */
  1727. static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
  1728. /* string for sysctl */
  1729. #define NUMA_ZONELIST_ORDER_LEN 16
  1730. char numa_zonelist_order[16] = "default";
  1731. /*
  1732. * interface for configure zonelist ordering.
  1733. * command line option "numa_zonelist_order"
  1734. * = "[dD]efault - default, automatic configuration.
  1735. * = "[nN]ode - order by node locality, then by zone within node
  1736. * = "[zZ]one - order by zone, then by locality within zone
  1737. */
  1738. static int __parse_numa_zonelist_order(char *s)
  1739. {
  1740. if (*s == 'd' || *s == 'D') {
  1741. user_zonelist_order = ZONELIST_ORDER_DEFAULT;
  1742. } else if (*s == 'n' || *s == 'N') {
  1743. user_zonelist_order = ZONELIST_ORDER_NODE;
  1744. } else if (*s == 'z' || *s == 'Z') {
  1745. user_zonelist_order = ZONELIST_ORDER_ZONE;
  1746. } else {
  1747. printk(KERN_WARNING
  1748. "Ignoring invalid numa_zonelist_order value: "
  1749. "%s\n", s);
  1750. return -EINVAL;
  1751. }
  1752. return 0;
  1753. }
  1754. static __init int setup_numa_zonelist_order(char *s)
  1755. {
  1756. if (s)
  1757. return __parse_numa_zonelist_order(s);
  1758. return 0;
  1759. }
  1760. early_param("numa_zonelist_order", setup_numa_zonelist_order);
  1761. /*
  1762. * sysctl handler for numa_zonelist_order
  1763. */
  1764. int numa_zonelist_order_handler(ctl_table *table, int write,
  1765. struct file *file, void __user *buffer, size_t *length,
  1766. loff_t *ppos)
  1767. {
  1768. char saved_string[NUMA_ZONELIST_ORDER_LEN];
  1769. int ret;
  1770. if (write)
  1771. strncpy(saved_string, (char*)table->data,
  1772. NUMA_ZONELIST_ORDER_LEN);
  1773. ret = proc_dostring(table, write, file, buffer, length, ppos);
  1774. if (ret)
  1775. return ret;
  1776. if (write) {
  1777. int oldval = user_zonelist_order;
  1778. if (__parse_numa_zonelist_order((char*)table->data)) {
  1779. /*
  1780. * bogus value. restore saved string
  1781. */
  1782. strncpy((char*)table->data, saved_string,
  1783. NUMA_ZONELIST_ORDER_LEN);
  1784. user_zonelist_order = oldval;
  1785. } else if (oldval != user_zonelist_order)
  1786. build_all_zonelists();
  1787. }
  1788. return 0;
  1789. }
  1790. #define MAX_NODE_LOAD (num_online_nodes())
  1791. static int node_load[MAX_NUMNODES];
  1792. /**
  1793. * find_next_best_node - find the next node that should appear in a given node's fallback list
  1794. * @node: node whose fallback list we're appending
  1795. * @used_node_mask: nodemask_t of already used nodes
  1796. *
  1797. * We use a number of factors to determine which is the next node that should
  1798. * appear on a given node's fallback list. The node should not have appeared
  1799. * already in @node's fallback list, and it should be the next closest node
  1800. * according to the distance array (which contains arbitrary distance values
  1801. * from each node to each node in the system), and should also prefer nodes
  1802. * with no CPUs, since presumably they'll have very little allocation pressure
  1803. * on them otherwise.
  1804. * It returns -1 if no node is found.
  1805. */
  1806. static int find_next_best_node(int node, nodemask_t *used_node_mask)
  1807. {
  1808. int n, val;
  1809. int min_val = INT_MAX;
  1810. int best_node = -1;
  1811. node_to_cpumask_ptr(tmp, 0);
  1812. /* Use the local node if we haven't already */
  1813. if (!node_isset(node, *used_node_mask)) {
  1814. node_set(node, *used_node_mask);
  1815. return node;
  1816. }
  1817. for_each_node_state(n, N_HIGH_MEMORY) {
  1818. /* Don't want a node to appear more than once */
  1819. if (node_isset(n, *used_node_mask))
  1820. continue;
  1821. /* Use the distance array to find the distance */
  1822. val = node_distance(node, n);
  1823. /* Penalize nodes under us ("prefer the next node") */
  1824. val += (n < node);
  1825. /* Give preference to headless and unused nodes */
  1826. node_to_cpumask_ptr_next(tmp, n);
  1827. if (!cpus_empty(*tmp))
  1828. val += PENALTY_FOR_NODE_WITH_CPUS;
  1829. /* Slight preference for less loaded node */
  1830. val *= (MAX_NODE_LOAD*MAX_NUMNODES);
  1831. val += node_load[n];
  1832. if (val < min_val) {
  1833. min_val = val;
  1834. best_node = n;
  1835. }
  1836. }
  1837. if (best_node >= 0)
  1838. node_set(best_node, *used_node_mask);
  1839. return best_node;
  1840. }
  1841. /*
  1842. * Build zonelists ordered by node and zones within node.
  1843. * This results in maximum locality--normal zone overflows into local
  1844. * DMA zone, if any--but risks exhausting DMA zone.
  1845. */
  1846. static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
  1847. {
  1848. int j;
  1849. struct zonelist *zonelist;
  1850. zonelist = &pgdat->node_zonelists[0];
  1851. for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
  1852. ;
  1853. j = build_zonelists_node(NODE_DATA(node), zonelist, j,
  1854. MAX_NR_ZONES - 1);
  1855. zonelist->_zonerefs[j].zone = NULL;
  1856. zonelist->_zonerefs[j].zone_idx = 0;
  1857. }
  1858. /*
  1859. * Build gfp_thisnode zonelists
  1860. */
  1861. static void build_thisnode_zonelists(pg_data_t *pgdat)
  1862. {
  1863. int j;
  1864. struct zonelist *zonelist;
  1865. zonelist = &pgdat->node_zonelists[1];
  1866. j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
  1867. zonelist->_zonerefs[j].zone = NULL;
  1868. zonelist->_zonerefs[j].zone_idx = 0;
  1869. }
  1870. /*
  1871. * Build zonelists ordered by zone and nodes within zones.
  1872. * This results in conserving DMA zone[s] until all Normal memory is
  1873. * exhausted, but results in overflowing to remote node while memory
  1874. * may still exist in local DMA zone.
  1875. */
  1876. static int node_order[MAX_NUMNODES];
  1877. static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
  1878. {
  1879. int pos, j, node;
  1880. int zone_type; /* needs to be signed */
  1881. struct zone *z;
  1882. struct zonelist *zonelist;
  1883. zonelist = &pgdat->node_zonelists[0];
  1884. pos = 0;
  1885. for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
  1886. for (j = 0; j < nr_nodes; j++) {
  1887. node = node_order[j];
  1888. z = &NODE_DATA(node)->node_zones[zone_type];
  1889. if (populated_zone(z)) {
  1890. zoneref_set_zone(z,
  1891. &zonelist->_zonerefs[pos++]);
  1892. check_highest_zone(zone_type);
  1893. }
  1894. }
  1895. }
  1896. zonelist->_zonerefs[pos].zone = NULL;
  1897. zonelist->_zonerefs[pos].zone_idx = 0;
  1898. }
  1899. static int default_zonelist_order(void)
  1900. {
  1901. int nid, zone_type;
  1902. unsigned long low_kmem_size,total_size;
  1903. struct zone *z;
  1904. int average_size;
  1905. /*
  1906. * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
  1907. * If they are really small and used heavily, the system can fall
  1908. * into OOM very easily.
  1909. * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
  1910. */
  1911. /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
  1912. low_kmem_size = 0;
  1913. total_size = 0;
  1914. for_each_online_node(nid) {
  1915. for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
  1916. z = &NODE_DATA(nid)->node_zones[zone_type];
  1917. if (populated_zone(z)) {
  1918. if (zone_type < ZONE_NORMAL)
  1919. low_kmem_size += z->present_pages;
  1920. total_size += z->present_pages;
  1921. }
  1922. }
  1923. }
  1924. if (!low_kmem_size || /* there are no DMA area. */
  1925. low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
  1926. return ZONELIST_ORDER_NODE;
  1927. /*
  1928. * look into each node's config.
  1929. * If there is a node whose DMA/DMA32 memory is very big area on
  1930. * local memory, NODE_ORDER may be suitable.
  1931. */
  1932. average_size = total_size /
  1933. (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
  1934. for_each_online_node(nid) {
  1935. low_kmem_size = 0;
  1936. total_size = 0;
  1937. for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
  1938. z = &NODE_DATA(nid)->node_zones[zone_type];
  1939. if (populated_zone(z)) {
  1940. if (zone_type < ZONE_NORMAL)
  1941. low_kmem_size += z->present_pages;
  1942. total_size += z->present_pages;
  1943. }
  1944. }
  1945. if (low_kmem_size &&
  1946. total_size > average_size && /* ignore small node */
  1947. low_kmem_size > total_size * 70/100)
  1948. return ZONELIST_ORDER_NODE;
  1949. }
  1950. return ZONELIST_ORDER_ZONE;
  1951. }
  1952. static void set_zonelist_order(void)
  1953. {
  1954. if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
  1955. current_zonelist_order = default_zonelist_order();
  1956. else
  1957. current_zonelist_order = user_zonelist_order;
  1958. }
  1959. static void build_zonelists(pg_data_t *pgdat)
  1960. {
  1961. int j, node, load;
  1962. enum zone_type i;
  1963. nodemask_t used_mask;
  1964. int local_node, prev_node;
  1965. struct zonelist *zonelist;
  1966. int order = current_zonelist_order;
  1967. /* initialize zonelists */
  1968. for (i = 0; i < MAX_ZONELISTS; i++) {
  1969. zonelist = pgdat->node_zonelists + i;
  1970. zonelist->_zonerefs[0].zone = NULL;
  1971. zonelist->_zonerefs[0].zone_idx = 0;
  1972. }
  1973. /* NUMA-aware ordering of nodes */
  1974. local_node = pgdat->node_id;
  1975. load = num_online_nodes();
  1976. prev_node = local_node;
  1977. nodes_clear(used_mask);
  1978. memset(node_load, 0, sizeof(node_load));
  1979. memset(node_order, 0, sizeof(node_order));
  1980. j = 0;
  1981. while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
  1982. int distance = node_distance(local_node, node);
  1983. /*
  1984. * If another node is sufficiently far away then it is better
  1985. * to reclaim pages in a zone before going off node.
  1986. */
  1987. if (distance > RECLAIM_DISTANCE)
  1988. zone_reclaim_mode = 1;
  1989. /*
  1990. * We don't want to pressure a particular node.
  1991. * So adding penalty to the first node in same
  1992. * distance group to make it round-robin.
  1993. */
  1994. if (distance != node_distance(local_node, prev_node))
  1995. node_load[node] = load;
  1996. prev_node = node;
  1997. load--;
  1998. if (order == ZONELIST_ORDER_NODE)
  1999. build_zonelists_in_node_order(pgdat, node);
  2000. else
  2001. node_order[j++] = node; /* remember order */
  2002. }
  2003. if (order == ZONELIST_ORDER_ZONE) {
  2004. /* calculate node order -- i.e., DMA last! */
  2005. build_zonelists_in_zone_order(pgdat, j);
  2006. }
  2007. build_thisnode_zonelists(pgdat);
  2008. }
  2009. /* Construct the zonelist performance cache - see further mmzone.h */
  2010. static void build_zonelist_cache(pg_data_t *pgdat)
  2011. {
  2012. struct zonelist *zonelist;
  2013. struct zonelist_cache *zlc;
  2014. struct zoneref *z;
  2015. zonelist = &pgdat->node_zonelists[0];
  2016. zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
  2017. bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
  2018. for (z = zonelist->_zonerefs; z->zone; z++)
  2019. zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
  2020. }
  2021. #else /* CONFIG_NUMA */
  2022. static void set_zonelist_order(void)
  2023. {
  2024. current_zonelist_order = ZONELIST_ORDER_ZONE;
  2025. }
  2026. static void build_zonelists(pg_data_t *pgdat)
  2027. {
  2028. int node, local_node;
  2029. enum zone_type j;
  2030. struct zonelist *zonelist;
  2031. local_node = pgdat->node_id;
  2032. zonelist = &pgdat->node_zonelists[0];
  2033. j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
  2034. /*
  2035. * Now we build the zonelist so that it contains the zones
  2036. * of all the other nodes.
  2037. * We don't want to pressure a particular node, so when
  2038. * building the zones for node N, we make sure that the
  2039. * zones coming right after the local ones are those from
  2040. * node N+1 (modulo N)
  2041. */
  2042. for (node = local_node + 1; node < MAX_NUMNODES; node++) {
  2043. if (!node_online(node))
  2044. continue;
  2045. j = build_zonelists_node(NODE_DATA(node), zonelist, j,
  2046. MAX_NR_ZONES - 1);
  2047. }
  2048. for (node = 0; node < local_node; node++) {
  2049. if (!node_online(node))
  2050. continue;
  2051. j = build_zonelists_node(NODE_DATA(node), zonelist, j,
  2052. MAX_NR_ZONES - 1);
  2053. }
  2054. zonelist->_zonerefs[j].zone = NULL;
  2055. zonelist->_zonerefs[j].zone_idx = 0;
  2056. }
  2057. /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
  2058. static void build_zonelist_cache(pg_data_t *pgdat)
  2059. {
  2060. pgdat->node_zonelists[0].zlcache_ptr = NULL;
  2061. }
  2062. #endif /* CONFIG_NUMA */
  2063. /* return values int ....just for stop_machine() */
  2064. static int __build_all_zonelists(void *dummy)
  2065. {
  2066. int nid;
  2067. for_each_online_node(nid) {
  2068. pg_data_t *pgdat = NODE_DATA(nid);
  2069. build_zonelists(pgdat);
  2070. build_zonelist_cache(pgdat);
  2071. }
  2072. return 0;
  2073. }
  2074. void build_all_zonelists(void)
  2075. {
  2076. set_zonelist_order();
  2077. if (system_state == SYSTEM_BOOTING) {
  2078. __build_all_zonelists(NULL);
  2079. mminit_verify_zonelist();
  2080. cpuset_init_current_mems_allowed();
  2081. } else {
  2082. /* we have to stop all cpus to guarantee there is no user
  2083. of zonelist */
  2084. stop_machine(__build_all_zonelists, NULL, NULL);
  2085. /* cpuset refresh routine should be here */
  2086. }
  2087. vm_total_pages = nr_free_pagecache_pages();
  2088. /*
  2089. * Disable grouping by mobility if the number of pages in the
  2090. * system is too low to allow the mechanism to work. It would be
  2091. * more accurate, but expensive to check per-zone. This check is
  2092. * made on memory-hotadd so a system can start with mobility
  2093. * disabled and enable it later
  2094. */
  2095. if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
  2096. page_group_by_mobility_disabled = 1;
  2097. else
  2098. page_group_by_mobility_disabled = 0;
  2099. printk("Built %i zonelists in %s order, mobility grouping %s. "
  2100. "Total pages: %ld\n",
  2101. num_online_nodes(),
  2102. zonelist_order_name[current_zonelist_order],
  2103. page_group_by_mobility_disabled ? "off" : "on",
  2104. vm_total_pages);
  2105. #ifdef CONFIG_NUMA
  2106. printk("Policy zone: %s\n", zone_names[policy_zone]);
  2107. #endif
  2108. }
  2109. /*
  2110. * Helper functions to size the waitqueue hash table.
  2111. * Essentially these want to choose hash table sizes sufficiently
  2112. * large so that collisions trying to wait on pages are rare.
  2113. * But in fact, the number of active page waitqueues on typical
  2114. * systems is ridiculously low, less than 200. So this is even
  2115. * conservative, even though it seems large.
  2116. *
  2117. * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
  2118. * waitqueues, i.e. the size of the waitq table given the number of pages.
  2119. */
  2120. #define PAGES_PER_WAITQUEUE 256
  2121. #ifndef CONFIG_MEMORY_HOTPLUG
  2122. static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
  2123. {
  2124. unsigned long size = 1;
  2125. pages /= PAGES_PER_WAITQUEUE;
  2126. while (size < pages)
  2127. size <<= 1;
  2128. /*
  2129. * Once we have dozens or even hundreds of threads sleeping
  2130. * on IO we've got bigger problems than wait queue collision.
  2131. * Limit the size of the wait table to a reasonable size.
  2132. */
  2133. size = min(size, 4096UL);
  2134. return max(size, 4UL);
  2135. }
  2136. #else
  2137. /*
  2138. * A zone's size might be changed by hot-add, so it is not possible to determine
  2139. * a suitable size for its wait_table. So we use the maximum size now.
  2140. *
  2141. * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
  2142. *
  2143. * i386 (preemption config) : 4096 x 16 = 64Kbyte.
  2144. * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
  2145. * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
  2146. *
  2147. * The maximum entries are prepared when a zone's memory is (512K + 256) pages
  2148. * or more by the traditional way. (See above). It equals:
  2149. *
  2150. * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
  2151. * ia64(16K page size) : = ( 8G + 4M)byte.
  2152. * powerpc (64K page size) : = (32G +16M)byte.
  2153. */
  2154. static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
  2155. {
  2156. return 4096UL;
  2157. }
  2158. #endif
  2159. /*
  2160. * This is an integer logarithm so that shifts can be used later
  2161. * to extract the more random high bits from the multiplicative
  2162. * hash function before the remainder is taken.
  2163. */
  2164. static inline unsigned long wait_table_bits(unsigned long size)
  2165. {
  2166. return ffz(~size);
  2167. }
  2168. #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
  2169. /*
  2170. * Mark a number of pageblocks as MIGRATE_RESERVE. The number
  2171. * of blocks reserved is based on zone->pages_min. The memory within the
  2172. * reserve will tend to store contiguous free pages. Setting min_free_kbytes
  2173. * higher will lead to a bigger reserve which will get freed as contiguous
  2174. * blocks as reclaim kicks in
  2175. */
  2176. static void setup_zone_migrate_reserve(struct zone *zone)
  2177. {
  2178. unsigned long start_pfn, pfn, end_pfn;
  2179. struct page *page;
  2180. unsigned long reserve, block_migratetype;
  2181. /* Get the start pfn, end pfn and the number of blocks to reserve */
  2182. start_pfn = zone->zone_start_pfn;
  2183. end_pfn = start_pfn + zone->spanned_pages;
  2184. reserve = roundup(zone->pages_min, pageblock_nr_pages) >>
  2185. pageblock_order;
  2186. for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
  2187. if (!pfn_valid(pfn))
  2188. continue;
  2189. page = pfn_to_page(pfn);
  2190. /* Watch out for overlapping nodes */
  2191. if (page_to_nid(page) != zone_to_nid(zone))
  2192. continue;
  2193. /* Blocks with reserved pages will never free, skip them. */
  2194. if (PageReserved(page))
  2195. continue;
  2196. block_migratetype = get_pageblock_migratetype(page);
  2197. /* If this block is reserved, account for it */
  2198. if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
  2199. reserve--;
  2200. continue;
  2201. }
  2202. /* Suitable for reserving if this block is movable */
  2203. if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
  2204. set_pageblock_migratetype(page, MIGRATE_RESERVE);
  2205. move_freepages_block(zone, page, MIGRATE_RESERVE);
  2206. reserve--;
  2207. continue;
  2208. }
  2209. /*
  2210. * If the reserve is met and this is a previous reserved block,
  2211. * take it back
  2212. */
  2213. if (block_migratetype == MIGRATE_RESERVE) {
  2214. set_pageblock_migratetype(page, MIGRATE_MOVABLE);
  2215. move_freepages_block(zone, page, MIGRATE_MOVABLE);
  2216. }
  2217. }
  2218. }
  2219. /*
  2220. * Initially all pages are reserved - free ones are freed
  2221. * up by free_all_bootmem() once the early boot process is
  2222. * done. Non-atomic initialization, single-pass.
  2223. */
  2224. void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
  2225. unsigned long start_pfn, enum memmap_context context)
  2226. {
  2227. struct page *page;
  2228. unsigned long end_pfn = start_pfn + size;
  2229. unsigned long pfn;
  2230. struct zone *z;
  2231. z = &NODE_DATA(nid)->node_zones[zone];
  2232. for (pfn = start_pfn; pfn < end_pfn; pfn++) {
  2233. /*
  2234. * There can be holes in boot-time mem_map[]s
  2235. * handed to this function. They do not
  2236. * exist on hotplugged memory.
  2237. */
  2238. if (context == MEMMAP_EARLY) {
  2239. if (!early_pfn_valid(pfn))
  2240. continue;
  2241. if (!early_pfn_in_nid(pfn, nid))
  2242. continue;
  2243. }
  2244. page = pfn_to_page(pfn);
  2245. set_page_links(page, zone, nid, pfn);
  2246. mminit_verify_page_links(page, zone, nid, pfn);
  2247. init_page_count(page);
  2248. reset_page_mapcount(page);
  2249. SetPageReserved(page);
  2250. /*
  2251. * Mark the block movable so that blocks are reserved for
  2252. * movable at startup. This will force kernel allocations
  2253. * to reserve their blocks rather than leaking throughout
  2254. * the address space during boot when many long-lived
  2255. * kernel allocations are made. Later some blocks near
  2256. * the start are marked MIGRATE_RESERVE by
  2257. * setup_zone_migrate_reserve()
  2258. *
  2259. * bitmap is created for zone's valid pfn range. but memmap
  2260. * can be created for invalid pages (for alignment)
  2261. * check here not to call set_pageblock_migratetype() against
  2262. * pfn out of zone.
  2263. */
  2264. if ((z->zone_start_pfn <= pfn)
  2265. && (pfn < z->zone_start_pfn + z->spanned_pages)
  2266. && !(pfn & (pageblock_nr_pages - 1)))
  2267. set_pageblock_migratetype(page, MIGRATE_MOVABLE);
  2268. INIT_LIST_HEAD(&page->lru);
  2269. #ifdef WANT_PAGE_VIRTUAL
  2270. /* The shift won't overflow because ZONE_NORMAL is below 4G. */
  2271. if (!is_highmem_idx(zone))
  2272. set_page_address(page, __va(pfn << PAGE_SHIFT));
  2273. #endif
  2274. }
  2275. }
  2276. static void __meminit zone_init_free_lists(struct zone *zone)
  2277. {
  2278. int order, t;
  2279. for_each_migratetype_order(order, t) {
  2280. INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
  2281. zone->free_area[order].nr_free = 0;
  2282. }
  2283. }
  2284. #ifndef __HAVE_ARCH_MEMMAP_INIT
  2285. #define memmap_init(size, nid, zone, start_pfn) \
  2286. memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
  2287. #endif
  2288. static int zone_batchsize(struct zone *zone)
  2289. {
  2290. int batch;
  2291. /*
  2292. * The per-cpu-pages pools are set to around 1000th of the
  2293. * size of the zone. But no more than 1/2 of a meg.
  2294. *
  2295. * OK, so we don't know how big the cache is. So guess.
  2296. */
  2297. batch = zone->present_pages / 1024;
  2298. if (batch * PAGE_SIZE > 512 * 1024)
  2299. batch = (512 * 1024) / PAGE_SIZE;
  2300. batch /= 4; /* We effectively *= 4 below */
  2301. if (batch < 1)
  2302. batch = 1;
  2303. /*
  2304. * Clamp the batch to a 2^n - 1 value. Having a power
  2305. * of 2 value was found to be more likely to have
  2306. * suboptimal cache aliasing properties in some cases.
  2307. *
  2308. * For example if 2 tasks are alternately allocating
  2309. * batches of pages, one task can end up with a lot
  2310. * of pages of one half of the possible page colors
  2311. * and the other with pages of the other colors.
  2312. */
  2313. batch = (1 << (fls(batch + batch/2)-1)) - 1;
  2314. return batch;
  2315. }
  2316. static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
  2317. {
  2318. struct per_cpu_pages *pcp;
  2319. memset(p, 0, sizeof(*p));
  2320. pcp = &p->pcp;
  2321. pcp->count = 0;
  2322. pcp->high = 6 * batch;
  2323. pcp->batch = max(1UL, 1 * batch);
  2324. INIT_LIST_HEAD(&pcp->list);
  2325. }
  2326. /*
  2327. * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
  2328. * to the value high for the pageset p.
  2329. */
  2330. static void setup_pagelist_highmark(struct per_cpu_pageset *p,
  2331. unsigned long high)
  2332. {
  2333. struct per_cpu_pages *pcp;
  2334. pcp = &p->pcp;
  2335. pcp->high = high;
  2336. pcp->batch = max(1UL, high/4);
  2337. if ((high/4) > (PAGE_SHIFT * 8))
  2338. pcp->batch = PAGE_SHIFT * 8;
  2339. }
  2340. #ifdef CONFIG_NUMA
  2341. /*
  2342. * Boot pageset table. One per cpu which is going to be used for all
  2343. * zones and all nodes. The parameters will be set in such a way
  2344. * that an item put on a list will immediately be handed over to
  2345. * the buddy list. This is safe since pageset manipulation is done
  2346. * with interrupts disabled.
  2347. *
  2348. * Some NUMA counter updates may also be caught by the boot pagesets.
  2349. *
  2350. * The boot_pagesets must be kept even after bootup is complete for
  2351. * unused processors and/or zones. They do play a role for bootstrapping
  2352. * hotplugged processors.
  2353. *
  2354. * zoneinfo_show() and maybe other functions do
  2355. * not check if the processor is online before following the pageset pointer.
  2356. * Other parts of the kernel may not check if the zone is available.
  2357. */
  2358. static struct per_cpu_pageset boot_pageset[NR_CPUS];
  2359. /*
  2360. * Dynamically allocate memory for the
  2361. * per cpu pageset array in struct zone.
  2362. */
  2363. static int __cpuinit process_zones(int cpu)
  2364. {
  2365. struct zone *zone, *dzone;
  2366. int node = cpu_to_node(cpu);
  2367. node_set_state(node, N_CPU); /* this node has a cpu */
  2368. for_each_zone(zone) {
  2369. if (!populated_zone(zone))
  2370. continue;
  2371. zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
  2372. GFP_KERNEL, node);
  2373. if (!zone_pcp(zone, cpu))
  2374. goto bad;
  2375. setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
  2376. if (percpu_pagelist_fraction)
  2377. setup_pagelist_highmark(zone_pcp(zone, cpu),
  2378. (zone->present_pages / percpu_pagelist_fraction));
  2379. }
  2380. return 0;
  2381. bad:
  2382. for_each_zone(dzone) {
  2383. if (!populated_zone(dzone))
  2384. continue;
  2385. if (dzone == zone)
  2386. break;
  2387. kfree(zone_pcp(dzone, cpu));
  2388. zone_pcp(dzone, cpu) = NULL;
  2389. }
  2390. return -ENOMEM;
  2391. }
  2392. static inline void free_zone_pagesets(int cpu)
  2393. {
  2394. struct zone *zone;
  2395. for_each_zone(zone) {
  2396. struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
  2397. /* Free per_cpu_pageset if it is slab allocated */
  2398. if (pset != &boot_pageset[cpu])
  2399. kfree(pset);
  2400. zone_pcp(zone, cpu) = NULL;
  2401. }
  2402. }
  2403. static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
  2404. unsigned long action,
  2405. void *hcpu)
  2406. {
  2407. int cpu = (long)hcpu;
  2408. int ret = NOTIFY_OK;
  2409. switch (action) {
  2410. case CPU_UP_PREPARE:
  2411. case CPU_UP_PREPARE_FROZEN:
  2412. if (process_zones(cpu))
  2413. ret = NOTIFY_BAD;
  2414. break;
  2415. case CPU_UP_CANCELED:
  2416. case CPU_UP_CANCELED_FROZEN:
  2417. case CPU_DEAD:
  2418. case CPU_DEAD_FROZEN:
  2419. free_zone_pagesets(cpu);
  2420. break;
  2421. default:
  2422. break;
  2423. }
  2424. return ret;
  2425. }
  2426. static struct notifier_block __cpuinitdata pageset_notifier =
  2427. { &pageset_cpuup_callback, NULL, 0 };
  2428. void __init setup_per_cpu_pageset(void)
  2429. {
  2430. int err;
  2431. /* Initialize per_cpu_pageset for cpu 0.
  2432. * A cpuup callback will do this for every cpu
  2433. * as it comes online
  2434. */
  2435. err = process_zones(smp_processor_id());
  2436. BUG_ON(err);
  2437. register_cpu_notifier(&pageset_notifier);
  2438. }
  2439. #endif
  2440. static noinline __init_refok
  2441. int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
  2442. {
  2443. int i;
  2444. struct pglist_data *pgdat = zone->zone_pgdat;
  2445. size_t alloc_size;
  2446. /*
  2447. * The per-page waitqueue mechanism uses hashed waitqueues
  2448. * per zone.
  2449. */
  2450. zone->wait_table_hash_nr_entries =
  2451. wait_table_hash_nr_entries(zone_size_pages);
  2452. zone->wait_table_bits =
  2453. wait_table_bits(zone->wait_table_hash_nr_entries);
  2454. alloc_size = zone->wait_table_hash_nr_entries
  2455. * sizeof(wait_queue_head_t);
  2456. if (!slab_is_available()) {
  2457. zone->wait_table = (wait_queue_head_t *)
  2458. alloc_bootmem_node(pgdat, alloc_size);
  2459. } else {
  2460. /*
  2461. * This case means that a zone whose size was 0 gets new memory
  2462. * via memory hot-add.
  2463. * But it may be the case that a new node was hot-added. In
  2464. * this case vmalloc() will not be able to use this new node's
  2465. * memory - this wait_table must be initialized to use this new
  2466. * node itself as well.
  2467. * To use this new node's memory, further consideration will be
  2468. * necessary.
  2469. */
  2470. zone->wait_table = vmalloc(alloc_size);
  2471. }
  2472. if (!zone->wait_table)
  2473. return -ENOMEM;
  2474. for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
  2475. init_waitqueue_head(zone->wait_table + i);
  2476. return 0;
  2477. }
  2478. static __meminit void zone_pcp_init(struct zone *zone)
  2479. {
  2480. int cpu;
  2481. unsigned long batch = zone_batchsize(zone);
  2482. for (cpu = 0; cpu < NR_CPUS; cpu++) {
  2483. #ifdef CONFIG_NUMA
  2484. /* Early boot. Slab allocator not functional yet */
  2485. zone_pcp(zone, cpu) = &boot_pageset[cpu];
  2486. setup_pageset(&boot_pageset[cpu],0);
  2487. #else
  2488. setup_pageset(zone_pcp(zone,cpu), batch);
  2489. #endif
  2490. }
  2491. if (zone->present_pages)
  2492. printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
  2493. zone->name, zone->present_pages, batch);
  2494. }
  2495. __meminit int init_currently_empty_zone(struct zone *zone,
  2496. unsigned long zone_start_pfn,
  2497. unsigned long size,
  2498. enum memmap_context context)
  2499. {
  2500. struct pglist_data *pgdat = zone->zone_pgdat;
  2501. int ret;
  2502. ret = zone_wait_table_init(zone, size);
  2503. if (ret)
  2504. return ret;
  2505. pgdat->nr_zones = zone_idx(zone) + 1;
  2506. zone->zone_start_pfn = zone_start_pfn;
  2507. mminit_dprintk(MMINIT_TRACE, "memmap_init",
  2508. "Initialising map node %d zone %lu pfns %lu -> %lu\n",
  2509. pgdat->node_id,
  2510. (unsigned long)zone_idx(zone),
  2511. zone_start_pfn, (zone_start_pfn + size));
  2512. zone_init_free_lists(zone);
  2513. return 0;
  2514. }
  2515. #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
  2516. /*
  2517. * Basic iterator support. Return the first range of PFNs for a node
  2518. * Note: nid == MAX_NUMNODES returns first region regardless of node
  2519. */
  2520. static int __meminit first_active_region_index_in_nid(int nid)
  2521. {
  2522. int i;
  2523. for (i = 0; i < nr_nodemap_entries; i++)
  2524. if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
  2525. return i;
  2526. return -1;
  2527. }
  2528. /*
  2529. * Basic iterator support. Return the next active range of PFNs for a node
  2530. * Note: nid == MAX_NUMNODES returns next region regardless of node
  2531. */
  2532. static int __meminit next_active_region_index_in_nid(int index, int nid)
  2533. {
  2534. for (index = index + 1; index < nr_nodemap_entries; index++)
  2535. if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
  2536. return index;
  2537. return -1;
  2538. }
  2539. #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
  2540. /*
  2541. * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
  2542. * Architectures may implement their own version but if add_active_range()
  2543. * was used and there are no special requirements, this is a convenient
  2544. * alternative
  2545. */
  2546. int __meminit early_pfn_to_nid(unsigned long pfn)
  2547. {
  2548. int i;
  2549. for (i = 0; i < nr_nodemap_entries; i++) {
  2550. unsigned long start_pfn = early_node_map[i].start_pfn;
  2551. unsigned long end_pfn = early_node_map[i].end_pfn;
  2552. if (start_pfn <= pfn && pfn < end_pfn)
  2553. return early_node_map[i].nid;
  2554. }
  2555. return 0;
  2556. }
  2557. #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
  2558. /* Basic iterator support to walk early_node_map[] */
  2559. #define for_each_active_range_index_in_nid(i, nid) \
  2560. for (i = first_active_region_index_in_nid(nid); i != -1; \
  2561. i = next_active_region_index_in_nid(i, nid))
  2562. /**
  2563. * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
  2564. * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
  2565. * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
  2566. *
  2567. * If an architecture guarantees that all ranges registered with
  2568. * add_active_ranges() contain no holes and may be freed, this
  2569. * this function may be used instead of calling free_bootmem() manually.
  2570. */
  2571. void __init free_bootmem_with_active_regions(int nid,
  2572. unsigned long max_low_pfn)
  2573. {
  2574. int i;
  2575. for_each_active_range_index_in_nid(i, nid) {
  2576. unsigned long size_pages = 0;
  2577. unsigned long end_pfn = early_node_map[i].end_pfn;
  2578. if (early_node_map[i].start_pfn >= max_low_pfn)
  2579. continue;
  2580. if (end_pfn > max_low_pfn)
  2581. end_pfn = max_low_pfn;
  2582. size_pages = end_pfn - early_node_map[i].start_pfn;
  2583. free_bootmem_node(NODE_DATA(early_node_map[i].nid),
  2584. PFN_PHYS(early_node_map[i].start_pfn),
  2585. size_pages << PAGE_SHIFT);
  2586. }
  2587. }
  2588. void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data)
  2589. {
  2590. int i;
  2591. int ret;
  2592. for_each_active_range_index_in_nid(i, nid) {
  2593. ret = work_fn(early_node_map[i].start_pfn,
  2594. early_node_map[i].end_pfn, data);
  2595. if (ret)
  2596. break;
  2597. }
  2598. }
  2599. /**
  2600. * sparse_memory_present_with_active_regions - Call memory_present for each active range
  2601. * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
  2602. *
  2603. * If an architecture guarantees that all ranges registered with
  2604. * add_active_ranges() contain no holes and may be freed, this
  2605. * function may be used instead of calling memory_present() manually.
  2606. */
  2607. void __init sparse_memory_present_with_active_regions(int nid)
  2608. {
  2609. int i;
  2610. for_each_active_range_index_in_nid(i, nid)
  2611. memory_present(early_node_map[i].nid,
  2612. early_node_map[i].start_pfn,
  2613. early_node_map[i].end_pfn);
  2614. }
  2615. /**
  2616. * push_node_boundaries - Push node boundaries to at least the requested boundary
  2617. * @nid: The nid of the node to push the boundary for
  2618. * @start_pfn: The start pfn of the node
  2619. * @end_pfn: The end pfn of the node
  2620. *
  2621. * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
  2622. * time. Specifically, on x86_64, SRAT will report ranges that can potentially
  2623. * be hotplugged even though no physical memory exists. This function allows
  2624. * an arch to push out the node boundaries so mem_map is allocated that can
  2625. * be used later.
  2626. */
  2627. #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
  2628. void __init push_node_boundaries(unsigned int nid,
  2629. unsigned long start_pfn, unsigned long end_pfn)
  2630. {
  2631. mminit_dprintk(MMINIT_TRACE, "zoneboundary",
  2632. "Entering push_node_boundaries(%u, %lu, %lu)\n",
  2633. nid, start_pfn, end_pfn);
  2634. /* Initialise the boundary for this node if necessary */
  2635. if (node_boundary_end_pfn[nid] == 0)
  2636. node_boundary_start_pfn[nid] = -1UL;
  2637. /* Update the boundaries */
  2638. if (node_boundary_start_pfn[nid] > start_pfn)
  2639. node_boundary_start_pfn[nid] = start_pfn;
  2640. if (node_boundary_end_pfn[nid] < end_pfn)
  2641. node_boundary_end_pfn[nid] = end_pfn;
  2642. }
  2643. /* If necessary, push the node boundary out for reserve hotadd */
  2644. static void __meminit account_node_boundary(unsigned int nid,
  2645. unsigned long *start_pfn, unsigned long *end_pfn)
  2646. {
  2647. mminit_dprintk(MMINIT_TRACE, "zoneboundary",
  2648. "Entering account_node_boundary(%u, %lu, %lu)\n",
  2649. nid, *start_pfn, *end_pfn);
  2650. /* Return if boundary information has not been provided */
  2651. if (node_boundary_end_pfn[nid] == 0)
  2652. return;
  2653. /* Check the boundaries and update if necessary */
  2654. if (node_boundary_start_pfn[nid] < *start_pfn)
  2655. *start_pfn = node_boundary_start_pfn[nid];
  2656. if (node_boundary_end_pfn[nid] > *end_pfn)
  2657. *end_pfn = node_boundary_end_pfn[nid];
  2658. }
  2659. #else
  2660. void __init push_node_boundaries(unsigned int nid,
  2661. unsigned long start_pfn, unsigned long end_pfn) {}
  2662. static void __meminit account_node_boundary(unsigned int nid,
  2663. unsigned long *start_pfn, unsigned long *end_pfn) {}
  2664. #endif
  2665. /**
  2666. * get_pfn_range_for_nid - Return the start and end page frames for a node
  2667. * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
  2668. * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
  2669. * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
  2670. *
  2671. * It returns the start and end page frame of a node based on information
  2672. * provided by an arch calling add_active_range(). If called for a node
  2673. * with no available memory, a warning is printed and the start and end
  2674. * PFNs will be 0.
  2675. */
  2676. void __meminit get_pfn_range_for_nid(unsigned int nid,
  2677. unsigned long *start_pfn, unsigned long *end_pfn)
  2678. {
  2679. int i;
  2680. *start_pfn = -1UL;
  2681. *end_pfn = 0;
  2682. for_each_active_range_index_in_nid(i, nid) {
  2683. *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
  2684. *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
  2685. }
  2686. if (*start_pfn == -1UL)
  2687. *start_pfn = 0;
  2688. /* Push the node boundaries out if requested */
  2689. account_node_boundary(nid, start_pfn, end_pfn);
  2690. }
  2691. /*
  2692. * This finds a zone that can be used for ZONE_MOVABLE pages. The
  2693. * assumption is made that zones within a node are ordered in monotonic
  2694. * increasing memory addresses so that the "highest" populated zone is used
  2695. */
  2696. static void __init find_usable_zone_for_movable(void)
  2697. {
  2698. int zone_index;
  2699. for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
  2700. if (zone_index == ZONE_MOVABLE)
  2701. continue;
  2702. if (arch_zone_highest_possible_pfn[zone_index] >
  2703. arch_zone_lowest_possible_pfn[zone_index])
  2704. break;
  2705. }
  2706. VM_BUG_ON(zone_index == -1);
  2707. movable_zone = zone_index;
  2708. }
  2709. /*
  2710. * The zone ranges provided by the architecture do not include ZONE_MOVABLE
  2711. * because it is sized independant of architecture. Unlike the other zones,
  2712. * the starting point for ZONE_MOVABLE is not fixed. It may be different
  2713. * in each node depending on the size of each node and how evenly kernelcore
  2714. * is distributed. This helper function adjusts the zone ranges
  2715. * provided by the architecture for a given node by using the end of the
  2716. * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
  2717. * zones within a node are in order of monotonic increases memory addresses
  2718. */
  2719. static void __meminit adjust_zone_range_for_zone_movable(int nid,
  2720. unsigned long zone_type,
  2721. unsigned long node_start_pfn,
  2722. unsigned long node_end_pfn,
  2723. unsigned long *zone_start_pfn,
  2724. unsigned long *zone_end_pfn)
  2725. {
  2726. /* Only adjust if ZONE_MOVABLE is on this node */
  2727. if (zone_movable_pfn[nid]) {
  2728. /* Size ZONE_MOVABLE */
  2729. if (zone_type == ZONE_MOVABLE) {
  2730. *zone_start_pfn = zone_movable_pfn[nid];
  2731. *zone_end_pfn = min(node_end_pfn,
  2732. arch_zone_highest_possible_pfn[movable_zone]);
  2733. /* Adjust for ZONE_MOVABLE starting within this range */
  2734. } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
  2735. *zone_end_pfn > zone_movable_pfn[nid]) {
  2736. *zone_end_pfn = zone_movable_pfn[nid];
  2737. /* Check if this whole range is within ZONE_MOVABLE */
  2738. } else if (*zone_start_pfn >= zone_movable_pfn[nid])
  2739. *zone_start_pfn = *zone_end_pfn;
  2740. }
  2741. }
  2742. /*
  2743. * Return the number of pages a zone spans in a node, including holes
  2744. * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
  2745. */
  2746. static unsigned long __meminit zone_spanned_pages_in_node(int nid,
  2747. unsigned long zone_type,
  2748. unsigned long *ignored)
  2749. {
  2750. unsigned long node_start_pfn, node_end_pfn;
  2751. unsigned long zone_start_pfn, zone_end_pfn;
  2752. /* Get the start and end of the node and zone */
  2753. get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
  2754. zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
  2755. zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
  2756. adjust_zone_range_for_zone_movable(nid, zone_type,
  2757. node_start_pfn, node_end_pfn,
  2758. &zone_start_pfn, &zone_end_pfn);
  2759. /* Check that this node has pages within the zone's required range */
  2760. if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
  2761. return 0;
  2762. /* Move the zone boundaries inside the node if necessary */
  2763. zone_end_pfn = min(zone_end_pfn, node_end_pfn);
  2764. zone_start_pfn = max(zone_start_pfn, node_start_pfn);
  2765. /* Return the spanned pages */
  2766. return zone_end_pfn - zone_start_pfn;
  2767. }
  2768. /*
  2769. * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
  2770. * then all holes in the requested range will be accounted for.
  2771. */
  2772. static unsigned long __meminit __absent_pages_in_range(int nid,
  2773. unsigned long range_start_pfn,
  2774. unsigned long range_end_pfn)
  2775. {
  2776. int i = 0;
  2777. unsigned long prev_end_pfn = 0, hole_pages = 0;
  2778. unsigned long start_pfn;
  2779. /* Find the end_pfn of the first active range of pfns in the node */
  2780. i = first_active_region_index_in_nid(nid);
  2781. if (i == -1)
  2782. return 0;
  2783. prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
  2784. /* Account for ranges before physical memory on this node */
  2785. if (early_node_map[i].start_pfn > range_start_pfn)
  2786. hole_pages = prev_end_pfn - range_start_pfn;
  2787. /* Find all holes for the zone within the node */
  2788. for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
  2789. /* No need to continue if prev_end_pfn is outside the zone */
  2790. if (prev_end_pfn >= range_end_pfn)
  2791. break;
  2792. /* Make sure the end of the zone is not within the hole */
  2793. start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
  2794. prev_end_pfn = max(prev_end_pfn, range_start_pfn);
  2795. /* Update the hole size cound and move on */
  2796. if (start_pfn > range_start_pfn) {
  2797. BUG_ON(prev_end_pfn > start_pfn);
  2798. hole_pages += start_pfn - prev_end_pfn;
  2799. }
  2800. prev_end_pfn = early_node_map[i].end_pfn;
  2801. }
  2802. /* Account for ranges past physical memory on this node */
  2803. if (range_end_pfn > prev_end_pfn)
  2804. hole_pages += range_end_pfn -
  2805. max(range_start_pfn, prev_end_pfn);
  2806. return hole_pages;
  2807. }
  2808. /**
  2809. * absent_pages_in_range - Return number of page frames in holes within a range
  2810. * @start_pfn: The start PFN to start searching for holes
  2811. * @end_pfn: The end PFN to stop searching for holes
  2812. *
  2813. * It returns the number of pages frames in memory holes within a range.
  2814. */
  2815. unsigned long __init absent_pages_in_range(unsigned long start_pfn,
  2816. unsigned long end_pfn)
  2817. {
  2818. return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
  2819. }
  2820. /* Return the number of page frames in holes in a zone on a node */
  2821. static unsigned long __meminit zone_absent_pages_in_node(int nid,
  2822. unsigned long zone_type,
  2823. unsigned long *ignored)
  2824. {
  2825. unsigned long node_start_pfn, node_end_pfn;
  2826. unsigned long zone_start_pfn, zone_end_pfn;
  2827. get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
  2828. zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
  2829. node_start_pfn);
  2830. zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
  2831. node_end_pfn);
  2832. adjust_zone_range_for_zone_movable(nid, zone_type,
  2833. node_start_pfn, node_end_pfn,
  2834. &zone_start_pfn, &zone_end_pfn);
  2835. return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
  2836. }
  2837. #else
  2838. static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
  2839. unsigned long zone_type,
  2840. unsigned long *zones_size)
  2841. {
  2842. return zones_size[zone_type];
  2843. }
  2844. static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
  2845. unsigned long zone_type,
  2846. unsigned long *zholes_size)
  2847. {
  2848. if (!zholes_size)
  2849. return 0;
  2850. return zholes_size[zone_type];
  2851. }
  2852. #endif
  2853. static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
  2854. unsigned long *zones_size, unsigned long *zholes_size)
  2855. {
  2856. unsigned long realtotalpages, totalpages = 0;
  2857. enum zone_type i;
  2858. for (i = 0; i < MAX_NR_ZONES; i++)
  2859. totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
  2860. zones_size);
  2861. pgdat->node_spanned_pages = totalpages;
  2862. realtotalpages = totalpages;
  2863. for (i = 0; i < MAX_NR_ZONES; i++)
  2864. realtotalpages -=
  2865. zone_absent_pages_in_node(pgdat->node_id, i,
  2866. zholes_size);
  2867. pgdat->node_present_pages = realtotalpages;
  2868. printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
  2869. realtotalpages);
  2870. }
  2871. #ifndef CONFIG_SPARSEMEM
  2872. /*
  2873. * Calculate the size of the zone->blockflags rounded to an unsigned long
  2874. * Start by making sure zonesize is a multiple of pageblock_order by rounding
  2875. * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
  2876. * round what is now in bits to nearest long in bits, then return it in
  2877. * bytes.
  2878. */
  2879. static unsigned long __init usemap_size(unsigned long zonesize)
  2880. {
  2881. unsigned long usemapsize;
  2882. usemapsize = roundup(zonesize, pageblock_nr_pages);
  2883. usemapsize = usemapsize >> pageblock_order;
  2884. usemapsize *= NR_PAGEBLOCK_BITS;
  2885. usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
  2886. return usemapsize / 8;
  2887. }
  2888. static void __init setup_usemap(struct pglist_data *pgdat,
  2889. struct zone *zone, unsigned long zonesize)
  2890. {
  2891. unsigned long usemapsize = usemap_size(zonesize);
  2892. zone->pageblock_flags = NULL;
  2893. if (usemapsize) {
  2894. zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
  2895. memset(zone->pageblock_flags, 0, usemapsize);
  2896. }
  2897. }
  2898. #else
  2899. static void inline setup_usemap(struct pglist_data *pgdat,
  2900. struct zone *zone, unsigned long zonesize) {}
  2901. #endif /* CONFIG_SPARSEMEM */
  2902. #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
  2903. /* Return a sensible default order for the pageblock size. */
  2904. static inline int pageblock_default_order(void)
  2905. {
  2906. if (HPAGE_SHIFT > PAGE_SHIFT)
  2907. return HUGETLB_PAGE_ORDER;
  2908. return MAX_ORDER-1;
  2909. }
  2910. /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
  2911. static inline void __init set_pageblock_order(unsigned int order)
  2912. {
  2913. /* Check that pageblock_nr_pages has not already been setup */
  2914. if (pageblock_order)
  2915. return;
  2916. /*
  2917. * Assume the largest contiguous order of interest is a huge page.
  2918. * This value may be variable depending on boot parameters on IA64
  2919. */
  2920. pageblock_order = order;
  2921. }
  2922. #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
  2923. /*
  2924. * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
  2925. * and pageblock_default_order() are unused as pageblock_order is set
  2926. * at compile-time. See include/linux/pageblock-flags.h for the values of
  2927. * pageblock_order based on the kernel config
  2928. */
  2929. static inline int pageblock_default_order(unsigned int order)
  2930. {
  2931. return MAX_ORDER-1;
  2932. }
  2933. #define set_pageblock_order(x) do {} while (0)
  2934. #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
  2935. /*
  2936. * Set up the zone data structures:
  2937. * - mark all pages reserved
  2938. * - mark all memory queues empty
  2939. * - clear the memory bitmaps
  2940. */
  2941. static void __paginginit free_area_init_core(struct pglist_data *pgdat,
  2942. unsigned long *zones_size, unsigned long *zholes_size)
  2943. {
  2944. enum zone_type j;
  2945. int nid = pgdat->node_id;
  2946. unsigned long zone_start_pfn = pgdat->node_start_pfn;
  2947. int ret;
  2948. pgdat_resize_init(pgdat);
  2949. pgdat->nr_zones = 0;
  2950. init_waitqueue_head(&pgdat->kswapd_wait);
  2951. pgdat->kswapd_max_order = 0;
  2952. for (j = 0; j < MAX_NR_ZONES; j++) {
  2953. struct zone *zone = pgdat->node_zones + j;
  2954. unsigned long size, realsize, memmap_pages;
  2955. enum lru_list l;
  2956. size = zone_spanned_pages_in_node(nid, j, zones_size);
  2957. realsize = size - zone_absent_pages_in_node(nid, j,
  2958. zholes_size);
  2959. /*
  2960. * Adjust realsize so that it accounts for how much memory
  2961. * is used by this zone for memmap. This affects the watermark
  2962. * and per-cpu initialisations
  2963. */
  2964. memmap_pages =
  2965. PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
  2966. if (realsize >= memmap_pages) {
  2967. realsize -= memmap_pages;
  2968. mminit_dprintk(MMINIT_TRACE, "memmap_init",
  2969. "%s zone: %lu pages used for memmap\n",
  2970. zone_names[j], memmap_pages);
  2971. } else
  2972. printk(KERN_WARNING
  2973. " %s zone: %lu pages exceeds realsize %lu\n",
  2974. zone_names[j], memmap_pages, realsize);
  2975. /* Account for reserved pages */
  2976. if (j == 0 && realsize > dma_reserve) {
  2977. realsize -= dma_reserve;
  2978. mminit_dprintk(MMINIT_TRACE, "memmap_init",
  2979. "%s zone: %lu pages reserved\n",
  2980. zone_names[0], dma_reserve);
  2981. }
  2982. if (!is_highmem_idx(j))
  2983. nr_kernel_pages += realsize;
  2984. nr_all_pages += realsize;
  2985. zone->spanned_pages = size;
  2986. zone->present_pages = realsize;
  2987. #ifdef CONFIG_NUMA
  2988. zone->node = nid;
  2989. zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
  2990. / 100;
  2991. zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
  2992. #endif
  2993. zone->name = zone_names[j];
  2994. spin_lock_init(&zone->lock);
  2995. spin_lock_init(&zone->lru_lock);
  2996. zone_seqlock_init(zone);
  2997. zone->zone_pgdat = pgdat;
  2998. zone->prev_priority = DEF_PRIORITY;
  2999. zone_pcp_init(zone);
  3000. for_each_lru(l) {
  3001. INIT_LIST_HEAD(&zone->lru[l].list);
  3002. zone->lru[l].nr_scan = 0;
  3003. }
  3004. zone->recent_rotated[0] = 0;
  3005. zone->recent_rotated[1] = 0;
  3006. zone->recent_scanned[0] = 0;
  3007. zone->recent_scanned[1] = 0;
  3008. zap_zone_vm_stats(zone);
  3009. zone->flags = 0;
  3010. if (!size)
  3011. continue;
  3012. set_pageblock_order(pageblock_default_order());
  3013. setup_usemap(pgdat, zone, size);
  3014. ret = init_currently_empty_zone(zone, zone_start_pfn,
  3015. size, MEMMAP_EARLY);
  3016. BUG_ON(ret);
  3017. memmap_init(size, nid, j, zone_start_pfn);
  3018. zone_start_pfn += size;
  3019. }
  3020. }
  3021. static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
  3022. {
  3023. /* Skip empty nodes */
  3024. if (!pgdat->node_spanned_pages)
  3025. return;
  3026. #ifdef CONFIG_FLAT_NODE_MEM_MAP
  3027. /* ia64 gets its own node_mem_map, before this, without bootmem */
  3028. if (!pgdat->node_mem_map) {
  3029. unsigned long size, start, end;
  3030. struct page *map;
  3031. /*
  3032. * The zone's endpoints aren't required to be MAX_ORDER
  3033. * aligned but the node_mem_map endpoints must be in order
  3034. * for the buddy allocator to function correctly.
  3035. */
  3036. start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
  3037. end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
  3038. end = ALIGN(end, MAX_ORDER_NR_PAGES);
  3039. size = (end - start) * sizeof(struct page);
  3040. map = alloc_remap(pgdat->node_id, size);
  3041. if (!map)
  3042. map = alloc_bootmem_node(pgdat, size);
  3043. pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
  3044. }
  3045. #ifndef CONFIG_NEED_MULTIPLE_NODES
  3046. /*
  3047. * With no DISCONTIG, the global mem_map is just set as node 0's
  3048. */
  3049. if (pgdat == NODE_DATA(0)) {
  3050. mem_map = NODE_DATA(0)->node_mem_map;
  3051. #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
  3052. if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
  3053. mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
  3054. #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
  3055. }
  3056. #endif
  3057. #endif /* CONFIG_FLAT_NODE_MEM_MAP */
  3058. }
  3059. void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
  3060. unsigned long node_start_pfn, unsigned long *zholes_size)
  3061. {
  3062. pg_data_t *pgdat = NODE_DATA(nid);
  3063. pgdat->node_id = nid;
  3064. pgdat->node_start_pfn = node_start_pfn;
  3065. calculate_node_totalpages(pgdat, zones_size, zholes_size);
  3066. alloc_node_mem_map(pgdat);
  3067. #ifdef CONFIG_FLAT_NODE_MEM_MAP
  3068. printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
  3069. nid, (unsigned long)pgdat,
  3070. (unsigned long)pgdat->node_mem_map);
  3071. #endif
  3072. free_area_init_core(pgdat, zones_size, zholes_size);
  3073. }
  3074. #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
  3075. #if MAX_NUMNODES > 1
  3076. /*
  3077. * Figure out the number of possible node ids.
  3078. */
  3079. static void __init setup_nr_node_ids(void)
  3080. {
  3081. unsigned int node;
  3082. unsigned int highest = 0;
  3083. for_each_node_mask(node, node_possible_map)
  3084. highest = node;
  3085. nr_node_ids = highest + 1;
  3086. }
  3087. #else
  3088. static inline void setup_nr_node_ids(void)
  3089. {
  3090. }
  3091. #endif
  3092. /**
  3093. * add_active_range - Register a range of PFNs backed by physical memory
  3094. * @nid: The node ID the range resides on
  3095. * @start_pfn: The start PFN of the available physical memory
  3096. * @end_pfn: The end PFN of the available physical memory
  3097. *
  3098. * These ranges are stored in an early_node_map[] and later used by
  3099. * free_area_init_nodes() to calculate zone sizes and holes. If the
  3100. * range spans a memory hole, it is up to the architecture to ensure
  3101. * the memory is not freed by the bootmem allocator. If possible
  3102. * the range being registered will be merged with existing ranges.
  3103. */
  3104. void __init add_active_range(unsigned int nid, unsigned long start_pfn,
  3105. unsigned long end_pfn)
  3106. {
  3107. int i;
  3108. mminit_dprintk(MMINIT_TRACE, "memory_register",
  3109. "Entering add_active_range(%d, %#lx, %#lx) "
  3110. "%d entries of %d used\n",
  3111. nid, start_pfn, end_pfn,
  3112. nr_nodemap_entries, MAX_ACTIVE_REGIONS);
  3113. mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
  3114. /* Merge with existing active regions if possible */
  3115. for (i = 0; i < nr_nodemap_entries; i++) {
  3116. if (early_node_map[i].nid != nid)
  3117. continue;
  3118. /* Skip if an existing region covers this new one */
  3119. if (start_pfn >= early_node_map[i].start_pfn &&
  3120. end_pfn <= early_node_map[i].end_pfn)
  3121. return;
  3122. /* Merge forward if suitable */
  3123. if (start_pfn <= early_node_map[i].end_pfn &&
  3124. end_pfn > early_node_map[i].end_pfn) {
  3125. early_node_map[i].end_pfn = end_pfn;
  3126. return;
  3127. }
  3128. /* Merge backward if suitable */
  3129. if (start_pfn < early_node_map[i].end_pfn &&
  3130. end_pfn >= early_node_map[i].start_pfn) {
  3131. early_node_map[i].start_pfn = start_pfn;
  3132. return;
  3133. }
  3134. }
  3135. /* Check that early_node_map is large enough */
  3136. if (i >= MAX_ACTIVE_REGIONS) {
  3137. printk(KERN_CRIT "More than %d memory regions, truncating\n",
  3138. MAX_ACTIVE_REGIONS);
  3139. return;
  3140. }
  3141. early_node_map[i].nid = nid;
  3142. early_node_map[i].start_pfn = start_pfn;
  3143. early_node_map[i].end_pfn = end_pfn;
  3144. nr_nodemap_entries = i + 1;
  3145. }
  3146. /**
  3147. * remove_active_range - Shrink an existing registered range of PFNs
  3148. * @nid: The node id the range is on that should be shrunk
  3149. * @start_pfn: The new PFN of the range
  3150. * @end_pfn: The new PFN of the range
  3151. *
  3152. * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
  3153. * The map is kept near the end physical page range that has already been
  3154. * registered. This function allows an arch to shrink an existing registered
  3155. * range.
  3156. */
  3157. void __init remove_active_range(unsigned int nid, unsigned long start_pfn,
  3158. unsigned long end_pfn)
  3159. {
  3160. int i, j;
  3161. int removed = 0;
  3162. printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n",
  3163. nid, start_pfn, end_pfn);
  3164. /* Find the old active region end and shrink */
  3165. for_each_active_range_index_in_nid(i, nid) {
  3166. if (early_node_map[i].start_pfn >= start_pfn &&
  3167. early_node_map[i].end_pfn <= end_pfn) {
  3168. /* clear it */
  3169. early_node_map[i].start_pfn = 0;
  3170. early_node_map[i].end_pfn = 0;
  3171. removed = 1;
  3172. continue;
  3173. }
  3174. if (early_node_map[i].start_pfn < start_pfn &&
  3175. early_node_map[i].end_pfn > start_pfn) {
  3176. unsigned long temp_end_pfn = early_node_map[i].end_pfn;
  3177. early_node_map[i].end_pfn = start_pfn;
  3178. if (temp_end_pfn > end_pfn)
  3179. add_active_range(nid, end_pfn, temp_end_pfn);
  3180. continue;
  3181. }
  3182. if (early_node_map[i].start_pfn >= start_pfn &&
  3183. early_node_map[i].end_pfn > end_pfn &&
  3184. early_node_map[i].start_pfn < end_pfn) {
  3185. early_node_map[i].start_pfn = end_pfn;
  3186. continue;
  3187. }
  3188. }
  3189. if (!removed)
  3190. return;
  3191. /* remove the blank ones */
  3192. for (i = nr_nodemap_entries - 1; i > 0; i--) {
  3193. if (early_node_map[i].nid != nid)
  3194. continue;
  3195. if (early_node_map[i].end_pfn)
  3196. continue;
  3197. /* we found it, get rid of it */
  3198. for (j = i; j < nr_nodemap_entries - 1; j++)
  3199. memcpy(&early_node_map[j], &early_node_map[j+1],
  3200. sizeof(early_node_map[j]));
  3201. j = nr_nodemap_entries - 1;
  3202. memset(&early_node_map[j], 0, sizeof(early_node_map[j]));
  3203. nr_nodemap_entries--;
  3204. }
  3205. }
  3206. /**
  3207. * remove_all_active_ranges - Remove all currently registered regions
  3208. *
  3209. * During discovery, it may be found that a table like SRAT is invalid
  3210. * and an alternative discovery method must be used. This function removes
  3211. * all currently registered regions.
  3212. */
  3213. void __init remove_all_active_ranges(void)
  3214. {
  3215. memset(early_node_map, 0, sizeof(early_node_map));
  3216. nr_nodemap_entries = 0;
  3217. #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
  3218. memset(node_boundary_start_pfn, 0, sizeof(node_boundary_start_pfn));
  3219. memset(node_boundary_end_pfn, 0, sizeof(node_boundary_end_pfn));
  3220. #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
  3221. }
  3222. /* Compare two active node_active_regions */
  3223. static int __init cmp_node_active_region(const void *a, const void *b)
  3224. {
  3225. struct node_active_region *arange = (struct node_active_region *)a;
  3226. struct node_active_region *brange = (struct node_active_region *)b;
  3227. /* Done this way to avoid overflows */
  3228. if (arange->start_pfn > brange->start_pfn)
  3229. return 1;
  3230. if (arange->start_pfn < brange->start_pfn)
  3231. return -1;
  3232. return 0;
  3233. }
  3234. /* sort the node_map by start_pfn */
  3235. static void __init sort_node_map(void)
  3236. {
  3237. sort(early_node_map, (size_t)nr_nodemap_entries,
  3238. sizeof(struct node_active_region),
  3239. cmp_node_active_region, NULL);
  3240. }
  3241. /* Find the lowest pfn for a node */
  3242. static unsigned long __init find_min_pfn_for_node(int nid)
  3243. {
  3244. int i;
  3245. unsigned long min_pfn = ULONG_MAX;
  3246. /* Assuming a sorted map, the first range found has the starting pfn */
  3247. for_each_active_range_index_in_nid(i, nid)
  3248. min_pfn = min(min_pfn, early_node_map[i].start_pfn);
  3249. if (min_pfn == ULONG_MAX) {
  3250. printk(KERN_WARNING
  3251. "Could not find start_pfn for node %d\n", nid);
  3252. return 0;
  3253. }
  3254. return min_pfn;
  3255. }
  3256. /**
  3257. * find_min_pfn_with_active_regions - Find the minimum PFN registered
  3258. *
  3259. * It returns the minimum PFN based on information provided via
  3260. * add_active_range().
  3261. */
  3262. unsigned long __init find_min_pfn_with_active_regions(void)
  3263. {
  3264. return find_min_pfn_for_node(MAX_NUMNODES);
  3265. }
  3266. /*
  3267. * early_calculate_totalpages()
  3268. * Sum pages in active regions for movable zone.
  3269. * Populate N_HIGH_MEMORY for calculating usable_nodes.
  3270. */
  3271. static unsigned long __init early_calculate_totalpages(void)
  3272. {
  3273. int i;
  3274. unsigned long totalpages = 0;
  3275. for (i = 0; i < nr_nodemap_entries; i++) {
  3276. unsigned long pages = early_node_map[i].end_pfn -
  3277. early_node_map[i].start_pfn;
  3278. totalpages += pages;
  3279. if (pages)
  3280. node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
  3281. }
  3282. return totalpages;
  3283. }
  3284. /*
  3285. * Find the PFN the Movable zone begins in each node. Kernel memory
  3286. * is spread evenly between nodes as long as the nodes have enough
  3287. * memory. When they don't, some nodes will have more kernelcore than
  3288. * others
  3289. */
  3290. static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
  3291. {
  3292. int i, nid;
  3293. unsigned long usable_startpfn;
  3294. unsigned long kernelcore_node, kernelcore_remaining;
  3295. unsigned long totalpages = early_calculate_totalpages();
  3296. int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
  3297. /*
  3298. * If movablecore was specified, calculate what size of
  3299. * kernelcore that corresponds so that memory usable for
  3300. * any allocation type is evenly spread. If both kernelcore
  3301. * and movablecore are specified, then the value of kernelcore
  3302. * will be used for required_kernelcore if it's greater than
  3303. * what movablecore would have allowed.
  3304. */
  3305. if (required_movablecore) {
  3306. unsigned long corepages;
  3307. /*
  3308. * Round-up so that ZONE_MOVABLE is at least as large as what
  3309. * was requested by the user
  3310. */
  3311. required_movablecore =
  3312. roundup(required_movablecore, MAX_ORDER_NR_PAGES);
  3313. corepages = totalpages - required_movablecore;
  3314. required_kernelcore = max(required_kernelcore, corepages);
  3315. }
  3316. /* If kernelcore was not specified, there is no ZONE_MOVABLE */
  3317. if (!required_kernelcore)
  3318. return;
  3319. /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
  3320. find_usable_zone_for_movable();
  3321. usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
  3322. restart:
  3323. /* Spread kernelcore memory as evenly as possible throughout nodes */
  3324. kernelcore_node = required_kernelcore / usable_nodes;
  3325. for_each_node_state(nid, N_HIGH_MEMORY) {
  3326. /*
  3327. * Recalculate kernelcore_node if the division per node
  3328. * now exceeds what is necessary to satisfy the requested
  3329. * amount of memory for the kernel
  3330. */
  3331. if (required_kernelcore < kernelcore_node)
  3332. kernelcore_node = required_kernelcore / usable_nodes;
  3333. /*
  3334. * As the map is walked, we track how much memory is usable
  3335. * by the kernel using kernelcore_remaining. When it is
  3336. * 0, the rest of the node is usable by ZONE_MOVABLE
  3337. */
  3338. kernelcore_remaining = kernelcore_node;
  3339. /* Go through each range of PFNs within this node */
  3340. for_each_active_range_index_in_nid(i, nid) {
  3341. unsigned long start_pfn, end_pfn;
  3342. unsigned long size_pages;
  3343. start_pfn = max(early_node_map[i].start_pfn,
  3344. zone_movable_pfn[nid]);
  3345. end_pfn = early_node_map[i].end_pfn;
  3346. if (start_pfn >= end_pfn)
  3347. continue;
  3348. /* Account for what is only usable for kernelcore */
  3349. if (start_pfn < usable_startpfn) {
  3350. unsigned long kernel_pages;
  3351. kernel_pages = min(end_pfn, usable_startpfn)
  3352. - start_pfn;
  3353. kernelcore_remaining -= min(kernel_pages,
  3354. kernelcore_remaining);
  3355. required_kernelcore -= min(kernel_pages,
  3356. required_kernelcore);
  3357. /* Continue if range is now fully accounted */
  3358. if (end_pfn <= usable_startpfn) {
  3359. /*
  3360. * Push zone_movable_pfn to the end so
  3361. * that if we have to rebalance
  3362. * kernelcore across nodes, we will
  3363. * not double account here
  3364. */
  3365. zone_movable_pfn[nid] = end_pfn;
  3366. continue;
  3367. }
  3368. start_pfn = usable_startpfn;
  3369. }
  3370. /*
  3371. * The usable PFN range for ZONE_MOVABLE is from
  3372. * start_pfn->end_pfn. Calculate size_pages as the
  3373. * number of pages used as kernelcore
  3374. */
  3375. size_pages = end_pfn - start_pfn;
  3376. if (size_pages > kernelcore_remaining)
  3377. size_pages = kernelcore_remaining;
  3378. zone_movable_pfn[nid] = start_pfn + size_pages;
  3379. /*
  3380. * Some kernelcore has been met, update counts and
  3381. * break if the kernelcore for this node has been
  3382. * satisified
  3383. */
  3384. required_kernelcore -= min(required_kernelcore,
  3385. size_pages);
  3386. kernelcore_remaining -= size_pages;
  3387. if (!kernelcore_remaining)
  3388. break;
  3389. }
  3390. }
  3391. /*
  3392. * If there is still required_kernelcore, we do another pass with one
  3393. * less node in the count. This will push zone_movable_pfn[nid] further
  3394. * along on the nodes that still have memory until kernelcore is
  3395. * satisified
  3396. */
  3397. usable_nodes--;
  3398. if (usable_nodes && required_kernelcore > usable_nodes)
  3399. goto restart;
  3400. /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
  3401. for (nid = 0; nid < MAX_NUMNODES; nid++)
  3402. zone_movable_pfn[nid] =
  3403. roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
  3404. }
  3405. /* Any regular memory on that node ? */
  3406. static void check_for_regular_memory(pg_data_t *pgdat)
  3407. {
  3408. #ifdef CONFIG_HIGHMEM
  3409. enum zone_type zone_type;
  3410. for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
  3411. struct zone *zone = &pgdat->node_zones[zone_type];
  3412. if (zone->present_pages)
  3413. node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
  3414. }
  3415. #endif
  3416. }
  3417. /**
  3418. * free_area_init_nodes - Initialise all pg_data_t and zone data
  3419. * @max_zone_pfn: an array of max PFNs for each zone
  3420. *
  3421. * This will call free_area_init_node() for each active node in the system.
  3422. * Using the page ranges provided by add_active_range(), the size of each
  3423. * zone in each node and their holes is calculated. If the maximum PFN
  3424. * between two adjacent zones match, it is assumed that the zone is empty.
  3425. * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
  3426. * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
  3427. * starts where the previous one ended. For example, ZONE_DMA32 starts
  3428. * at arch_max_dma_pfn.
  3429. */
  3430. void __init free_area_init_nodes(unsigned long *max_zone_pfn)
  3431. {
  3432. unsigned long nid;
  3433. int i;
  3434. /* Sort early_node_map as initialisation assumes it is sorted */
  3435. sort_node_map();
  3436. /* Record where the zone boundaries are */
  3437. memset(arch_zone_lowest_possible_pfn, 0,
  3438. sizeof(arch_zone_lowest_possible_pfn));
  3439. memset(arch_zone_highest_possible_pfn, 0,
  3440. sizeof(arch_zone_highest_possible_pfn));
  3441. arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
  3442. arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
  3443. for (i = 1; i < MAX_NR_ZONES; i++) {
  3444. if (i == ZONE_MOVABLE)
  3445. continue;
  3446. arch_zone_lowest_possible_pfn[i] =
  3447. arch_zone_highest_possible_pfn[i-1];
  3448. arch_zone_highest_possible_pfn[i] =
  3449. max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
  3450. }
  3451. arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
  3452. arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
  3453. /* Find the PFNs that ZONE_MOVABLE begins at in each node */
  3454. memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
  3455. find_zone_movable_pfns_for_nodes(zone_movable_pfn);
  3456. /* Print out the zone ranges */
  3457. printk("Zone PFN ranges:\n");
  3458. for (i = 0; i < MAX_NR_ZONES; i++) {
  3459. if (i == ZONE_MOVABLE)
  3460. continue;
  3461. printk(" %-8s %0#10lx -> %0#10lx\n",
  3462. zone_names[i],
  3463. arch_zone_lowest_possible_pfn[i],
  3464. arch_zone_highest_possible_pfn[i]);
  3465. }
  3466. /* Print out the PFNs ZONE_MOVABLE begins at in each node */
  3467. printk("Movable zone start PFN for each node\n");
  3468. for (i = 0; i < MAX_NUMNODES; i++) {
  3469. if (zone_movable_pfn[i])
  3470. printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
  3471. }
  3472. /* Print out the early_node_map[] */
  3473. printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
  3474. for (i = 0; i < nr_nodemap_entries; i++)
  3475. printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid,
  3476. early_node_map[i].start_pfn,
  3477. early_node_map[i].end_pfn);
  3478. /* Initialise every node */
  3479. mminit_verify_pageflags_layout();
  3480. setup_nr_node_ids();
  3481. for_each_online_node(nid) {
  3482. pg_data_t *pgdat = NODE_DATA(nid);
  3483. free_area_init_node(nid, NULL,
  3484. find_min_pfn_for_node(nid), NULL);
  3485. /* Any memory on that node */
  3486. if (pgdat->node_present_pages)
  3487. node_set_state(nid, N_HIGH_MEMORY);
  3488. check_for_regular_memory(pgdat);
  3489. }
  3490. }
  3491. static int __init cmdline_parse_core(char *p, unsigned long *core)
  3492. {
  3493. unsigned long long coremem;
  3494. if (!p)
  3495. return -EINVAL;
  3496. coremem = memparse(p, &p);
  3497. *core = coremem >> PAGE_SHIFT;
  3498. /* Paranoid check that UL is enough for the coremem value */
  3499. WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
  3500. return 0;
  3501. }
  3502. /*
  3503. * kernelcore=size sets the amount of memory for use for allocations that
  3504. * cannot be reclaimed or migrated.
  3505. */
  3506. static int __init cmdline_parse_kernelcore(char *p)
  3507. {
  3508. return cmdline_parse_core(p, &required_kernelcore);
  3509. }
  3510. /*
  3511. * movablecore=size sets the amount of memory for use for allocations that
  3512. * can be reclaimed or migrated.
  3513. */
  3514. static int __init cmdline_parse_movablecore(char *p)
  3515. {
  3516. return cmdline_parse_core(p, &required_movablecore);
  3517. }
  3518. early_param("kernelcore", cmdline_parse_kernelcore);
  3519. early_param("movablecore", cmdline_parse_movablecore);
  3520. #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
  3521. /**
  3522. * set_dma_reserve - set the specified number of pages reserved in the first zone
  3523. * @new_dma_reserve: The number of pages to mark reserved
  3524. *
  3525. * The per-cpu batchsize and zone watermarks are determined by present_pages.
  3526. * In the DMA zone, a significant percentage may be consumed by kernel image
  3527. * and other unfreeable allocations which can skew the watermarks badly. This
  3528. * function may optionally be used to account for unfreeable pages in the
  3529. * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
  3530. * smaller per-cpu batchsize.
  3531. */
  3532. void __init set_dma_reserve(unsigned long new_dma_reserve)
  3533. {
  3534. dma_reserve = new_dma_reserve;
  3535. }
  3536. #ifndef CONFIG_NEED_MULTIPLE_NODES
  3537. struct pglist_data __refdata contig_page_data = { .bdata = &bootmem_node_data[0] };
  3538. EXPORT_SYMBOL(contig_page_data);
  3539. #endif
  3540. void __init free_area_init(unsigned long *zones_size)
  3541. {
  3542. free_area_init_node(0, zones_size,
  3543. __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
  3544. }
  3545. static int page_alloc_cpu_notify(struct notifier_block *self,
  3546. unsigned long action, void *hcpu)
  3547. {
  3548. int cpu = (unsigned long)hcpu;
  3549. if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
  3550. drain_pages(cpu);
  3551. /*
  3552. * Spill the event counters of the dead processor
  3553. * into the current processors event counters.
  3554. * This artificially elevates the count of the current
  3555. * processor.
  3556. */
  3557. vm_events_fold_cpu(cpu);
  3558. /*
  3559. * Zero the differential counters of the dead processor
  3560. * so that the vm statistics are consistent.
  3561. *
  3562. * This is only okay since the processor is dead and cannot
  3563. * race with what we are doing.
  3564. */
  3565. refresh_cpu_vm_stats(cpu);
  3566. }
  3567. return NOTIFY_OK;
  3568. }
  3569. void __init page_alloc_init(void)
  3570. {
  3571. hotcpu_notifier(page_alloc_cpu_notify, 0);
  3572. }
  3573. /*
  3574. * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
  3575. * or min_free_kbytes changes.
  3576. */
  3577. static void calculate_totalreserve_pages(void)
  3578. {
  3579. struct pglist_data *pgdat;
  3580. unsigned long reserve_pages = 0;
  3581. enum zone_type i, j;
  3582. for_each_online_pgdat(pgdat) {
  3583. for (i = 0; i < MAX_NR_ZONES; i++) {
  3584. struct zone *zone = pgdat->node_zones + i;
  3585. unsigned long max = 0;
  3586. /* Find valid and maximum lowmem_reserve in the zone */
  3587. for (j = i; j < MAX_NR_ZONES; j++) {
  3588. if (zone->lowmem_reserve[j] > max)
  3589. max = zone->lowmem_reserve[j];
  3590. }
  3591. /* we treat pages_high as reserved pages. */
  3592. max += zone->pages_high;
  3593. if (max > zone->present_pages)
  3594. max = zone->present_pages;
  3595. reserve_pages += max;
  3596. }
  3597. }
  3598. totalreserve_pages = reserve_pages;
  3599. }
  3600. /*
  3601. * setup_per_zone_lowmem_reserve - called whenever
  3602. * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
  3603. * has a correct pages reserved value, so an adequate number of
  3604. * pages are left in the zone after a successful __alloc_pages().
  3605. */
  3606. static void setup_per_zone_lowmem_reserve(void)
  3607. {
  3608. struct pglist_data *pgdat;
  3609. enum zone_type j, idx;
  3610. for_each_online_pgdat(pgdat) {
  3611. for (j = 0; j < MAX_NR_ZONES; j++) {
  3612. struct zone *zone = pgdat->node_zones + j;
  3613. unsigned long present_pages = zone->present_pages;
  3614. zone->lowmem_reserve[j] = 0;
  3615. idx = j;
  3616. while (idx) {
  3617. struct zone *lower_zone;
  3618. idx--;
  3619. if (sysctl_lowmem_reserve_ratio[idx] < 1)
  3620. sysctl_lowmem_reserve_ratio[idx] = 1;
  3621. lower_zone = pgdat->node_zones + idx;
  3622. lower_zone->lowmem_reserve[j] = present_pages /
  3623. sysctl_lowmem_reserve_ratio[idx];
  3624. present_pages += lower_zone->present_pages;
  3625. }
  3626. }
  3627. }
  3628. /* update totalreserve_pages */
  3629. calculate_totalreserve_pages();
  3630. }
  3631. /**
  3632. * setup_per_zone_pages_min - called when min_free_kbytes changes.
  3633. *
  3634. * Ensures that the pages_{min,low,high} values for each zone are set correctly
  3635. * with respect to min_free_kbytes.
  3636. */
  3637. void setup_per_zone_pages_min(void)
  3638. {
  3639. unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
  3640. unsigned long lowmem_pages = 0;
  3641. struct zone *zone;
  3642. unsigned long flags;
  3643. /* Calculate total number of !ZONE_HIGHMEM pages */
  3644. for_each_zone(zone) {
  3645. if (!is_highmem(zone))
  3646. lowmem_pages += zone->present_pages;
  3647. }
  3648. for_each_zone(zone) {
  3649. u64 tmp;
  3650. spin_lock_irqsave(&zone->lru_lock, flags);
  3651. tmp = (u64)pages_min * zone->present_pages;
  3652. do_div(tmp, lowmem_pages);
  3653. if (is_highmem(zone)) {
  3654. /*
  3655. * __GFP_HIGH and PF_MEMALLOC allocations usually don't
  3656. * need highmem pages, so cap pages_min to a small
  3657. * value here.
  3658. *
  3659. * The (pages_high-pages_low) and (pages_low-pages_min)
  3660. * deltas controls asynch page reclaim, and so should
  3661. * not be capped for highmem.
  3662. */
  3663. int min_pages;
  3664. min_pages = zone->present_pages / 1024;
  3665. if (min_pages < SWAP_CLUSTER_MAX)
  3666. min_pages = SWAP_CLUSTER_MAX;
  3667. if (min_pages > 128)
  3668. min_pages = 128;
  3669. zone->pages_min = min_pages;
  3670. } else {
  3671. /*
  3672. * If it's a lowmem zone, reserve a number of pages
  3673. * proportionate to the zone's size.
  3674. */
  3675. zone->pages_min = tmp;
  3676. }
  3677. zone->pages_low = zone->pages_min + (tmp >> 2);
  3678. zone->pages_high = zone->pages_min + (tmp >> 1);
  3679. setup_zone_migrate_reserve(zone);
  3680. spin_unlock_irqrestore(&zone->lru_lock, flags);
  3681. }
  3682. /* update totalreserve_pages */
  3683. calculate_totalreserve_pages();
  3684. }
  3685. /*
  3686. * Initialise min_free_kbytes.
  3687. *
  3688. * For small machines we want it small (128k min). For large machines
  3689. * we want it large (64MB max). But it is not linear, because network
  3690. * bandwidth does not increase linearly with machine size. We use
  3691. *
  3692. * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
  3693. * min_free_kbytes = sqrt(lowmem_kbytes * 16)
  3694. *
  3695. * which yields
  3696. *
  3697. * 16MB: 512k
  3698. * 32MB: 724k
  3699. * 64MB: 1024k
  3700. * 128MB: 1448k
  3701. * 256MB: 2048k
  3702. * 512MB: 2896k
  3703. * 1024MB: 4096k
  3704. * 2048MB: 5792k
  3705. * 4096MB: 8192k
  3706. * 8192MB: 11584k
  3707. * 16384MB: 16384k
  3708. */
  3709. static int __init init_per_zone_pages_min(void)
  3710. {
  3711. unsigned long lowmem_kbytes;
  3712. lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
  3713. min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
  3714. if (min_free_kbytes < 128)
  3715. min_free_kbytes = 128;
  3716. if (min_free_kbytes > 65536)
  3717. min_free_kbytes = 65536;
  3718. setup_per_zone_pages_min();
  3719. setup_per_zone_lowmem_reserve();
  3720. return 0;
  3721. }
  3722. module_init(init_per_zone_pages_min)
  3723. /*
  3724. * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
  3725. * that we can call two helper functions whenever min_free_kbytes
  3726. * changes.
  3727. */
  3728. int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
  3729. struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
  3730. {
  3731. proc_dointvec(table, write, file, buffer, length, ppos);
  3732. if (write)
  3733. setup_per_zone_pages_min();
  3734. return 0;
  3735. }
  3736. #ifdef CONFIG_NUMA
  3737. int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
  3738. struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
  3739. {
  3740. struct zone *zone;
  3741. int rc;
  3742. rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
  3743. if (rc)
  3744. return rc;
  3745. for_each_zone(zone)
  3746. zone->min_unmapped_pages = (zone->present_pages *
  3747. sysctl_min_unmapped_ratio) / 100;
  3748. return 0;
  3749. }
  3750. int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
  3751. struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
  3752. {
  3753. struct zone *zone;
  3754. int rc;
  3755. rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
  3756. if (rc)
  3757. return rc;
  3758. for_each_zone(zone)
  3759. zone->min_slab_pages = (zone->present_pages *
  3760. sysctl_min_slab_ratio) / 100;
  3761. return 0;
  3762. }
  3763. #endif
  3764. /*
  3765. * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
  3766. * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
  3767. * whenever sysctl_lowmem_reserve_ratio changes.
  3768. *
  3769. * The reserve ratio obviously has absolutely no relation with the
  3770. * pages_min watermarks. The lowmem reserve ratio can only make sense
  3771. * if in function of the boot time zone sizes.
  3772. */
  3773. int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
  3774. struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
  3775. {
  3776. proc_dointvec_minmax(table, write, file, buffer, length, ppos);
  3777. setup_per_zone_lowmem_reserve();
  3778. return 0;
  3779. }
  3780. /*
  3781. * percpu_pagelist_fraction - changes the pcp->high for each zone on each
  3782. * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
  3783. * can have before it gets flushed back to buddy allocator.
  3784. */
  3785. int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
  3786. struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
  3787. {
  3788. struct zone *zone;
  3789. unsigned int cpu;
  3790. int ret;
  3791. ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
  3792. if (!write || (ret == -EINVAL))
  3793. return ret;
  3794. for_each_zone(zone) {
  3795. for_each_online_cpu(cpu) {
  3796. unsigned long high;
  3797. high = zone->present_pages / percpu_pagelist_fraction;
  3798. setup_pagelist_highmark(zone_pcp(zone, cpu), high);
  3799. }
  3800. }
  3801. return 0;
  3802. }
  3803. int hashdist = HASHDIST_DEFAULT;
  3804. #ifdef CONFIG_NUMA
  3805. static int __init set_hashdist(char *str)
  3806. {
  3807. if (!str)
  3808. return 0;
  3809. hashdist = simple_strtoul(str, &str, 0);
  3810. return 1;
  3811. }
  3812. __setup("hashdist=", set_hashdist);
  3813. #endif
  3814. /*
  3815. * allocate a large system hash table from bootmem
  3816. * - it is assumed that the hash table must contain an exact power-of-2
  3817. * quantity of entries
  3818. * - limit is the number of hash buckets, not the total allocation size
  3819. */
  3820. void *__init alloc_large_system_hash(const char *tablename,
  3821. unsigned long bucketsize,
  3822. unsigned long numentries,
  3823. int scale,
  3824. int flags,
  3825. unsigned int *_hash_shift,
  3826. unsigned int *_hash_mask,
  3827. unsigned long limit)
  3828. {
  3829. unsigned long long max = limit;
  3830. unsigned long log2qty, size;
  3831. void *table = NULL;
  3832. /* allow the kernel cmdline to have a say */
  3833. if (!numentries) {
  3834. /* round applicable memory size up to nearest megabyte */
  3835. numentries = nr_kernel_pages;
  3836. numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
  3837. numentries >>= 20 - PAGE_SHIFT;
  3838. numentries <<= 20 - PAGE_SHIFT;
  3839. /* limit to 1 bucket per 2^scale bytes of low memory */
  3840. if (scale > PAGE_SHIFT)
  3841. numentries >>= (scale - PAGE_SHIFT);
  3842. else
  3843. numentries <<= (PAGE_SHIFT - scale);
  3844. /* Make sure we've got at least a 0-order allocation.. */
  3845. if (unlikely((numentries * bucketsize) < PAGE_SIZE))
  3846. numentries = PAGE_SIZE / bucketsize;
  3847. }
  3848. numentries = roundup_pow_of_two(numentries);
  3849. /* limit allocation size to 1/16 total memory by default */
  3850. if (max == 0) {
  3851. max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
  3852. do_div(max, bucketsize);
  3853. }
  3854. if (numentries > max)
  3855. numentries = max;
  3856. log2qty = ilog2(numentries);
  3857. do {
  3858. size = bucketsize << log2qty;
  3859. if (flags & HASH_EARLY)
  3860. table = alloc_bootmem_nopanic(size);
  3861. else if (hashdist)
  3862. table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
  3863. else {
  3864. unsigned long order = get_order(size);
  3865. table = (void*) __get_free_pages(GFP_ATOMIC, order);
  3866. /*
  3867. * If bucketsize is not a power-of-two, we may free
  3868. * some pages at the end of hash table.
  3869. */
  3870. if (table) {
  3871. unsigned long alloc_end = (unsigned long)table +
  3872. (PAGE_SIZE << order);
  3873. unsigned long used = (unsigned long)table +
  3874. PAGE_ALIGN(size);
  3875. split_page(virt_to_page(table), order);
  3876. while (used < alloc_end) {
  3877. free_page(used);
  3878. used += PAGE_SIZE;
  3879. }
  3880. }
  3881. }
  3882. } while (!table && size > PAGE_SIZE && --log2qty);
  3883. if (!table)
  3884. panic("Failed to allocate %s hash table\n", tablename);
  3885. printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n",
  3886. tablename,
  3887. (1U << log2qty),
  3888. ilog2(size) - PAGE_SHIFT,
  3889. size);
  3890. if (_hash_shift)
  3891. *_hash_shift = log2qty;
  3892. if (_hash_mask)
  3893. *_hash_mask = (1 << log2qty) - 1;
  3894. return table;
  3895. }
  3896. #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
  3897. struct page *pfn_to_page(unsigned long pfn)
  3898. {
  3899. return __pfn_to_page(pfn);
  3900. }
  3901. unsigned long page_to_pfn(struct page *page)
  3902. {
  3903. return __page_to_pfn(page);
  3904. }
  3905. EXPORT_SYMBOL(pfn_to_page);
  3906. EXPORT_SYMBOL(page_to_pfn);
  3907. #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
  3908. /* Return a pointer to the bitmap storing bits affecting a block of pages */
  3909. static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
  3910. unsigned long pfn)
  3911. {
  3912. #ifdef CONFIG_SPARSEMEM
  3913. return __pfn_to_section(pfn)->pageblock_flags;
  3914. #else
  3915. return zone->pageblock_flags;
  3916. #endif /* CONFIG_SPARSEMEM */
  3917. }
  3918. static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
  3919. {
  3920. #ifdef CONFIG_SPARSEMEM
  3921. pfn &= (PAGES_PER_SECTION-1);
  3922. return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
  3923. #else
  3924. pfn = pfn - zone->zone_start_pfn;
  3925. return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
  3926. #endif /* CONFIG_SPARSEMEM */
  3927. }
  3928. /**
  3929. * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
  3930. * @page: The page within the block of interest
  3931. * @start_bitidx: The first bit of interest to retrieve
  3932. * @end_bitidx: The last bit of interest
  3933. * returns pageblock_bits flags
  3934. */
  3935. unsigned long get_pageblock_flags_group(struct page *page,
  3936. int start_bitidx, int end_bitidx)
  3937. {
  3938. struct zone *zone;
  3939. unsigned long *bitmap;
  3940. unsigned long pfn, bitidx;
  3941. unsigned long flags = 0;
  3942. unsigned long value = 1;
  3943. zone = page_zone(page);
  3944. pfn = page_to_pfn(page);
  3945. bitmap = get_pageblock_bitmap(zone, pfn);
  3946. bitidx = pfn_to_bitidx(zone, pfn);
  3947. for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
  3948. if (test_bit(bitidx + start_bitidx, bitmap))
  3949. flags |= value;
  3950. return flags;
  3951. }
  3952. /**
  3953. * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
  3954. * @page: The page within the block of interest
  3955. * @start_bitidx: The first bit of interest
  3956. * @end_bitidx: The last bit of interest
  3957. * @flags: The flags to set
  3958. */
  3959. void set_pageblock_flags_group(struct page *page, unsigned long flags,
  3960. int start_bitidx, int end_bitidx)
  3961. {
  3962. struct zone *zone;
  3963. unsigned long *bitmap;
  3964. unsigned long pfn, bitidx;
  3965. unsigned long value = 1;
  3966. zone = page_zone(page);
  3967. pfn = page_to_pfn(page);
  3968. bitmap = get_pageblock_bitmap(zone, pfn);
  3969. bitidx = pfn_to_bitidx(zone, pfn);
  3970. VM_BUG_ON(pfn < zone->zone_start_pfn);
  3971. VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
  3972. for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
  3973. if (flags & value)
  3974. __set_bit(bitidx + start_bitidx, bitmap);
  3975. else
  3976. __clear_bit(bitidx + start_bitidx, bitmap);
  3977. }
  3978. /*
  3979. * This is designed as sub function...plz see page_isolation.c also.
  3980. * set/clear page block's type to be ISOLATE.
  3981. * page allocater never alloc memory from ISOLATE block.
  3982. */
  3983. int set_migratetype_isolate(struct page *page)
  3984. {
  3985. struct zone *zone;
  3986. unsigned long flags;
  3987. int ret = -EBUSY;
  3988. zone = page_zone(page);
  3989. spin_lock_irqsave(&zone->lock, flags);
  3990. /*
  3991. * In future, more migrate types will be able to be isolation target.
  3992. */
  3993. if (get_pageblock_migratetype(page) != MIGRATE_MOVABLE)
  3994. goto out;
  3995. set_pageblock_migratetype(page, MIGRATE_ISOLATE);
  3996. move_freepages_block(zone, page, MIGRATE_ISOLATE);
  3997. ret = 0;
  3998. out:
  3999. spin_unlock_irqrestore(&zone->lock, flags);
  4000. if (!ret)
  4001. drain_all_pages();
  4002. return ret;
  4003. }
  4004. void unset_migratetype_isolate(struct page *page)
  4005. {
  4006. struct zone *zone;
  4007. unsigned long flags;
  4008. zone = page_zone(page);
  4009. spin_lock_irqsave(&zone->lock, flags);
  4010. if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
  4011. goto out;
  4012. set_pageblock_migratetype(page, MIGRATE_MOVABLE);
  4013. move_freepages_block(zone, page, MIGRATE_MOVABLE);
  4014. out:
  4015. spin_unlock_irqrestore(&zone->lock, flags);
  4016. }
  4017. #ifdef CONFIG_MEMORY_HOTREMOVE
  4018. /*
  4019. * All pages in the range must be isolated before calling this.
  4020. */
  4021. void
  4022. __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
  4023. {
  4024. struct page *page;
  4025. struct zone *zone;
  4026. int order, i;
  4027. unsigned long pfn;
  4028. unsigned long flags;
  4029. /* find the first valid pfn */
  4030. for (pfn = start_pfn; pfn < end_pfn; pfn++)
  4031. if (pfn_valid(pfn))
  4032. break;
  4033. if (pfn == end_pfn)
  4034. return;
  4035. zone = page_zone(pfn_to_page(pfn));
  4036. spin_lock_irqsave(&zone->lock, flags);
  4037. pfn = start_pfn;
  4038. while (pfn < end_pfn) {
  4039. if (!pfn_valid(pfn)) {
  4040. pfn++;
  4041. continue;
  4042. }
  4043. page = pfn_to_page(pfn);
  4044. BUG_ON(page_count(page));
  4045. BUG_ON(!PageBuddy(page));
  4046. order = page_order(page);
  4047. #ifdef CONFIG_DEBUG_VM
  4048. printk(KERN_INFO "remove from free list %lx %d %lx\n",
  4049. pfn, 1 << order, end_pfn);
  4050. #endif
  4051. list_del(&page->lru);
  4052. rmv_page_order(page);
  4053. zone->free_area[order].nr_free--;
  4054. __mod_zone_page_state(zone, NR_FREE_PAGES,
  4055. - (1UL << order));
  4056. for (i = 0; i < (1 << order); i++)
  4057. SetPageReserved((page+i));
  4058. pfn += (1 << order);
  4059. }
  4060. spin_unlock_irqrestore(&zone->lock, flags);
  4061. }
  4062. #endif