page_alloc.c 116 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489149014911492149314941495149614971498149915001501150215031504150515061507150815091510151115121513151415151516151715181519152015211522152315241525152615271528152915301531153215331534153515361537153815391540154115421543154415451546154715481549155015511552155315541555155615571558155915601561156215631564156515661567156815691570157115721573157415751576157715781579158015811582158315841585158615871588158915901591159215931594159515961597159815991600160116021603160416051606160716081609161016111612161316141615161616171618161916201621162216231624162516261627162816291630163116321633163416351636163716381639164016411642164316441645164616471648164916501651165216531654165516561657165816591660166116621663166416651666166716681669167016711672167316741675167616771678167916801681168216831684168516861687168816891690169116921693169416951696169716981699170017011702170317041705170617071708170917101711171217131714171517161717171817191720172117221723172417251726172717281729173017311732173317341735173617371738173917401741174217431744174517461747174817491750175117521753175417551756175717581759176017611762176317641765176617671768176917701771177217731774177517761777177817791780178117821783178417851786178717881789179017911792179317941795179617971798179918001801180218031804180518061807180818091810181118121813181418151816181718181819182018211822182318241825182618271828182918301831183218331834183518361837183818391840184118421843184418451846184718481849185018511852185318541855185618571858185918601861186218631864186518661867186818691870187118721873187418751876187718781879188018811882188318841885188618871888188918901891189218931894189518961897189818991900190119021903190419051906190719081909191019111912191319141915191619171918191919201921192219231924192519261927192819291930193119321933193419351936193719381939194019411942194319441945194619471948194919501951195219531954195519561957195819591960196119621963196419651966196719681969197019711972197319741975197619771978197919801981198219831984198519861987198819891990199119921993199419951996199719981999200020012002200320042005200620072008200920102011201220132014201520162017201820192020202120222023202420252026202720282029203020312032203320342035203620372038203920402041204220432044204520462047204820492050205120522053205420552056205720582059206020612062206320642065206620672068206920702071207220732074207520762077207820792080208120822083208420852086208720882089209020912092209320942095209620972098209921002101210221032104210521062107210821092110211121122113211421152116211721182119212021212122212321242125212621272128212921302131213221332134213521362137213821392140214121422143214421452146214721482149215021512152215321542155215621572158215921602161216221632164216521662167216821692170217121722173217421752176217721782179218021812182218321842185218621872188218921902191219221932194219521962197219821992200220122022203220422052206220722082209221022112212221322142215221622172218221922202221222222232224222522262227222822292230223122322233223422352236223722382239224022412242224322442245224622472248224922502251225222532254225522562257225822592260226122622263226422652266226722682269227022712272227322742275227622772278227922802281228222832284228522862287228822892290229122922293229422952296229722982299230023012302230323042305230623072308230923102311231223132314231523162317231823192320232123222323232423252326232723282329233023312332233323342335233623372338233923402341234223432344234523462347234823492350235123522353235423552356235723582359236023612362236323642365236623672368236923702371237223732374237523762377237823792380238123822383238423852386238723882389239023912392239323942395239623972398239924002401240224032404240524062407240824092410241124122413241424152416241724182419242024212422242324242425242624272428242924302431243224332434243524362437243824392440244124422443244424452446244724482449245024512452245324542455245624572458245924602461246224632464246524662467246824692470247124722473247424752476247724782479248024812482248324842485248624872488248924902491249224932494249524962497249824992500250125022503250425052506250725082509251025112512251325142515251625172518251925202521252225232524252525262527252825292530253125322533253425352536253725382539254025412542254325442545254625472548254925502551255225532554255525562557255825592560256125622563256425652566256725682569257025712572257325742575257625772578257925802581258225832584258525862587258825892590259125922593259425952596259725982599260026012602260326042605260626072608260926102611261226132614261526162617261826192620262126222623262426252626262726282629263026312632263326342635263626372638263926402641264226432644264526462647264826492650265126522653265426552656265726582659266026612662266326642665266626672668266926702671267226732674267526762677267826792680268126822683268426852686268726882689269026912692269326942695269626972698269927002701270227032704270527062707270827092710271127122713271427152716271727182719272027212722272327242725272627272728272927302731273227332734273527362737273827392740274127422743274427452746274727482749275027512752275327542755275627572758275927602761276227632764276527662767276827692770277127722773277427752776277727782779278027812782278327842785278627872788278927902791279227932794279527962797279827992800280128022803280428052806280728082809281028112812281328142815281628172818281928202821282228232824282528262827282828292830283128322833283428352836283728382839284028412842284328442845284628472848284928502851285228532854285528562857285828592860286128622863286428652866286728682869287028712872287328742875287628772878287928802881288228832884288528862887288828892890289128922893289428952896289728982899290029012902290329042905290629072908290929102911291229132914291529162917291829192920292129222923292429252926292729282929293029312932293329342935293629372938293929402941294229432944294529462947294829492950295129522953295429552956295729582959296029612962296329642965296629672968296929702971297229732974297529762977297829792980298129822983298429852986298729882989299029912992299329942995299629972998299930003001300230033004300530063007300830093010301130123013301430153016301730183019302030213022302330243025302630273028302930303031303230333034303530363037303830393040304130423043304430453046304730483049305030513052305330543055305630573058305930603061306230633064306530663067306830693070307130723073307430753076307730783079308030813082308330843085308630873088308930903091309230933094309530963097309830993100310131023103310431053106310731083109311031113112311331143115311631173118311931203121312231233124312531263127312831293130313131323133313431353136313731383139314031413142314331443145314631473148314931503151315231533154315531563157315831593160316131623163316431653166316731683169317031713172317331743175317631773178317931803181318231833184318531863187318831893190319131923193319431953196319731983199320032013202320332043205320632073208320932103211321232133214321532163217321832193220322132223223322432253226322732283229323032313232323332343235323632373238323932403241324232433244324532463247324832493250325132523253325432553256325732583259326032613262326332643265326632673268326932703271327232733274327532763277327832793280328132823283328432853286328732883289329032913292329332943295329632973298329933003301330233033304330533063307330833093310331133123313331433153316331733183319332033213322332333243325332633273328332933303331333233333334333533363337333833393340334133423343334433453346334733483349335033513352335333543355335633573358335933603361336233633364336533663367336833693370337133723373337433753376337733783379338033813382338333843385338633873388338933903391339233933394339533963397339833993400340134023403340434053406340734083409341034113412341334143415341634173418341934203421342234233424342534263427342834293430343134323433343434353436343734383439344034413442344334443445344634473448344934503451345234533454345534563457345834593460346134623463346434653466346734683469347034713472347334743475347634773478347934803481348234833484348534863487348834893490349134923493349434953496349734983499350035013502350335043505350635073508350935103511351235133514351535163517351835193520352135223523352435253526352735283529353035313532353335343535353635373538353935403541354235433544354535463547354835493550355135523553355435553556355735583559356035613562356335643565356635673568356935703571357235733574357535763577357835793580358135823583358435853586358735883589359035913592359335943595359635973598359936003601360236033604360536063607360836093610361136123613361436153616361736183619362036213622362336243625362636273628362936303631363236333634363536363637363836393640364136423643364436453646364736483649365036513652365336543655365636573658365936603661366236633664366536663667366836693670367136723673367436753676367736783679368036813682368336843685368636873688368936903691369236933694369536963697369836993700370137023703370437053706370737083709371037113712371337143715371637173718371937203721372237233724372537263727372837293730373137323733373437353736373737383739374037413742374337443745374637473748374937503751375237533754375537563757375837593760376137623763376437653766376737683769377037713772377337743775377637773778377937803781378237833784378537863787378837893790379137923793379437953796379737983799380038013802380338043805380638073808380938103811381238133814381538163817381838193820382138223823382438253826382738283829383038313832383338343835383638373838383938403841384238433844384538463847384838493850385138523853385438553856385738583859386038613862386338643865386638673868386938703871387238733874387538763877387838793880388138823883388438853886388738883889389038913892389338943895389638973898389939003901390239033904390539063907390839093910391139123913391439153916391739183919392039213922392339243925392639273928392939303931393239333934393539363937393839393940394139423943394439453946394739483949395039513952395339543955395639573958395939603961396239633964396539663967396839693970397139723973397439753976397739783979398039813982398339843985398639873988398939903991399239933994399539963997399839994000400140024003400440054006400740084009401040114012401340144015401640174018401940204021402240234024402540264027402840294030403140324033403440354036403740384039404040414042404340444045404640474048404940504051405240534054405540564057405840594060406140624063406440654066406740684069407040714072407340744075407640774078407940804081408240834084408540864087408840894090409140924093409440954096409740984099410041014102410341044105410641074108410941104111411241134114411541164117411841194120412141224123412441254126412741284129413041314132413341344135413641374138413941404141414241434144414541464147414841494150415141524153415441554156415741584159416041614162416341644165416641674168416941704171417241734174417541764177417841794180418141824183418441854186418741884189419041914192419341944195419641974198419942004201420242034204420542064207420842094210421142124213421442154216421742184219422042214222422342244225422642274228422942304231423242334234423542364237423842394240424142424243424442454246424742484249425042514252425342544255425642574258425942604261426242634264426542664267
  1. /*
  2. * linux/mm/page_alloc.c
  3. *
  4. * Manages the free list, the system allocates free pages here.
  5. * Note that kmalloc() lives in slab.c
  6. *
  7. * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
  8. * Swap reorganised 29.12.95, Stephen Tweedie
  9. * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
  10. * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
  11. * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
  12. * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
  13. * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
  14. * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
  15. */
  16. #include <linux/stddef.h>
  17. #include <linux/mm.h>
  18. #include <linux/swap.h>
  19. #include <linux/interrupt.h>
  20. #include <linux/pagemap.h>
  21. #include <linux/bootmem.h>
  22. #include <linux/compiler.h>
  23. #include <linux/kernel.h>
  24. #include <linux/module.h>
  25. #include <linux/suspend.h>
  26. #include <linux/pagevec.h>
  27. #include <linux/blkdev.h>
  28. #include <linux/slab.h>
  29. #include <linux/notifier.h>
  30. #include <linux/topology.h>
  31. #include <linux/sysctl.h>
  32. #include <linux/cpu.h>
  33. #include <linux/cpuset.h>
  34. #include <linux/memory_hotplug.h>
  35. #include <linux/nodemask.h>
  36. #include <linux/vmalloc.h>
  37. #include <linux/mempolicy.h>
  38. #include <linux/stop_machine.h>
  39. #include <linux/sort.h>
  40. #include <linux/pfn.h>
  41. #include <linux/backing-dev.h>
  42. #include <linux/fault-inject.h>
  43. #include <asm/tlbflush.h>
  44. #include <asm/div64.h>
  45. #include "internal.h"
  46. /*
  47. * Array of node states.
  48. */
  49. nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
  50. [N_POSSIBLE] = NODE_MASK_ALL,
  51. [N_ONLINE] = { { [0] = 1UL } },
  52. #ifndef CONFIG_NUMA
  53. [N_NORMAL_MEMORY] = { { [0] = 1UL } },
  54. #ifdef CONFIG_HIGHMEM
  55. [N_HIGH_MEMORY] = { { [0] = 1UL } },
  56. #endif
  57. [N_CPU] = { { [0] = 1UL } },
  58. #endif /* NUMA */
  59. };
  60. EXPORT_SYMBOL(node_states);
  61. unsigned long totalram_pages __read_mostly;
  62. unsigned long totalreserve_pages __read_mostly;
  63. long nr_swap_pages;
  64. int percpu_pagelist_fraction;
  65. static void __free_pages_ok(struct page *page, unsigned int order);
  66. /*
  67. * results with 256, 32 in the lowmem_reserve sysctl:
  68. * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
  69. * 1G machine -> (16M dma, 784M normal, 224M high)
  70. * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
  71. * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
  72. * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
  73. *
  74. * TBD: should special case ZONE_DMA32 machines here - in those we normally
  75. * don't need any ZONE_NORMAL reservation
  76. */
  77. int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
  78. #ifdef CONFIG_ZONE_DMA
  79. 256,
  80. #endif
  81. #ifdef CONFIG_ZONE_DMA32
  82. 256,
  83. #endif
  84. #ifdef CONFIG_HIGHMEM
  85. 32,
  86. #endif
  87. 32,
  88. };
  89. EXPORT_SYMBOL(totalram_pages);
  90. static char * const zone_names[MAX_NR_ZONES] = {
  91. #ifdef CONFIG_ZONE_DMA
  92. "DMA",
  93. #endif
  94. #ifdef CONFIG_ZONE_DMA32
  95. "DMA32",
  96. #endif
  97. "Normal",
  98. #ifdef CONFIG_HIGHMEM
  99. "HighMem",
  100. #endif
  101. "Movable",
  102. };
  103. int min_free_kbytes = 1024;
  104. unsigned long __meminitdata nr_kernel_pages;
  105. unsigned long __meminitdata nr_all_pages;
  106. static unsigned long __meminitdata dma_reserve;
  107. #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
  108. /*
  109. * MAX_ACTIVE_REGIONS determines the maxmimum number of distinct
  110. * ranges of memory (RAM) that may be registered with add_active_range().
  111. * Ranges passed to add_active_range() will be merged if possible
  112. * so the number of times add_active_range() can be called is
  113. * related to the number of nodes and the number of holes
  114. */
  115. #ifdef CONFIG_MAX_ACTIVE_REGIONS
  116. /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
  117. #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
  118. #else
  119. #if MAX_NUMNODES >= 32
  120. /* If there can be many nodes, allow up to 50 holes per node */
  121. #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
  122. #else
  123. /* By default, allow up to 256 distinct regions */
  124. #define MAX_ACTIVE_REGIONS 256
  125. #endif
  126. #endif
  127. static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS];
  128. static int __meminitdata nr_nodemap_entries;
  129. static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
  130. static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
  131. #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
  132. static unsigned long __meminitdata node_boundary_start_pfn[MAX_NUMNODES];
  133. static unsigned long __meminitdata node_boundary_end_pfn[MAX_NUMNODES];
  134. #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
  135. unsigned long __initdata required_kernelcore;
  136. unsigned long __initdata required_movablecore;
  137. unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
  138. /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
  139. int movable_zone;
  140. EXPORT_SYMBOL(movable_zone);
  141. #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
  142. #if MAX_NUMNODES > 1
  143. int nr_node_ids __read_mostly = MAX_NUMNODES;
  144. EXPORT_SYMBOL(nr_node_ids);
  145. #endif
  146. #ifdef CONFIG_PAGE_GROUP_BY_MOBILITY
  147. static inline int get_pageblock_migratetype(struct page *page)
  148. {
  149. return get_pageblock_flags_group(page, PB_migrate, PB_migrate_end);
  150. }
  151. static void set_pageblock_migratetype(struct page *page, int migratetype)
  152. {
  153. set_pageblock_flags_group(page, (unsigned long)migratetype,
  154. PB_migrate, PB_migrate_end);
  155. }
  156. static inline int gfpflags_to_migratetype(gfp_t gfp_flags)
  157. {
  158. return ((gfp_flags & __GFP_MOVABLE) != 0);
  159. }
  160. #else
  161. static inline int get_pageblock_migratetype(struct page *page)
  162. {
  163. return MIGRATE_UNMOVABLE;
  164. }
  165. static void set_pageblock_migratetype(struct page *page, int migratetype)
  166. {
  167. }
  168. static inline int gfpflags_to_migratetype(gfp_t gfp_flags)
  169. {
  170. return MIGRATE_UNMOVABLE;
  171. }
  172. #endif /* CONFIG_PAGE_GROUP_BY_MOBILITY */
  173. #ifdef CONFIG_DEBUG_VM
  174. static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
  175. {
  176. int ret = 0;
  177. unsigned seq;
  178. unsigned long pfn = page_to_pfn(page);
  179. do {
  180. seq = zone_span_seqbegin(zone);
  181. if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
  182. ret = 1;
  183. else if (pfn < zone->zone_start_pfn)
  184. ret = 1;
  185. } while (zone_span_seqretry(zone, seq));
  186. return ret;
  187. }
  188. static int page_is_consistent(struct zone *zone, struct page *page)
  189. {
  190. if (!pfn_valid_within(page_to_pfn(page)))
  191. return 0;
  192. if (zone != page_zone(page))
  193. return 0;
  194. return 1;
  195. }
  196. /*
  197. * Temporary debugging check for pages not lying within a given zone.
  198. */
  199. static int bad_range(struct zone *zone, struct page *page)
  200. {
  201. if (page_outside_zone_boundaries(zone, page))
  202. return 1;
  203. if (!page_is_consistent(zone, page))
  204. return 1;
  205. return 0;
  206. }
  207. #else
  208. static inline int bad_range(struct zone *zone, struct page *page)
  209. {
  210. return 0;
  211. }
  212. #endif
  213. static void bad_page(struct page *page)
  214. {
  215. printk(KERN_EMERG "Bad page state in process '%s'\n"
  216. KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
  217. KERN_EMERG "Trying to fix it up, but a reboot is needed\n"
  218. KERN_EMERG "Backtrace:\n",
  219. current->comm, page, (int)(2*sizeof(unsigned long)),
  220. (unsigned long)page->flags, page->mapping,
  221. page_mapcount(page), page_count(page));
  222. dump_stack();
  223. page->flags &= ~(1 << PG_lru |
  224. 1 << PG_private |
  225. 1 << PG_locked |
  226. 1 << PG_active |
  227. 1 << PG_dirty |
  228. 1 << PG_reclaim |
  229. 1 << PG_slab |
  230. 1 << PG_swapcache |
  231. 1 << PG_writeback |
  232. 1 << PG_buddy );
  233. set_page_count(page, 0);
  234. reset_page_mapcount(page);
  235. page->mapping = NULL;
  236. add_taint(TAINT_BAD_PAGE);
  237. }
  238. /*
  239. * Higher-order pages are called "compound pages". They are structured thusly:
  240. *
  241. * The first PAGE_SIZE page is called the "head page".
  242. *
  243. * The remaining PAGE_SIZE pages are called "tail pages".
  244. *
  245. * All pages have PG_compound set. All pages have their ->private pointing at
  246. * the head page (even the head page has this).
  247. *
  248. * The first tail page's ->lru.next holds the address of the compound page's
  249. * put_page() function. Its ->lru.prev holds the order of allocation.
  250. * This usage means that zero-order pages may not be compound.
  251. */
  252. static void free_compound_page(struct page *page)
  253. {
  254. __free_pages_ok(page, compound_order(page));
  255. }
  256. static void prep_compound_page(struct page *page, unsigned long order)
  257. {
  258. int i;
  259. int nr_pages = 1 << order;
  260. set_compound_page_dtor(page, free_compound_page);
  261. set_compound_order(page, order);
  262. __SetPageHead(page);
  263. for (i = 1; i < nr_pages; i++) {
  264. struct page *p = page + i;
  265. __SetPageTail(p);
  266. p->first_page = page;
  267. }
  268. }
  269. static void destroy_compound_page(struct page *page, unsigned long order)
  270. {
  271. int i;
  272. int nr_pages = 1 << order;
  273. if (unlikely(compound_order(page) != order))
  274. bad_page(page);
  275. if (unlikely(!PageHead(page)))
  276. bad_page(page);
  277. __ClearPageHead(page);
  278. for (i = 1; i < nr_pages; i++) {
  279. struct page *p = page + i;
  280. if (unlikely(!PageTail(p) |
  281. (p->first_page != page)))
  282. bad_page(page);
  283. __ClearPageTail(p);
  284. }
  285. }
  286. static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
  287. {
  288. int i;
  289. VM_BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM);
  290. /*
  291. * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
  292. * and __GFP_HIGHMEM from hard or soft interrupt context.
  293. */
  294. VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
  295. for (i = 0; i < (1 << order); i++)
  296. clear_highpage(page + i);
  297. }
  298. /*
  299. * function for dealing with page's order in buddy system.
  300. * zone->lock is already acquired when we use these.
  301. * So, we don't need atomic page->flags operations here.
  302. */
  303. static inline unsigned long page_order(struct page *page)
  304. {
  305. return page_private(page);
  306. }
  307. static inline void set_page_order(struct page *page, int order)
  308. {
  309. set_page_private(page, order);
  310. __SetPageBuddy(page);
  311. }
  312. static inline void rmv_page_order(struct page *page)
  313. {
  314. __ClearPageBuddy(page);
  315. set_page_private(page, 0);
  316. }
  317. /*
  318. * Locate the struct page for both the matching buddy in our
  319. * pair (buddy1) and the combined O(n+1) page they form (page).
  320. *
  321. * 1) Any buddy B1 will have an order O twin B2 which satisfies
  322. * the following equation:
  323. * B2 = B1 ^ (1 << O)
  324. * For example, if the starting buddy (buddy2) is #8 its order
  325. * 1 buddy is #10:
  326. * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
  327. *
  328. * 2) Any buddy B will have an order O+1 parent P which
  329. * satisfies the following equation:
  330. * P = B & ~(1 << O)
  331. *
  332. * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
  333. */
  334. static inline struct page *
  335. __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
  336. {
  337. unsigned long buddy_idx = page_idx ^ (1 << order);
  338. return page + (buddy_idx - page_idx);
  339. }
  340. static inline unsigned long
  341. __find_combined_index(unsigned long page_idx, unsigned int order)
  342. {
  343. return (page_idx & ~(1 << order));
  344. }
  345. /*
  346. * This function checks whether a page is free && is the buddy
  347. * we can do coalesce a page and its buddy if
  348. * (a) the buddy is not in a hole &&
  349. * (b) the buddy is in the buddy system &&
  350. * (c) a page and its buddy have the same order &&
  351. * (d) a page and its buddy are in the same zone.
  352. *
  353. * For recording whether a page is in the buddy system, we use PG_buddy.
  354. * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
  355. *
  356. * For recording page's order, we use page_private(page).
  357. */
  358. static inline int page_is_buddy(struct page *page, struct page *buddy,
  359. int order)
  360. {
  361. if (!pfn_valid_within(page_to_pfn(buddy)))
  362. return 0;
  363. if (page_zone_id(page) != page_zone_id(buddy))
  364. return 0;
  365. if (PageBuddy(buddy) && page_order(buddy) == order) {
  366. BUG_ON(page_count(buddy) != 0);
  367. return 1;
  368. }
  369. return 0;
  370. }
  371. /*
  372. * Freeing function for a buddy system allocator.
  373. *
  374. * The concept of a buddy system is to maintain direct-mapped table
  375. * (containing bit values) for memory blocks of various "orders".
  376. * The bottom level table contains the map for the smallest allocatable
  377. * units of memory (here, pages), and each level above it describes
  378. * pairs of units from the levels below, hence, "buddies".
  379. * At a high level, all that happens here is marking the table entry
  380. * at the bottom level available, and propagating the changes upward
  381. * as necessary, plus some accounting needed to play nicely with other
  382. * parts of the VM system.
  383. * At each level, we keep a list of pages, which are heads of continuous
  384. * free pages of length of (1 << order) and marked with PG_buddy. Page's
  385. * order is recorded in page_private(page) field.
  386. * So when we are allocating or freeing one, we can derive the state of the
  387. * other. That is, if we allocate a small block, and both were
  388. * free, the remainder of the region must be split into blocks.
  389. * If a block is freed, and its buddy is also free, then this
  390. * triggers coalescing into a block of larger size.
  391. *
  392. * -- wli
  393. */
  394. static inline void __free_one_page(struct page *page,
  395. struct zone *zone, unsigned int order)
  396. {
  397. unsigned long page_idx;
  398. int order_size = 1 << order;
  399. int migratetype = get_pageblock_migratetype(page);
  400. if (unlikely(PageCompound(page)))
  401. destroy_compound_page(page, order);
  402. page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
  403. VM_BUG_ON(page_idx & (order_size - 1));
  404. VM_BUG_ON(bad_range(zone, page));
  405. __mod_zone_page_state(zone, NR_FREE_PAGES, order_size);
  406. while (order < MAX_ORDER-1) {
  407. unsigned long combined_idx;
  408. struct page *buddy;
  409. buddy = __page_find_buddy(page, page_idx, order);
  410. if (!page_is_buddy(page, buddy, order))
  411. break; /* Move the buddy up one level. */
  412. list_del(&buddy->lru);
  413. zone->free_area[order].nr_free--;
  414. rmv_page_order(buddy);
  415. combined_idx = __find_combined_index(page_idx, order);
  416. page = page + (combined_idx - page_idx);
  417. page_idx = combined_idx;
  418. order++;
  419. }
  420. set_page_order(page, order);
  421. list_add(&page->lru,
  422. &zone->free_area[order].free_list[migratetype]);
  423. zone->free_area[order].nr_free++;
  424. }
  425. static inline int free_pages_check(struct page *page)
  426. {
  427. if (unlikely(page_mapcount(page) |
  428. (page->mapping != NULL) |
  429. (page_count(page) != 0) |
  430. (page->flags & (
  431. 1 << PG_lru |
  432. 1 << PG_private |
  433. 1 << PG_locked |
  434. 1 << PG_active |
  435. 1 << PG_slab |
  436. 1 << PG_swapcache |
  437. 1 << PG_writeback |
  438. 1 << PG_reserved |
  439. 1 << PG_buddy ))))
  440. bad_page(page);
  441. if (PageDirty(page))
  442. __ClearPageDirty(page);
  443. /*
  444. * For now, we report if PG_reserved was found set, but do not
  445. * clear it, and do not free the page. But we shall soon need
  446. * to do more, for when the ZERO_PAGE count wraps negative.
  447. */
  448. return PageReserved(page);
  449. }
  450. /*
  451. * Frees a list of pages.
  452. * Assumes all pages on list are in same zone, and of same order.
  453. * count is the number of pages to free.
  454. *
  455. * If the zone was previously in an "all pages pinned" state then look to
  456. * see if this freeing clears that state.
  457. *
  458. * And clear the zone's pages_scanned counter, to hold off the "all pages are
  459. * pinned" detection logic.
  460. */
  461. static void free_pages_bulk(struct zone *zone, int count,
  462. struct list_head *list, int order)
  463. {
  464. spin_lock(&zone->lock);
  465. zone->all_unreclaimable = 0;
  466. zone->pages_scanned = 0;
  467. while (count--) {
  468. struct page *page;
  469. VM_BUG_ON(list_empty(list));
  470. page = list_entry(list->prev, struct page, lru);
  471. /* have to delete it as __free_one_page list manipulates */
  472. list_del(&page->lru);
  473. __free_one_page(page, zone, order);
  474. }
  475. spin_unlock(&zone->lock);
  476. }
  477. static void free_one_page(struct zone *zone, struct page *page, int order)
  478. {
  479. spin_lock(&zone->lock);
  480. zone->all_unreclaimable = 0;
  481. zone->pages_scanned = 0;
  482. __free_one_page(page, zone, order);
  483. spin_unlock(&zone->lock);
  484. }
  485. static void __free_pages_ok(struct page *page, unsigned int order)
  486. {
  487. unsigned long flags;
  488. int i;
  489. int reserved = 0;
  490. for (i = 0 ; i < (1 << order) ; ++i)
  491. reserved += free_pages_check(page + i);
  492. if (reserved)
  493. return;
  494. if (!PageHighMem(page))
  495. debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
  496. arch_free_page(page, order);
  497. kernel_map_pages(page, 1 << order, 0);
  498. local_irq_save(flags);
  499. __count_vm_events(PGFREE, 1 << order);
  500. free_one_page(page_zone(page), page, order);
  501. local_irq_restore(flags);
  502. }
  503. /*
  504. * permit the bootmem allocator to evade page validation on high-order frees
  505. */
  506. void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order)
  507. {
  508. if (order == 0) {
  509. __ClearPageReserved(page);
  510. set_page_count(page, 0);
  511. set_page_refcounted(page);
  512. __free_page(page);
  513. } else {
  514. int loop;
  515. prefetchw(page);
  516. for (loop = 0; loop < BITS_PER_LONG; loop++) {
  517. struct page *p = &page[loop];
  518. if (loop + 1 < BITS_PER_LONG)
  519. prefetchw(p + 1);
  520. __ClearPageReserved(p);
  521. set_page_count(p, 0);
  522. }
  523. set_page_refcounted(page);
  524. __free_pages(page, order);
  525. }
  526. }
  527. /*
  528. * The order of subdivision here is critical for the IO subsystem.
  529. * Please do not alter this order without good reasons and regression
  530. * testing. Specifically, as large blocks of memory are subdivided,
  531. * the order in which smaller blocks are delivered depends on the order
  532. * they're subdivided in this function. This is the primary factor
  533. * influencing the order in which pages are delivered to the IO
  534. * subsystem according to empirical testing, and this is also justified
  535. * by considering the behavior of a buddy system containing a single
  536. * large block of memory acted on by a series of small allocations.
  537. * This behavior is a critical factor in sglist merging's success.
  538. *
  539. * -- wli
  540. */
  541. static inline void expand(struct zone *zone, struct page *page,
  542. int low, int high, struct free_area *area,
  543. int migratetype)
  544. {
  545. unsigned long size = 1 << high;
  546. while (high > low) {
  547. area--;
  548. high--;
  549. size >>= 1;
  550. VM_BUG_ON(bad_range(zone, &page[size]));
  551. list_add(&page[size].lru, &area->free_list[migratetype]);
  552. area->nr_free++;
  553. set_page_order(&page[size], high);
  554. }
  555. }
  556. /*
  557. * This page is about to be returned from the page allocator
  558. */
  559. static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
  560. {
  561. if (unlikely(page_mapcount(page) |
  562. (page->mapping != NULL) |
  563. (page_count(page) != 0) |
  564. (page->flags & (
  565. 1 << PG_lru |
  566. 1 << PG_private |
  567. 1 << PG_locked |
  568. 1 << PG_active |
  569. 1 << PG_dirty |
  570. 1 << PG_slab |
  571. 1 << PG_swapcache |
  572. 1 << PG_writeback |
  573. 1 << PG_reserved |
  574. 1 << PG_buddy ))))
  575. bad_page(page);
  576. /*
  577. * For now, we report if PG_reserved was found set, but do not
  578. * clear it, and do not allocate the page: as a safety net.
  579. */
  580. if (PageReserved(page))
  581. return 1;
  582. page->flags &= ~(1 << PG_uptodate | 1 << PG_error | 1 << PG_readahead |
  583. 1 << PG_referenced | 1 << PG_arch_1 |
  584. 1 << PG_owner_priv_1 | 1 << PG_mappedtodisk);
  585. set_page_private(page, 0);
  586. set_page_refcounted(page);
  587. arch_alloc_page(page, order);
  588. kernel_map_pages(page, 1 << order, 1);
  589. if (gfp_flags & __GFP_ZERO)
  590. prep_zero_page(page, order, gfp_flags);
  591. if (order && (gfp_flags & __GFP_COMP))
  592. prep_compound_page(page, order);
  593. return 0;
  594. }
  595. #ifdef CONFIG_PAGE_GROUP_BY_MOBILITY
  596. /*
  597. * This array describes the order lists are fallen back to when
  598. * the free lists for the desirable migrate type are depleted
  599. */
  600. static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
  601. [MIGRATE_UNMOVABLE] = { MIGRATE_MOVABLE },
  602. [MIGRATE_MOVABLE] = { MIGRATE_UNMOVABLE },
  603. };
  604. /* Remove an element from the buddy allocator from the fallback list */
  605. static struct page *__rmqueue_fallback(struct zone *zone, int order,
  606. int start_migratetype)
  607. {
  608. struct free_area * area;
  609. int current_order;
  610. struct page *page;
  611. int migratetype, i;
  612. /* Find the largest possible block of pages in the other list */
  613. for (current_order = MAX_ORDER-1; current_order >= order;
  614. --current_order) {
  615. for (i = 0; i < MIGRATE_TYPES - 1; i++) {
  616. migratetype = fallbacks[start_migratetype][i];
  617. area = &(zone->free_area[current_order]);
  618. if (list_empty(&area->free_list[migratetype]))
  619. continue;
  620. page = list_entry(area->free_list[migratetype].next,
  621. struct page, lru);
  622. area->nr_free--;
  623. /*
  624. * If breaking a large block of pages, place the buddies
  625. * on the preferred allocation list
  626. */
  627. if (unlikely(current_order >= MAX_ORDER / 2))
  628. migratetype = start_migratetype;
  629. /* Remove the page from the freelists */
  630. list_del(&page->lru);
  631. rmv_page_order(page);
  632. __mod_zone_page_state(zone, NR_FREE_PAGES,
  633. -(1UL << order));
  634. if (current_order == MAX_ORDER - 1)
  635. set_pageblock_migratetype(page,
  636. start_migratetype);
  637. expand(zone, page, order, current_order, area, migratetype);
  638. return page;
  639. }
  640. }
  641. return NULL;
  642. }
  643. #else
  644. static struct page *__rmqueue_fallback(struct zone *zone, int order,
  645. int start_migratetype)
  646. {
  647. return NULL;
  648. }
  649. #endif /* CONFIG_PAGE_GROUP_BY_MOBILITY */
  650. /*
  651. * Do the hard work of removing an element from the buddy allocator.
  652. * Call me with the zone->lock already held.
  653. */
  654. static struct page *__rmqueue(struct zone *zone, unsigned int order,
  655. int migratetype)
  656. {
  657. struct free_area * area;
  658. unsigned int current_order;
  659. struct page *page;
  660. /* Find a page of the appropriate size in the preferred list */
  661. for (current_order = order; current_order < MAX_ORDER; ++current_order) {
  662. area = &(zone->free_area[current_order]);
  663. if (list_empty(&area->free_list[migratetype]))
  664. continue;
  665. page = list_entry(area->free_list[migratetype].next,
  666. struct page, lru);
  667. list_del(&page->lru);
  668. rmv_page_order(page);
  669. area->nr_free--;
  670. __mod_zone_page_state(zone, NR_FREE_PAGES, - (1UL << order));
  671. expand(zone, page, order, current_order, area, migratetype);
  672. goto got_page;
  673. }
  674. page = __rmqueue_fallback(zone, order, migratetype);
  675. got_page:
  676. return page;
  677. }
  678. /*
  679. * Obtain a specified number of elements from the buddy allocator, all under
  680. * a single hold of the lock, for efficiency. Add them to the supplied list.
  681. * Returns the number of new pages which were placed at *list.
  682. */
  683. static int rmqueue_bulk(struct zone *zone, unsigned int order,
  684. unsigned long count, struct list_head *list,
  685. int migratetype)
  686. {
  687. int i;
  688. spin_lock(&zone->lock);
  689. for (i = 0; i < count; ++i) {
  690. struct page *page = __rmqueue(zone, order, migratetype);
  691. if (unlikely(page == NULL))
  692. break;
  693. list_add(&page->lru, list);
  694. set_page_private(page, migratetype);
  695. }
  696. spin_unlock(&zone->lock);
  697. return i;
  698. }
  699. #ifdef CONFIG_NUMA
  700. /*
  701. * Called from the vmstat counter updater to drain pagesets of this
  702. * currently executing processor on remote nodes after they have
  703. * expired.
  704. *
  705. * Note that this function must be called with the thread pinned to
  706. * a single processor.
  707. */
  708. void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
  709. {
  710. unsigned long flags;
  711. int to_drain;
  712. local_irq_save(flags);
  713. if (pcp->count >= pcp->batch)
  714. to_drain = pcp->batch;
  715. else
  716. to_drain = pcp->count;
  717. free_pages_bulk(zone, to_drain, &pcp->list, 0);
  718. pcp->count -= to_drain;
  719. local_irq_restore(flags);
  720. }
  721. #endif
  722. static void __drain_pages(unsigned int cpu)
  723. {
  724. unsigned long flags;
  725. struct zone *zone;
  726. int i;
  727. for_each_zone(zone) {
  728. struct per_cpu_pageset *pset;
  729. if (!populated_zone(zone))
  730. continue;
  731. pset = zone_pcp(zone, cpu);
  732. for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
  733. struct per_cpu_pages *pcp;
  734. pcp = &pset->pcp[i];
  735. local_irq_save(flags);
  736. free_pages_bulk(zone, pcp->count, &pcp->list, 0);
  737. pcp->count = 0;
  738. local_irq_restore(flags);
  739. }
  740. }
  741. }
  742. #ifdef CONFIG_HIBERNATION
  743. void mark_free_pages(struct zone *zone)
  744. {
  745. unsigned long pfn, max_zone_pfn;
  746. unsigned long flags;
  747. int order, t;
  748. struct list_head *curr;
  749. if (!zone->spanned_pages)
  750. return;
  751. spin_lock_irqsave(&zone->lock, flags);
  752. max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
  753. for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
  754. if (pfn_valid(pfn)) {
  755. struct page *page = pfn_to_page(pfn);
  756. if (!swsusp_page_is_forbidden(page))
  757. swsusp_unset_page_free(page);
  758. }
  759. for_each_migratetype_order(order, t) {
  760. list_for_each(curr, &zone->free_area[order].free_list[t]) {
  761. unsigned long i;
  762. pfn = page_to_pfn(list_entry(curr, struct page, lru));
  763. for (i = 0; i < (1UL << order); i++)
  764. swsusp_set_page_free(pfn_to_page(pfn + i));
  765. }
  766. }
  767. spin_unlock_irqrestore(&zone->lock, flags);
  768. }
  769. #endif /* CONFIG_PM */
  770. #if defined(CONFIG_HIBERNATION) || defined(CONFIG_PAGE_GROUP_BY_MOBILITY)
  771. /*
  772. * Spill all of this CPU's per-cpu pages back into the buddy allocator.
  773. */
  774. void drain_local_pages(void)
  775. {
  776. unsigned long flags;
  777. local_irq_save(flags);
  778. __drain_pages(smp_processor_id());
  779. local_irq_restore(flags);
  780. }
  781. void smp_drain_local_pages(void *arg)
  782. {
  783. drain_local_pages();
  784. }
  785. /*
  786. * Spill all the per-cpu pages from all CPUs back into the buddy allocator
  787. */
  788. void drain_all_local_pages(void)
  789. {
  790. unsigned long flags;
  791. local_irq_save(flags);
  792. __drain_pages(smp_processor_id());
  793. local_irq_restore(flags);
  794. smp_call_function(smp_drain_local_pages, NULL, 0, 1);
  795. }
  796. #else
  797. void drain_all_local_pages(void) {}
  798. #endif /* CONFIG_HIBERNATION || CONFIG_PAGE_GROUP_BY_MOBILITY */
  799. /*
  800. * Free a 0-order page
  801. */
  802. static void fastcall free_hot_cold_page(struct page *page, int cold)
  803. {
  804. struct zone *zone = page_zone(page);
  805. struct per_cpu_pages *pcp;
  806. unsigned long flags;
  807. if (PageAnon(page))
  808. page->mapping = NULL;
  809. if (free_pages_check(page))
  810. return;
  811. if (!PageHighMem(page))
  812. debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
  813. arch_free_page(page, 0);
  814. kernel_map_pages(page, 1, 0);
  815. pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
  816. local_irq_save(flags);
  817. __count_vm_event(PGFREE);
  818. list_add(&page->lru, &pcp->list);
  819. set_page_private(page, get_pageblock_migratetype(page));
  820. pcp->count++;
  821. if (pcp->count >= pcp->high) {
  822. free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
  823. pcp->count -= pcp->batch;
  824. }
  825. local_irq_restore(flags);
  826. put_cpu();
  827. }
  828. void fastcall free_hot_page(struct page *page)
  829. {
  830. free_hot_cold_page(page, 0);
  831. }
  832. void fastcall free_cold_page(struct page *page)
  833. {
  834. free_hot_cold_page(page, 1);
  835. }
  836. /*
  837. * split_page takes a non-compound higher-order page, and splits it into
  838. * n (1<<order) sub-pages: page[0..n]
  839. * Each sub-page must be freed individually.
  840. *
  841. * Note: this is probably too low level an operation for use in drivers.
  842. * Please consult with lkml before using this in your driver.
  843. */
  844. void split_page(struct page *page, unsigned int order)
  845. {
  846. int i;
  847. VM_BUG_ON(PageCompound(page));
  848. VM_BUG_ON(!page_count(page));
  849. for (i = 1; i < (1 << order); i++)
  850. set_page_refcounted(page + i);
  851. }
  852. /*
  853. * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
  854. * we cheat by calling it from here, in the order > 0 path. Saves a branch
  855. * or two.
  856. */
  857. static struct page *buffered_rmqueue(struct zonelist *zonelist,
  858. struct zone *zone, int order, gfp_t gfp_flags)
  859. {
  860. unsigned long flags;
  861. struct page *page;
  862. int cold = !!(gfp_flags & __GFP_COLD);
  863. int cpu;
  864. int migratetype = gfpflags_to_migratetype(gfp_flags);
  865. again:
  866. cpu = get_cpu();
  867. if (likely(order == 0)) {
  868. struct per_cpu_pages *pcp;
  869. pcp = &zone_pcp(zone, cpu)->pcp[cold];
  870. local_irq_save(flags);
  871. if (!pcp->count) {
  872. pcp->count = rmqueue_bulk(zone, 0,
  873. pcp->batch, &pcp->list, migratetype);
  874. if (unlikely(!pcp->count))
  875. goto failed;
  876. }
  877. #ifdef CONFIG_PAGE_GROUP_BY_MOBILITY
  878. /* Find a page of the appropriate migrate type */
  879. list_for_each_entry(page, &pcp->list, lru)
  880. if (page_private(page) == migratetype)
  881. break;
  882. /* Allocate more to the pcp list if necessary */
  883. if (unlikely(&page->lru == &pcp->list)) {
  884. pcp->count += rmqueue_bulk(zone, 0,
  885. pcp->batch, &pcp->list, migratetype);
  886. page = list_entry(pcp->list.next, struct page, lru);
  887. }
  888. #else
  889. page = list_entry(pcp->list.next, struct page, lru);
  890. #endif /* CONFIG_PAGE_GROUP_BY_MOBILITY */
  891. list_del(&page->lru);
  892. pcp->count--;
  893. } else {
  894. spin_lock_irqsave(&zone->lock, flags);
  895. page = __rmqueue(zone, order, migratetype);
  896. spin_unlock(&zone->lock);
  897. if (!page)
  898. goto failed;
  899. }
  900. __count_zone_vm_events(PGALLOC, zone, 1 << order);
  901. zone_statistics(zonelist, zone);
  902. local_irq_restore(flags);
  903. put_cpu();
  904. VM_BUG_ON(bad_range(zone, page));
  905. if (prep_new_page(page, order, gfp_flags))
  906. goto again;
  907. return page;
  908. failed:
  909. local_irq_restore(flags);
  910. put_cpu();
  911. return NULL;
  912. }
  913. #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
  914. #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
  915. #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
  916. #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
  917. #define ALLOC_HARDER 0x10 /* try to alloc harder */
  918. #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
  919. #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
  920. #ifdef CONFIG_FAIL_PAGE_ALLOC
  921. static struct fail_page_alloc_attr {
  922. struct fault_attr attr;
  923. u32 ignore_gfp_highmem;
  924. u32 ignore_gfp_wait;
  925. u32 min_order;
  926. #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
  927. struct dentry *ignore_gfp_highmem_file;
  928. struct dentry *ignore_gfp_wait_file;
  929. struct dentry *min_order_file;
  930. #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
  931. } fail_page_alloc = {
  932. .attr = FAULT_ATTR_INITIALIZER,
  933. .ignore_gfp_wait = 1,
  934. .ignore_gfp_highmem = 1,
  935. .min_order = 1,
  936. };
  937. static int __init setup_fail_page_alloc(char *str)
  938. {
  939. return setup_fault_attr(&fail_page_alloc.attr, str);
  940. }
  941. __setup("fail_page_alloc=", setup_fail_page_alloc);
  942. static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
  943. {
  944. if (order < fail_page_alloc.min_order)
  945. return 0;
  946. if (gfp_mask & __GFP_NOFAIL)
  947. return 0;
  948. if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
  949. return 0;
  950. if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
  951. return 0;
  952. return should_fail(&fail_page_alloc.attr, 1 << order);
  953. }
  954. #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
  955. static int __init fail_page_alloc_debugfs(void)
  956. {
  957. mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
  958. struct dentry *dir;
  959. int err;
  960. err = init_fault_attr_dentries(&fail_page_alloc.attr,
  961. "fail_page_alloc");
  962. if (err)
  963. return err;
  964. dir = fail_page_alloc.attr.dentries.dir;
  965. fail_page_alloc.ignore_gfp_wait_file =
  966. debugfs_create_bool("ignore-gfp-wait", mode, dir,
  967. &fail_page_alloc.ignore_gfp_wait);
  968. fail_page_alloc.ignore_gfp_highmem_file =
  969. debugfs_create_bool("ignore-gfp-highmem", mode, dir,
  970. &fail_page_alloc.ignore_gfp_highmem);
  971. fail_page_alloc.min_order_file =
  972. debugfs_create_u32("min-order", mode, dir,
  973. &fail_page_alloc.min_order);
  974. if (!fail_page_alloc.ignore_gfp_wait_file ||
  975. !fail_page_alloc.ignore_gfp_highmem_file ||
  976. !fail_page_alloc.min_order_file) {
  977. err = -ENOMEM;
  978. debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
  979. debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
  980. debugfs_remove(fail_page_alloc.min_order_file);
  981. cleanup_fault_attr_dentries(&fail_page_alloc.attr);
  982. }
  983. return err;
  984. }
  985. late_initcall(fail_page_alloc_debugfs);
  986. #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
  987. #else /* CONFIG_FAIL_PAGE_ALLOC */
  988. static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
  989. {
  990. return 0;
  991. }
  992. #endif /* CONFIG_FAIL_PAGE_ALLOC */
  993. /*
  994. * Return 1 if free pages are above 'mark'. This takes into account the order
  995. * of the allocation.
  996. */
  997. int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
  998. int classzone_idx, int alloc_flags)
  999. {
  1000. /* free_pages my go negative - that's OK */
  1001. long min = mark;
  1002. long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1;
  1003. int o;
  1004. if (alloc_flags & ALLOC_HIGH)
  1005. min -= min / 2;
  1006. if (alloc_flags & ALLOC_HARDER)
  1007. min -= min / 4;
  1008. if (free_pages <= min + z->lowmem_reserve[classzone_idx])
  1009. return 0;
  1010. for (o = 0; o < order; o++) {
  1011. /* At the next order, this order's pages become unavailable */
  1012. free_pages -= z->free_area[o].nr_free << o;
  1013. /* Require fewer higher order pages to be free */
  1014. min >>= 1;
  1015. if (free_pages <= min)
  1016. return 0;
  1017. }
  1018. return 1;
  1019. }
  1020. #ifdef CONFIG_NUMA
  1021. /*
  1022. * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
  1023. * skip over zones that are not allowed by the cpuset, or that have
  1024. * been recently (in last second) found to be nearly full. See further
  1025. * comments in mmzone.h. Reduces cache footprint of zonelist scans
  1026. * that have to skip over alot of full or unallowed zones.
  1027. *
  1028. * If the zonelist cache is present in the passed in zonelist, then
  1029. * returns a pointer to the allowed node mask (either the current
  1030. * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
  1031. *
  1032. * If the zonelist cache is not available for this zonelist, does
  1033. * nothing and returns NULL.
  1034. *
  1035. * If the fullzones BITMAP in the zonelist cache is stale (more than
  1036. * a second since last zap'd) then we zap it out (clear its bits.)
  1037. *
  1038. * We hold off even calling zlc_setup, until after we've checked the
  1039. * first zone in the zonelist, on the theory that most allocations will
  1040. * be satisfied from that first zone, so best to examine that zone as
  1041. * quickly as we can.
  1042. */
  1043. static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
  1044. {
  1045. struct zonelist_cache *zlc; /* cached zonelist speedup info */
  1046. nodemask_t *allowednodes; /* zonelist_cache approximation */
  1047. zlc = zonelist->zlcache_ptr;
  1048. if (!zlc)
  1049. return NULL;
  1050. if (jiffies - zlc->last_full_zap > 1 * HZ) {
  1051. bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
  1052. zlc->last_full_zap = jiffies;
  1053. }
  1054. allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
  1055. &cpuset_current_mems_allowed :
  1056. &node_states[N_HIGH_MEMORY];
  1057. return allowednodes;
  1058. }
  1059. /*
  1060. * Given 'z' scanning a zonelist, run a couple of quick checks to see
  1061. * if it is worth looking at further for free memory:
  1062. * 1) Check that the zone isn't thought to be full (doesn't have its
  1063. * bit set in the zonelist_cache fullzones BITMAP).
  1064. * 2) Check that the zones node (obtained from the zonelist_cache
  1065. * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
  1066. * Return true (non-zero) if zone is worth looking at further, or
  1067. * else return false (zero) if it is not.
  1068. *
  1069. * This check -ignores- the distinction between various watermarks,
  1070. * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
  1071. * found to be full for any variation of these watermarks, it will
  1072. * be considered full for up to one second by all requests, unless
  1073. * we are so low on memory on all allowed nodes that we are forced
  1074. * into the second scan of the zonelist.
  1075. *
  1076. * In the second scan we ignore this zonelist cache and exactly
  1077. * apply the watermarks to all zones, even it is slower to do so.
  1078. * We are low on memory in the second scan, and should leave no stone
  1079. * unturned looking for a free page.
  1080. */
  1081. static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
  1082. nodemask_t *allowednodes)
  1083. {
  1084. struct zonelist_cache *zlc; /* cached zonelist speedup info */
  1085. int i; /* index of *z in zonelist zones */
  1086. int n; /* node that zone *z is on */
  1087. zlc = zonelist->zlcache_ptr;
  1088. if (!zlc)
  1089. return 1;
  1090. i = z - zonelist->zones;
  1091. n = zlc->z_to_n[i];
  1092. /* This zone is worth trying if it is allowed but not full */
  1093. return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
  1094. }
  1095. /*
  1096. * Given 'z' scanning a zonelist, set the corresponding bit in
  1097. * zlc->fullzones, so that subsequent attempts to allocate a page
  1098. * from that zone don't waste time re-examining it.
  1099. */
  1100. static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
  1101. {
  1102. struct zonelist_cache *zlc; /* cached zonelist speedup info */
  1103. int i; /* index of *z in zonelist zones */
  1104. zlc = zonelist->zlcache_ptr;
  1105. if (!zlc)
  1106. return;
  1107. i = z - zonelist->zones;
  1108. set_bit(i, zlc->fullzones);
  1109. }
  1110. #else /* CONFIG_NUMA */
  1111. static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
  1112. {
  1113. return NULL;
  1114. }
  1115. static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
  1116. nodemask_t *allowednodes)
  1117. {
  1118. return 1;
  1119. }
  1120. static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
  1121. {
  1122. }
  1123. #endif /* CONFIG_NUMA */
  1124. /*
  1125. * get_page_from_freelist goes through the zonelist trying to allocate
  1126. * a page.
  1127. */
  1128. static struct page *
  1129. get_page_from_freelist(gfp_t gfp_mask, unsigned int order,
  1130. struct zonelist *zonelist, int alloc_flags)
  1131. {
  1132. struct zone **z;
  1133. struct page *page = NULL;
  1134. int classzone_idx = zone_idx(zonelist->zones[0]);
  1135. struct zone *zone;
  1136. nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
  1137. int zlc_active = 0; /* set if using zonelist_cache */
  1138. int did_zlc_setup = 0; /* just call zlc_setup() one time */
  1139. enum zone_type highest_zoneidx = -1; /* Gets set for policy zonelists */
  1140. zonelist_scan:
  1141. /*
  1142. * Scan zonelist, looking for a zone with enough free.
  1143. * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
  1144. */
  1145. z = zonelist->zones;
  1146. do {
  1147. /*
  1148. * In NUMA, this could be a policy zonelist which contains
  1149. * zones that may not be allowed by the current gfp_mask.
  1150. * Check the zone is allowed by the current flags
  1151. */
  1152. if (unlikely(alloc_should_filter_zonelist(zonelist))) {
  1153. if (highest_zoneidx == -1)
  1154. highest_zoneidx = gfp_zone(gfp_mask);
  1155. if (zone_idx(*z) > highest_zoneidx)
  1156. continue;
  1157. }
  1158. if (NUMA_BUILD && zlc_active &&
  1159. !zlc_zone_worth_trying(zonelist, z, allowednodes))
  1160. continue;
  1161. zone = *z;
  1162. if ((alloc_flags & ALLOC_CPUSET) &&
  1163. !cpuset_zone_allowed_softwall(zone, gfp_mask))
  1164. goto try_next_zone;
  1165. if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
  1166. unsigned long mark;
  1167. if (alloc_flags & ALLOC_WMARK_MIN)
  1168. mark = zone->pages_min;
  1169. else if (alloc_flags & ALLOC_WMARK_LOW)
  1170. mark = zone->pages_low;
  1171. else
  1172. mark = zone->pages_high;
  1173. if (!zone_watermark_ok(zone, order, mark,
  1174. classzone_idx, alloc_flags)) {
  1175. if (!zone_reclaim_mode ||
  1176. !zone_reclaim(zone, gfp_mask, order))
  1177. goto this_zone_full;
  1178. }
  1179. }
  1180. page = buffered_rmqueue(zonelist, zone, order, gfp_mask);
  1181. if (page)
  1182. break;
  1183. this_zone_full:
  1184. if (NUMA_BUILD)
  1185. zlc_mark_zone_full(zonelist, z);
  1186. try_next_zone:
  1187. if (NUMA_BUILD && !did_zlc_setup) {
  1188. /* we do zlc_setup after the first zone is tried */
  1189. allowednodes = zlc_setup(zonelist, alloc_flags);
  1190. zlc_active = 1;
  1191. did_zlc_setup = 1;
  1192. }
  1193. } while (*(++z) != NULL);
  1194. if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
  1195. /* Disable zlc cache for second zonelist scan */
  1196. zlc_active = 0;
  1197. goto zonelist_scan;
  1198. }
  1199. return page;
  1200. }
  1201. /*
  1202. * This is the 'heart' of the zoned buddy allocator.
  1203. */
  1204. struct page * fastcall
  1205. __alloc_pages(gfp_t gfp_mask, unsigned int order,
  1206. struct zonelist *zonelist)
  1207. {
  1208. const gfp_t wait = gfp_mask & __GFP_WAIT;
  1209. struct zone **z;
  1210. struct page *page;
  1211. struct reclaim_state reclaim_state;
  1212. struct task_struct *p = current;
  1213. int do_retry;
  1214. int alloc_flags;
  1215. int did_some_progress;
  1216. might_sleep_if(wait);
  1217. if (should_fail_alloc_page(gfp_mask, order))
  1218. return NULL;
  1219. restart:
  1220. z = zonelist->zones; /* the list of zones suitable for gfp_mask */
  1221. if (unlikely(*z == NULL)) {
  1222. /*
  1223. * Happens if we have an empty zonelist as a result of
  1224. * GFP_THISNODE being used on a memoryless node
  1225. */
  1226. return NULL;
  1227. }
  1228. page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
  1229. zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET);
  1230. if (page)
  1231. goto got_pg;
  1232. /*
  1233. * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
  1234. * __GFP_NOWARN set) should not cause reclaim since the subsystem
  1235. * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
  1236. * using a larger set of nodes after it has established that the
  1237. * allowed per node queues are empty and that nodes are
  1238. * over allocated.
  1239. */
  1240. if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
  1241. goto nopage;
  1242. for (z = zonelist->zones; *z; z++)
  1243. wakeup_kswapd(*z, order);
  1244. /*
  1245. * OK, we're below the kswapd watermark and have kicked background
  1246. * reclaim. Now things get more complex, so set up alloc_flags according
  1247. * to how we want to proceed.
  1248. *
  1249. * The caller may dip into page reserves a bit more if the caller
  1250. * cannot run direct reclaim, or if the caller has realtime scheduling
  1251. * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
  1252. * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
  1253. */
  1254. alloc_flags = ALLOC_WMARK_MIN;
  1255. if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
  1256. alloc_flags |= ALLOC_HARDER;
  1257. if (gfp_mask & __GFP_HIGH)
  1258. alloc_flags |= ALLOC_HIGH;
  1259. if (wait)
  1260. alloc_flags |= ALLOC_CPUSET;
  1261. /*
  1262. * Go through the zonelist again. Let __GFP_HIGH and allocations
  1263. * coming from realtime tasks go deeper into reserves.
  1264. *
  1265. * This is the last chance, in general, before the goto nopage.
  1266. * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
  1267. * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
  1268. */
  1269. page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
  1270. if (page)
  1271. goto got_pg;
  1272. /* This allocation should allow future memory freeing. */
  1273. rebalance:
  1274. if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
  1275. && !in_interrupt()) {
  1276. if (!(gfp_mask & __GFP_NOMEMALLOC)) {
  1277. nofail_alloc:
  1278. /* go through the zonelist yet again, ignoring mins */
  1279. page = get_page_from_freelist(gfp_mask, order,
  1280. zonelist, ALLOC_NO_WATERMARKS);
  1281. if (page)
  1282. goto got_pg;
  1283. if (gfp_mask & __GFP_NOFAIL) {
  1284. congestion_wait(WRITE, HZ/50);
  1285. goto nofail_alloc;
  1286. }
  1287. }
  1288. goto nopage;
  1289. }
  1290. /* Atomic allocations - we can't balance anything */
  1291. if (!wait)
  1292. goto nopage;
  1293. cond_resched();
  1294. /* We now go into synchronous reclaim */
  1295. cpuset_memory_pressure_bump();
  1296. p->flags |= PF_MEMALLOC;
  1297. reclaim_state.reclaimed_slab = 0;
  1298. p->reclaim_state = &reclaim_state;
  1299. did_some_progress = try_to_free_pages(zonelist->zones, order, gfp_mask);
  1300. p->reclaim_state = NULL;
  1301. p->flags &= ~PF_MEMALLOC;
  1302. cond_resched();
  1303. if (order != 0)
  1304. drain_all_local_pages();
  1305. if (likely(did_some_progress)) {
  1306. page = get_page_from_freelist(gfp_mask, order,
  1307. zonelist, alloc_flags);
  1308. if (page)
  1309. goto got_pg;
  1310. } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
  1311. /*
  1312. * Go through the zonelist yet one more time, keep
  1313. * very high watermark here, this is only to catch
  1314. * a parallel oom killing, we must fail if we're still
  1315. * under heavy pressure.
  1316. */
  1317. page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
  1318. zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET);
  1319. if (page)
  1320. goto got_pg;
  1321. /* The OOM killer will not help higher order allocs so fail */
  1322. if (order > PAGE_ALLOC_COSTLY_ORDER)
  1323. goto nopage;
  1324. out_of_memory(zonelist, gfp_mask, order);
  1325. goto restart;
  1326. }
  1327. /*
  1328. * Don't let big-order allocations loop unless the caller explicitly
  1329. * requests that. Wait for some write requests to complete then retry.
  1330. *
  1331. * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
  1332. * <= 3, but that may not be true in other implementations.
  1333. */
  1334. do_retry = 0;
  1335. if (!(gfp_mask & __GFP_NORETRY)) {
  1336. if ((order <= PAGE_ALLOC_COSTLY_ORDER) ||
  1337. (gfp_mask & __GFP_REPEAT))
  1338. do_retry = 1;
  1339. if (gfp_mask & __GFP_NOFAIL)
  1340. do_retry = 1;
  1341. }
  1342. if (do_retry) {
  1343. congestion_wait(WRITE, HZ/50);
  1344. goto rebalance;
  1345. }
  1346. nopage:
  1347. if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
  1348. printk(KERN_WARNING "%s: page allocation failure."
  1349. " order:%d, mode:0x%x\n",
  1350. p->comm, order, gfp_mask);
  1351. dump_stack();
  1352. show_mem();
  1353. }
  1354. got_pg:
  1355. return page;
  1356. }
  1357. EXPORT_SYMBOL(__alloc_pages);
  1358. /*
  1359. * Common helper functions.
  1360. */
  1361. fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
  1362. {
  1363. struct page * page;
  1364. page = alloc_pages(gfp_mask, order);
  1365. if (!page)
  1366. return 0;
  1367. return (unsigned long) page_address(page);
  1368. }
  1369. EXPORT_SYMBOL(__get_free_pages);
  1370. fastcall unsigned long get_zeroed_page(gfp_t gfp_mask)
  1371. {
  1372. struct page * page;
  1373. /*
  1374. * get_zeroed_page() returns a 32-bit address, which cannot represent
  1375. * a highmem page
  1376. */
  1377. VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
  1378. page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
  1379. if (page)
  1380. return (unsigned long) page_address(page);
  1381. return 0;
  1382. }
  1383. EXPORT_SYMBOL(get_zeroed_page);
  1384. void __pagevec_free(struct pagevec *pvec)
  1385. {
  1386. int i = pagevec_count(pvec);
  1387. while (--i >= 0)
  1388. free_hot_cold_page(pvec->pages[i], pvec->cold);
  1389. }
  1390. fastcall void __free_pages(struct page *page, unsigned int order)
  1391. {
  1392. if (put_page_testzero(page)) {
  1393. if (order == 0)
  1394. free_hot_page(page);
  1395. else
  1396. __free_pages_ok(page, order);
  1397. }
  1398. }
  1399. EXPORT_SYMBOL(__free_pages);
  1400. fastcall void free_pages(unsigned long addr, unsigned int order)
  1401. {
  1402. if (addr != 0) {
  1403. VM_BUG_ON(!virt_addr_valid((void *)addr));
  1404. __free_pages(virt_to_page((void *)addr), order);
  1405. }
  1406. }
  1407. EXPORT_SYMBOL(free_pages);
  1408. static unsigned int nr_free_zone_pages(int offset)
  1409. {
  1410. /* Just pick one node, since fallback list is circular */
  1411. pg_data_t *pgdat = NODE_DATA(numa_node_id());
  1412. unsigned int sum = 0;
  1413. struct zonelist *zonelist = pgdat->node_zonelists + offset;
  1414. struct zone **zonep = zonelist->zones;
  1415. struct zone *zone;
  1416. for (zone = *zonep++; zone; zone = *zonep++) {
  1417. unsigned long size = zone->present_pages;
  1418. unsigned long high = zone->pages_high;
  1419. if (size > high)
  1420. sum += size - high;
  1421. }
  1422. return sum;
  1423. }
  1424. /*
  1425. * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
  1426. */
  1427. unsigned int nr_free_buffer_pages(void)
  1428. {
  1429. return nr_free_zone_pages(gfp_zone(GFP_USER));
  1430. }
  1431. EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
  1432. /*
  1433. * Amount of free RAM allocatable within all zones
  1434. */
  1435. unsigned int nr_free_pagecache_pages(void)
  1436. {
  1437. return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
  1438. }
  1439. static inline void show_node(struct zone *zone)
  1440. {
  1441. if (NUMA_BUILD)
  1442. printk("Node %d ", zone_to_nid(zone));
  1443. }
  1444. void si_meminfo(struct sysinfo *val)
  1445. {
  1446. val->totalram = totalram_pages;
  1447. val->sharedram = 0;
  1448. val->freeram = global_page_state(NR_FREE_PAGES);
  1449. val->bufferram = nr_blockdev_pages();
  1450. val->totalhigh = totalhigh_pages;
  1451. val->freehigh = nr_free_highpages();
  1452. val->mem_unit = PAGE_SIZE;
  1453. }
  1454. EXPORT_SYMBOL(si_meminfo);
  1455. #ifdef CONFIG_NUMA
  1456. void si_meminfo_node(struct sysinfo *val, int nid)
  1457. {
  1458. pg_data_t *pgdat = NODE_DATA(nid);
  1459. val->totalram = pgdat->node_present_pages;
  1460. val->freeram = node_page_state(nid, NR_FREE_PAGES);
  1461. #ifdef CONFIG_HIGHMEM
  1462. val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
  1463. val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
  1464. NR_FREE_PAGES);
  1465. #else
  1466. val->totalhigh = 0;
  1467. val->freehigh = 0;
  1468. #endif
  1469. val->mem_unit = PAGE_SIZE;
  1470. }
  1471. #endif
  1472. #define K(x) ((x) << (PAGE_SHIFT-10))
  1473. /*
  1474. * Show free area list (used inside shift_scroll-lock stuff)
  1475. * We also calculate the percentage fragmentation. We do this by counting the
  1476. * memory on each free list with the exception of the first item on the list.
  1477. */
  1478. void show_free_areas(void)
  1479. {
  1480. int cpu;
  1481. struct zone *zone;
  1482. for_each_zone(zone) {
  1483. if (!populated_zone(zone))
  1484. continue;
  1485. show_node(zone);
  1486. printk("%s per-cpu:\n", zone->name);
  1487. for_each_online_cpu(cpu) {
  1488. struct per_cpu_pageset *pageset;
  1489. pageset = zone_pcp(zone, cpu);
  1490. printk("CPU %4d: Hot: hi:%5d, btch:%4d usd:%4d "
  1491. "Cold: hi:%5d, btch:%4d usd:%4d\n",
  1492. cpu, pageset->pcp[0].high,
  1493. pageset->pcp[0].batch, pageset->pcp[0].count,
  1494. pageset->pcp[1].high, pageset->pcp[1].batch,
  1495. pageset->pcp[1].count);
  1496. }
  1497. }
  1498. printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n"
  1499. " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
  1500. global_page_state(NR_ACTIVE),
  1501. global_page_state(NR_INACTIVE),
  1502. global_page_state(NR_FILE_DIRTY),
  1503. global_page_state(NR_WRITEBACK),
  1504. global_page_state(NR_UNSTABLE_NFS),
  1505. global_page_state(NR_FREE_PAGES),
  1506. global_page_state(NR_SLAB_RECLAIMABLE) +
  1507. global_page_state(NR_SLAB_UNRECLAIMABLE),
  1508. global_page_state(NR_FILE_MAPPED),
  1509. global_page_state(NR_PAGETABLE),
  1510. global_page_state(NR_BOUNCE));
  1511. for_each_zone(zone) {
  1512. int i;
  1513. if (!populated_zone(zone))
  1514. continue;
  1515. show_node(zone);
  1516. printk("%s"
  1517. " free:%lukB"
  1518. " min:%lukB"
  1519. " low:%lukB"
  1520. " high:%lukB"
  1521. " active:%lukB"
  1522. " inactive:%lukB"
  1523. " present:%lukB"
  1524. " pages_scanned:%lu"
  1525. " all_unreclaimable? %s"
  1526. "\n",
  1527. zone->name,
  1528. K(zone_page_state(zone, NR_FREE_PAGES)),
  1529. K(zone->pages_min),
  1530. K(zone->pages_low),
  1531. K(zone->pages_high),
  1532. K(zone_page_state(zone, NR_ACTIVE)),
  1533. K(zone_page_state(zone, NR_INACTIVE)),
  1534. K(zone->present_pages),
  1535. zone->pages_scanned,
  1536. (zone->all_unreclaimable ? "yes" : "no")
  1537. );
  1538. printk("lowmem_reserve[]:");
  1539. for (i = 0; i < MAX_NR_ZONES; i++)
  1540. printk(" %lu", zone->lowmem_reserve[i]);
  1541. printk("\n");
  1542. }
  1543. for_each_zone(zone) {
  1544. unsigned long nr[MAX_ORDER], flags, order, total = 0;
  1545. if (!populated_zone(zone))
  1546. continue;
  1547. show_node(zone);
  1548. printk("%s: ", zone->name);
  1549. spin_lock_irqsave(&zone->lock, flags);
  1550. for (order = 0; order < MAX_ORDER; order++) {
  1551. nr[order] = zone->free_area[order].nr_free;
  1552. total += nr[order] << order;
  1553. }
  1554. spin_unlock_irqrestore(&zone->lock, flags);
  1555. for (order = 0; order < MAX_ORDER; order++)
  1556. printk("%lu*%lukB ", nr[order], K(1UL) << order);
  1557. printk("= %lukB\n", K(total));
  1558. }
  1559. show_swap_cache_info();
  1560. }
  1561. /*
  1562. * Builds allocation fallback zone lists.
  1563. *
  1564. * Add all populated zones of a node to the zonelist.
  1565. */
  1566. static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
  1567. int nr_zones, enum zone_type zone_type)
  1568. {
  1569. struct zone *zone;
  1570. BUG_ON(zone_type >= MAX_NR_ZONES);
  1571. zone_type++;
  1572. do {
  1573. zone_type--;
  1574. zone = pgdat->node_zones + zone_type;
  1575. if (populated_zone(zone)) {
  1576. zonelist->zones[nr_zones++] = zone;
  1577. check_highest_zone(zone_type);
  1578. }
  1579. } while (zone_type);
  1580. return nr_zones;
  1581. }
  1582. /*
  1583. * zonelist_order:
  1584. * 0 = automatic detection of better ordering.
  1585. * 1 = order by ([node] distance, -zonetype)
  1586. * 2 = order by (-zonetype, [node] distance)
  1587. *
  1588. * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
  1589. * the same zonelist. So only NUMA can configure this param.
  1590. */
  1591. #define ZONELIST_ORDER_DEFAULT 0
  1592. #define ZONELIST_ORDER_NODE 1
  1593. #define ZONELIST_ORDER_ZONE 2
  1594. /* zonelist order in the kernel.
  1595. * set_zonelist_order() will set this to NODE or ZONE.
  1596. */
  1597. static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
  1598. static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
  1599. #ifdef CONFIG_NUMA
  1600. /* The value user specified ....changed by config */
  1601. static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
  1602. /* string for sysctl */
  1603. #define NUMA_ZONELIST_ORDER_LEN 16
  1604. char numa_zonelist_order[16] = "default";
  1605. /*
  1606. * interface for configure zonelist ordering.
  1607. * command line option "numa_zonelist_order"
  1608. * = "[dD]efault - default, automatic configuration.
  1609. * = "[nN]ode - order by node locality, then by zone within node
  1610. * = "[zZ]one - order by zone, then by locality within zone
  1611. */
  1612. static int __parse_numa_zonelist_order(char *s)
  1613. {
  1614. if (*s == 'd' || *s == 'D') {
  1615. user_zonelist_order = ZONELIST_ORDER_DEFAULT;
  1616. } else if (*s == 'n' || *s == 'N') {
  1617. user_zonelist_order = ZONELIST_ORDER_NODE;
  1618. } else if (*s == 'z' || *s == 'Z') {
  1619. user_zonelist_order = ZONELIST_ORDER_ZONE;
  1620. } else {
  1621. printk(KERN_WARNING
  1622. "Ignoring invalid numa_zonelist_order value: "
  1623. "%s\n", s);
  1624. return -EINVAL;
  1625. }
  1626. return 0;
  1627. }
  1628. static __init int setup_numa_zonelist_order(char *s)
  1629. {
  1630. if (s)
  1631. return __parse_numa_zonelist_order(s);
  1632. return 0;
  1633. }
  1634. early_param("numa_zonelist_order", setup_numa_zonelist_order);
  1635. /*
  1636. * sysctl handler for numa_zonelist_order
  1637. */
  1638. int numa_zonelist_order_handler(ctl_table *table, int write,
  1639. struct file *file, void __user *buffer, size_t *length,
  1640. loff_t *ppos)
  1641. {
  1642. char saved_string[NUMA_ZONELIST_ORDER_LEN];
  1643. int ret;
  1644. if (write)
  1645. strncpy(saved_string, (char*)table->data,
  1646. NUMA_ZONELIST_ORDER_LEN);
  1647. ret = proc_dostring(table, write, file, buffer, length, ppos);
  1648. if (ret)
  1649. return ret;
  1650. if (write) {
  1651. int oldval = user_zonelist_order;
  1652. if (__parse_numa_zonelist_order((char*)table->data)) {
  1653. /*
  1654. * bogus value. restore saved string
  1655. */
  1656. strncpy((char*)table->data, saved_string,
  1657. NUMA_ZONELIST_ORDER_LEN);
  1658. user_zonelist_order = oldval;
  1659. } else if (oldval != user_zonelist_order)
  1660. build_all_zonelists();
  1661. }
  1662. return 0;
  1663. }
  1664. #define MAX_NODE_LOAD (num_online_nodes())
  1665. static int node_load[MAX_NUMNODES];
  1666. /**
  1667. * find_next_best_node - find the next node that should appear in a given node's fallback list
  1668. * @node: node whose fallback list we're appending
  1669. * @used_node_mask: nodemask_t of already used nodes
  1670. *
  1671. * We use a number of factors to determine which is the next node that should
  1672. * appear on a given node's fallback list. The node should not have appeared
  1673. * already in @node's fallback list, and it should be the next closest node
  1674. * according to the distance array (which contains arbitrary distance values
  1675. * from each node to each node in the system), and should also prefer nodes
  1676. * with no CPUs, since presumably they'll have very little allocation pressure
  1677. * on them otherwise.
  1678. * It returns -1 if no node is found.
  1679. */
  1680. static int find_next_best_node(int node, nodemask_t *used_node_mask)
  1681. {
  1682. int n, val;
  1683. int min_val = INT_MAX;
  1684. int best_node = -1;
  1685. /* Use the local node if we haven't already */
  1686. if (!node_isset(node, *used_node_mask)) {
  1687. node_set(node, *used_node_mask);
  1688. return node;
  1689. }
  1690. for_each_node_state(n, N_HIGH_MEMORY) {
  1691. cpumask_t tmp;
  1692. /* Don't want a node to appear more than once */
  1693. if (node_isset(n, *used_node_mask))
  1694. continue;
  1695. /* Use the distance array to find the distance */
  1696. val = node_distance(node, n);
  1697. /* Penalize nodes under us ("prefer the next node") */
  1698. val += (n < node);
  1699. /* Give preference to headless and unused nodes */
  1700. tmp = node_to_cpumask(n);
  1701. if (!cpus_empty(tmp))
  1702. val += PENALTY_FOR_NODE_WITH_CPUS;
  1703. /* Slight preference for less loaded node */
  1704. val *= (MAX_NODE_LOAD*MAX_NUMNODES);
  1705. val += node_load[n];
  1706. if (val < min_val) {
  1707. min_val = val;
  1708. best_node = n;
  1709. }
  1710. }
  1711. if (best_node >= 0)
  1712. node_set(best_node, *used_node_mask);
  1713. return best_node;
  1714. }
  1715. /*
  1716. * Build zonelists ordered by node and zones within node.
  1717. * This results in maximum locality--normal zone overflows into local
  1718. * DMA zone, if any--but risks exhausting DMA zone.
  1719. */
  1720. static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
  1721. {
  1722. enum zone_type i;
  1723. int j;
  1724. struct zonelist *zonelist;
  1725. for (i = 0; i < MAX_NR_ZONES; i++) {
  1726. zonelist = pgdat->node_zonelists + i;
  1727. for (j = 0; zonelist->zones[j] != NULL; j++)
  1728. ;
  1729. j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
  1730. zonelist->zones[j] = NULL;
  1731. }
  1732. }
  1733. /*
  1734. * Build gfp_thisnode zonelists
  1735. */
  1736. static void build_thisnode_zonelists(pg_data_t *pgdat)
  1737. {
  1738. enum zone_type i;
  1739. int j;
  1740. struct zonelist *zonelist;
  1741. for (i = 0; i < MAX_NR_ZONES; i++) {
  1742. zonelist = pgdat->node_zonelists + MAX_NR_ZONES + i;
  1743. j = build_zonelists_node(pgdat, zonelist, 0, i);
  1744. zonelist->zones[j] = NULL;
  1745. }
  1746. }
  1747. /*
  1748. * Build zonelists ordered by zone and nodes within zones.
  1749. * This results in conserving DMA zone[s] until all Normal memory is
  1750. * exhausted, but results in overflowing to remote node while memory
  1751. * may still exist in local DMA zone.
  1752. */
  1753. static int node_order[MAX_NUMNODES];
  1754. static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
  1755. {
  1756. enum zone_type i;
  1757. int pos, j, node;
  1758. int zone_type; /* needs to be signed */
  1759. struct zone *z;
  1760. struct zonelist *zonelist;
  1761. for (i = 0; i < MAX_NR_ZONES; i++) {
  1762. zonelist = pgdat->node_zonelists + i;
  1763. pos = 0;
  1764. for (zone_type = i; zone_type >= 0; zone_type--) {
  1765. for (j = 0; j < nr_nodes; j++) {
  1766. node = node_order[j];
  1767. z = &NODE_DATA(node)->node_zones[zone_type];
  1768. if (populated_zone(z)) {
  1769. zonelist->zones[pos++] = z;
  1770. check_highest_zone(zone_type);
  1771. }
  1772. }
  1773. }
  1774. zonelist->zones[pos] = NULL;
  1775. }
  1776. }
  1777. static int default_zonelist_order(void)
  1778. {
  1779. int nid, zone_type;
  1780. unsigned long low_kmem_size,total_size;
  1781. struct zone *z;
  1782. int average_size;
  1783. /*
  1784. * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
  1785. * If they are really small and used heavily, the system can fall
  1786. * into OOM very easily.
  1787. * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
  1788. */
  1789. /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
  1790. low_kmem_size = 0;
  1791. total_size = 0;
  1792. for_each_online_node(nid) {
  1793. for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
  1794. z = &NODE_DATA(nid)->node_zones[zone_type];
  1795. if (populated_zone(z)) {
  1796. if (zone_type < ZONE_NORMAL)
  1797. low_kmem_size += z->present_pages;
  1798. total_size += z->present_pages;
  1799. }
  1800. }
  1801. }
  1802. if (!low_kmem_size || /* there are no DMA area. */
  1803. low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
  1804. return ZONELIST_ORDER_NODE;
  1805. /*
  1806. * look into each node's config.
  1807. * If there is a node whose DMA/DMA32 memory is very big area on
  1808. * local memory, NODE_ORDER may be suitable.
  1809. */
  1810. average_size = total_size /
  1811. (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
  1812. for_each_online_node(nid) {
  1813. low_kmem_size = 0;
  1814. total_size = 0;
  1815. for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
  1816. z = &NODE_DATA(nid)->node_zones[zone_type];
  1817. if (populated_zone(z)) {
  1818. if (zone_type < ZONE_NORMAL)
  1819. low_kmem_size += z->present_pages;
  1820. total_size += z->present_pages;
  1821. }
  1822. }
  1823. if (low_kmem_size &&
  1824. total_size > average_size && /* ignore small node */
  1825. low_kmem_size > total_size * 70/100)
  1826. return ZONELIST_ORDER_NODE;
  1827. }
  1828. return ZONELIST_ORDER_ZONE;
  1829. }
  1830. static void set_zonelist_order(void)
  1831. {
  1832. if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
  1833. current_zonelist_order = default_zonelist_order();
  1834. else
  1835. current_zonelist_order = user_zonelist_order;
  1836. }
  1837. static void build_zonelists(pg_data_t *pgdat)
  1838. {
  1839. int j, node, load;
  1840. enum zone_type i;
  1841. nodemask_t used_mask;
  1842. int local_node, prev_node;
  1843. struct zonelist *zonelist;
  1844. int order = current_zonelist_order;
  1845. /* initialize zonelists */
  1846. for (i = 0; i < MAX_ZONELISTS; i++) {
  1847. zonelist = pgdat->node_zonelists + i;
  1848. zonelist->zones[0] = NULL;
  1849. }
  1850. /* NUMA-aware ordering of nodes */
  1851. local_node = pgdat->node_id;
  1852. load = num_online_nodes();
  1853. prev_node = local_node;
  1854. nodes_clear(used_mask);
  1855. memset(node_load, 0, sizeof(node_load));
  1856. memset(node_order, 0, sizeof(node_order));
  1857. j = 0;
  1858. while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
  1859. int distance = node_distance(local_node, node);
  1860. /*
  1861. * If another node is sufficiently far away then it is better
  1862. * to reclaim pages in a zone before going off node.
  1863. */
  1864. if (distance > RECLAIM_DISTANCE)
  1865. zone_reclaim_mode = 1;
  1866. /*
  1867. * We don't want to pressure a particular node.
  1868. * So adding penalty to the first node in same
  1869. * distance group to make it round-robin.
  1870. */
  1871. if (distance != node_distance(local_node, prev_node))
  1872. node_load[node] = load;
  1873. prev_node = node;
  1874. load--;
  1875. if (order == ZONELIST_ORDER_NODE)
  1876. build_zonelists_in_node_order(pgdat, node);
  1877. else
  1878. node_order[j++] = node; /* remember order */
  1879. }
  1880. if (order == ZONELIST_ORDER_ZONE) {
  1881. /* calculate node order -- i.e., DMA last! */
  1882. build_zonelists_in_zone_order(pgdat, j);
  1883. }
  1884. build_thisnode_zonelists(pgdat);
  1885. }
  1886. /* Construct the zonelist performance cache - see further mmzone.h */
  1887. static void build_zonelist_cache(pg_data_t *pgdat)
  1888. {
  1889. int i;
  1890. for (i = 0; i < MAX_NR_ZONES; i++) {
  1891. struct zonelist *zonelist;
  1892. struct zonelist_cache *zlc;
  1893. struct zone **z;
  1894. zonelist = pgdat->node_zonelists + i;
  1895. zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
  1896. bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
  1897. for (z = zonelist->zones; *z; z++)
  1898. zlc->z_to_n[z - zonelist->zones] = zone_to_nid(*z);
  1899. }
  1900. }
  1901. #else /* CONFIG_NUMA */
  1902. static void set_zonelist_order(void)
  1903. {
  1904. current_zonelist_order = ZONELIST_ORDER_ZONE;
  1905. }
  1906. static void build_zonelists(pg_data_t *pgdat)
  1907. {
  1908. int node, local_node;
  1909. enum zone_type i,j;
  1910. local_node = pgdat->node_id;
  1911. for (i = 0; i < MAX_NR_ZONES; i++) {
  1912. struct zonelist *zonelist;
  1913. zonelist = pgdat->node_zonelists + i;
  1914. j = build_zonelists_node(pgdat, zonelist, 0, i);
  1915. /*
  1916. * Now we build the zonelist so that it contains the zones
  1917. * of all the other nodes.
  1918. * We don't want to pressure a particular node, so when
  1919. * building the zones for node N, we make sure that the
  1920. * zones coming right after the local ones are those from
  1921. * node N+1 (modulo N)
  1922. */
  1923. for (node = local_node + 1; node < MAX_NUMNODES; node++) {
  1924. if (!node_online(node))
  1925. continue;
  1926. j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
  1927. }
  1928. for (node = 0; node < local_node; node++) {
  1929. if (!node_online(node))
  1930. continue;
  1931. j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
  1932. }
  1933. zonelist->zones[j] = NULL;
  1934. }
  1935. }
  1936. /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
  1937. static void build_zonelist_cache(pg_data_t *pgdat)
  1938. {
  1939. int i;
  1940. for (i = 0; i < MAX_NR_ZONES; i++)
  1941. pgdat->node_zonelists[i].zlcache_ptr = NULL;
  1942. }
  1943. #endif /* CONFIG_NUMA */
  1944. /* return values int ....just for stop_machine_run() */
  1945. static int __build_all_zonelists(void *dummy)
  1946. {
  1947. int nid;
  1948. for_each_online_node(nid) {
  1949. pg_data_t *pgdat = NODE_DATA(nid);
  1950. build_zonelists(pgdat);
  1951. build_zonelist_cache(pgdat);
  1952. }
  1953. return 0;
  1954. }
  1955. void build_all_zonelists(void)
  1956. {
  1957. set_zonelist_order();
  1958. if (system_state == SYSTEM_BOOTING) {
  1959. __build_all_zonelists(NULL);
  1960. cpuset_init_current_mems_allowed();
  1961. } else {
  1962. /* we have to stop all cpus to guaranntee there is no user
  1963. of zonelist */
  1964. stop_machine_run(__build_all_zonelists, NULL, NR_CPUS);
  1965. /* cpuset refresh routine should be here */
  1966. }
  1967. vm_total_pages = nr_free_pagecache_pages();
  1968. printk("Built %i zonelists in %s order. Total pages: %ld\n",
  1969. num_online_nodes(),
  1970. zonelist_order_name[current_zonelist_order],
  1971. vm_total_pages);
  1972. #ifdef CONFIG_NUMA
  1973. printk("Policy zone: %s\n", zone_names[policy_zone]);
  1974. #endif
  1975. }
  1976. /*
  1977. * Helper functions to size the waitqueue hash table.
  1978. * Essentially these want to choose hash table sizes sufficiently
  1979. * large so that collisions trying to wait on pages are rare.
  1980. * But in fact, the number of active page waitqueues on typical
  1981. * systems is ridiculously low, less than 200. So this is even
  1982. * conservative, even though it seems large.
  1983. *
  1984. * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
  1985. * waitqueues, i.e. the size of the waitq table given the number of pages.
  1986. */
  1987. #define PAGES_PER_WAITQUEUE 256
  1988. #ifndef CONFIG_MEMORY_HOTPLUG
  1989. static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
  1990. {
  1991. unsigned long size = 1;
  1992. pages /= PAGES_PER_WAITQUEUE;
  1993. while (size < pages)
  1994. size <<= 1;
  1995. /*
  1996. * Once we have dozens or even hundreds of threads sleeping
  1997. * on IO we've got bigger problems than wait queue collision.
  1998. * Limit the size of the wait table to a reasonable size.
  1999. */
  2000. size = min(size, 4096UL);
  2001. return max(size, 4UL);
  2002. }
  2003. #else
  2004. /*
  2005. * A zone's size might be changed by hot-add, so it is not possible to determine
  2006. * a suitable size for its wait_table. So we use the maximum size now.
  2007. *
  2008. * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
  2009. *
  2010. * i386 (preemption config) : 4096 x 16 = 64Kbyte.
  2011. * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
  2012. * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
  2013. *
  2014. * The maximum entries are prepared when a zone's memory is (512K + 256) pages
  2015. * or more by the traditional way. (See above). It equals:
  2016. *
  2017. * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
  2018. * ia64(16K page size) : = ( 8G + 4M)byte.
  2019. * powerpc (64K page size) : = (32G +16M)byte.
  2020. */
  2021. static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
  2022. {
  2023. return 4096UL;
  2024. }
  2025. #endif
  2026. /*
  2027. * This is an integer logarithm so that shifts can be used later
  2028. * to extract the more random high bits from the multiplicative
  2029. * hash function before the remainder is taken.
  2030. */
  2031. static inline unsigned long wait_table_bits(unsigned long size)
  2032. {
  2033. return ffz(~size);
  2034. }
  2035. #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
  2036. /*
  2037. * Initially all pages are reserved - free ones are freed
  2038. * up by free_all_bootmem() once the early boot process is
  2039. * done. Non-atomic initialization, single-pass.
  2040. */
  2041. void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
  2042. unsigned long start_pfn, enum memmap_context context)
  2043. {
  2044. struct page *page;
  2045. unsigned long end_pfn = start_pfn + size;
  2046. unsigned long pfn;
  2047. for (pfn = start_pfn; pfn < end_pfn; pfn++) {
  2048. /*
  2049. * There can be holes in boot-time mem_map[]s
  2050. * handed to this function. They do not
  2051. * exist on hotplugged memory.
  2052. */
  2053. if (context == MEMMAP_EARLY) {
  2054. if (!early_pfn_valid(pfn))
  2055. continue;
  2056. if (!early_pfn_in_nid(pfn, nid))
  2057. continue;
  2058. }
  2059. page = pfn_to_page(pfn);
  2060. set_page_links(page, zone, nid, pfn);
  2061. init_page_count(page);
  2062. reset_page_mapcount(page);
  2063. SetPageReserved(page);
  2064. /*
  2065. * Mark the block movable so that blocks are reserved for
  2066. * movable at startup. This will force kernel allocations
  2067. * to reserve their blocks rather than leaking throughout
  2068. * the address space during boot when many long-lived
  2069. * kernel allocations are made
  2070. */
  2071. set_pageblock_migratetype(page, MIGRATE_MOVABLE);
  2072. INIT_LIST_HEAD(&page->lru);
  2073. #ifdef WANT_PAGE_VIRTUAL
  2074. /* The shift won't overflow because ZONE_NORMAL is below 4G. */
  2075. if (!is_highmem_idx(zone))
  2076. set_page_address(page, __va(pfn << PAGE_SHIFT));
  2077. #endif
  2078. }
  2079. }
  2080. static void __meminit zone_init_free_lists(struct pglist_data *pgdat,
  2081. struct zone *zone, unsigned long size)
  2082. {
  2083. int order, t;
  2084. for_each_migratetype_order(order, t) {
  2085. INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
  2086. zone->free_area[order].nr_free = 0;
  2087. }
  2088. }
  2089. #ifndef __HAVE_ARCH_MEMMAP_INIT
  2090. #define memmap_init(size, nid, zone, start_pfn) \
  2091. memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
  2092. #endif
  2093. static int __devinit zone_batchsize(struct zone *zone)
  2094. {
  2095. int batch;
  2096. /*
  2097. * The per-cpu-pages pools are set to around 1000th of the
  2098. * size of the zone. But no more than 1/2 of a meg.
  2099. *
  2100. * OK, so we don't know how big the cache is. So guess.
  2101. */
  2102. batch = zone->present_pages / 1024;
  2103. if (batch * PAGE_SIZE > 512 * 1024)
  2104. batch = (512 * 1024) / PAGE_SIZE;
  2105. batch /= 4; /* We effectively *= 4 below */
  2106. if (batch < 1)
  2107. batch = 1;
  2108. /*
  2109. * Clamp the batch to a 2^n - 1 value. Having a power
  2110. * of 2 value was found to be more likely to have
  2111. * suboptimal cache aliasing properties in some cases.
  2112. *
  2113. * For example if 2 tasks are alternately allocating
  2114. * batches of pages, one task can end up with a lot
  2115. * of pages of one half of the possible page colors
  2116. * and the other with pages of the other colors.
  2117. */
  2118. batch = (1 << (fls(batch + batch/2)-1)) - 1;
  2119. return batch;
  2120. }
  2121. inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
  2122. {
  2123. struct per_cpu_pages *pcp;
  2124. memset(p, 0, sizeof(*p));
  2125. pcp = &p->pcp[0]; /* hot */
  2126. pcp->count = 0;
  2127. pcp->high = 6 * batch;
  2128. pcp->batch = max(1UL, 1 * batch);
  2129. INIT_LIST_HEAD(&pcp->list);
  2130. pcp = &p->pcp[1]; /* cold*/
  2131. pcp->count = 0;
  2132. pcp->high = 2 * batch;
  2133. pcp->batch = max(1UL, batch/2);
  2134. INIT_LIST_HEAD(&pcp->list);
  2135. }
  2136. /*
  2137. * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
  2138. * to the value high for the pageset p.
  2139. */
  2140. static void setup_pagelist_highmark(struct per_cpu_pageset *p,
  2141. unsigned long high)
  2142. {
  2143. struct per_cpu_pages *pcp;
  2144. pcp = &p->pcp[0]; /* hot list */
  2145. pcp->high = high;
  2146. pcp->batch = max(1UL, high/4);
  2147. if ((high/4) > (PAGE_SHIFT * 8))
  2148. pcp->batch = PAGE_SHIFT * 8;
  2149. }
  2150. #ifdef CONFIG_NUMA
  2151. /*
  2152. * Boot pageset table. One per cpu which is going to be used for all
  2153. * zones and all nodes. The parameters will be set in such a way
  2154. * that an item put on a list will immediately be handed over to
  2155. * the buddy list. This is safe since pageset manipulation is done
  2156. * with interrupts disabled.
  2157. *
  2158. * Some NUMA counter updates may also be caught by the boot pagesets.
  2159. *
  2160. * The boot_pagesets must be kept even after bootup is complete for
  2161. * unused processors and/or zones. They do play a role for bootstrapping
  2162. * hotplugged processors.
  2163. *
  2164. * zoneinfo_show() and maybe other functions do
  2165. * not check if the processor is online before following the pageset pointer.
  2166. * Other parts of the kernel may not check if the zone is available.
  2167. */
  2168. static struct per_cpu_pageset boot_pageset[NR_CPUS];
  2169. /*
  2170. * Dynamically allocate memory for the
  2171. * per cpu pageset array in struct zone.
  2172. */
  2173. static int __cpuinit process_zones(int cpu)
  2174. {
  2175. struct zone *zone, *dzone;
  2176. int node = cpu_to_node(cpu);
  2177. node_set_state(node, N_CPU); /* this node has a cpu */
  2178. for_each_zone(zone) {
  2179. if (!populated_zone(zone))
  2180. continue;
  2181. zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
  2182. GFP_KERNEL, node);
  2183. if (!zone_pcp(zone, cpu))
  2184. goto bad;
  2185. setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
  2186. if (percpu_pagelist_fraction)
  2187. setup_pagelist_highmark(zone_pcp(zone, cpu),
  2188. (zone->present_pages / percpu_pagelist_fraction));
  2189. }
  2190. return 0;
  2191. bad:
  2192. for_each_zone(dzone) {
  2193. if (!populated_zone(dzone))
  2194. continue;
  2195. if (dzone == zone)
  2196. break;
  2197. kfree(zone_pcp(dzone, cpu));
  2198. zone_pcp(dzone, cpu) = NULL;
  2199. }
  2200. return -ENOMEM;
  2201. }
  2202. static inline void free_zone_pagesets(int cpu)
  2203. {
  2204. struct zone *zone;
  2205. for_each_zone(zone) {
  2206. struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
  2207. /* Free per_cpu_pageset if it is slab allocated */
  2208. if (pset != &boot_pageset[cpu])
  2209. kfree(pset);
  2210. zone_pcp(zone, cpu) = NULL;
  2211. }
  2212. }
  2213. static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
  2214. unsigned long action,
  2215. void *hcpu)
  2216. {
  2217. int cpu = (long)hcpu;
  2218. int ret = NOTIFY_OK;
  2219. switch (action) {
  2220. case CPU_UP_PREPARE:
  2221. case CPU_UP_PREPARE_FROZEN:
  2222. if (process_zones(cpu))
  2223. ret = NOTIFY_BAD;
  2224. break;
  2225. case CPU_UP_CANCELED:
  2226. case CPU_UP_CANCELED_FROZEN:
  2227. case CPU_DEAD:
  2228. case CPU_DEAD_FROZEN:
  2229. free_zone_pagesets(cpu);
  2230. break;
  2231. default:
  2232. break;
  2233. }
  2234. return ret;
  2235. }
  2236. static struct notifier_block __cpuinitdata pageset_notifier =
  2237. { &pageset_cpuup_callback, NULL, 0 };
  2238. void __init setup_per_cpu_pageset(void)
  2239. {
  2240. int err;
  2241. /* Initialize per_cpu_pageset for cpu 0.
  2242. * A cpuup callback will do this for every cpu
  2243. * as it comes online
  2244. */
  2245. err = process_zones(smp_processor_id());
  2246. BUG_ON(err);
  2247. register_cpu_notifier(&pageset_notifier);
  2248. }
  2249. #endif
  2250. static noinline __init_refok
  2251. int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
  2252. {
  2253. int i;
  2254. struct pglist_data *pgdat = zone->zone_pgdat;
  2255. size_t alloc_size;
  2256. /*
  2257. * The per-page waitqueue mechanism uses hashed waitqueues
  2258. * per zone.
  2259. */
  2260. zone->wait_table_hash_nr_entries =
  2261. wait_table_hash_nr_entries(zone_size_pages);
  2262. zone->wait_table_bits =
  2263. wait_table_bits(zone->wait_table_hash_nr_entries);
  2264. alloc_size = zone->wait_table_hash_nr_entries
  2265. * sizeof(wait_queue_head_t);
  2266. if (system_state == SYSTEM_BOOTING) {
  2267. zone->wait_table = (wait_queue_head_t *)
  2268. alloc_bootmem_node(pgdat, alloc_size);
  2269. } else {
  2270. /*
  2271. * This case means that a zone whose size was 0 gets new memory
  2272. * via memory hot-add.
  2273. * But it may be the case that a new node was hot-added. In
  2274. * this case vmalloc() will not be able to use this new node's
  2275. * memory - this wait_table must be initialized to use this new
  2276. * node itself as well.
  2277. * To use this new node's memory, further consideration will be
  2278. * necessary.
  2279. */
  2280. zone->wait_table = vmalloc(alloc_size);
  2281. }
  2282. if (!zone->wait_table)
  2283. return -ENOMEM;
  2284. for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
  2285. init_waitqueue_head(zone->wait_table + i);
  2286. return 0;
  2287. }
  2288. static __meminit void zone_pcp_init(struct zone *zone)
  2289. {
  2290. int cpu;
  2291. unsigned long batch = zone_batchsize(zone);
  2292. for (cpu = 0; cpu < NR_CPUS; cpu++) {
  2293. #ifdef CONFIG_NUMA
  2294. /* Early boot. Slab allocator not functional yet */
  2295. zone_pcp(zone, cpu) = &boot_pageset[cpu];
  2296. setup_pageset(&boot_pageset[cpu],0);
  2297. #else
  2298. setup_pageset(zone_pcp(zone,cpu), batch);
  2299. #endif
  2300. }
  2301. if (zone->present_pages)
  2302. printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
  2303. zone->name, zone->present_pages, batch);
  2304. }
  2305. __meminit int init_currently_empty_zone(struct zone *zone,
  2306. unsigned long zone_start_pfn,
  2307. unsigned long size,
  2308. enum memmap_context context)
  2309. {
  2310. struct pglist_data *pgdat = zone->zone_pgdat;
  2311. int ret;
  2312. ret = zone_wait_table_init(zone, size);
  2313. if (ret)
  2314. return ret;
  2315. pgdat->nr_zones = zone_idx(zone) + 1;
  2316. zone->zone_start_pfn = zone_start_pfn;
  2317. memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
  2318. zone_init_free_lists(pgdat, zone, zone->spanned_pages);
  2319. return 0;
  2320. }
  2321. #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
  2322. /*
  2323. * Basic iterator support. Return the first range of PFNs for a node
  2324. * Note: nid == MAX_NUMNODES returns first region regardless of node
  2325. */
  2326. static int __meminit first_active_region_index_in_nid(int nid)
  2327. {
  2328. int i;
  2329. for (i = 0; i < nr_nodemap_entries; i++)
  2330. if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
  2331. return i;
  2332. return -1;
  2333. }
  2334. /*
  2335. * Basic iterator support. Return the next active range of PFNs for a node
  2336. * Note: nid == MAX_NUMNODES returns next region regardles of node
  2337. */
  2338. static int __meminit next_active_region_index_in_nid(int index, int nid)
  2339. {
  2340. for (index = index + 1; index < nr_nodemap_entries; index++)
  2341. if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
  2342. return index;
  2343. return -1;
  2344. }
  2345. #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
  2346. /*
  2347. * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
  2348. * Architectures may implement their own version but if add_active_range()
  2349. * was used and there are no special requirements, this is a convenient
  2350. * alternative
  2351. */
  2352. int __meminit early_pfn_to_nid(unsigned long pfn)
  2353. {
  2354. int i;
  2355. for (i = 0; i < nr_nodemap_entries; i++) {
  2356. unsigned long start_pfn = early_node_map[i].start_pfn;
  2357. unsigned long end_pfn = early_node_map[i].end_pfn;
  2358. if (start_pfn <= pfn && pfn < end_pfn)
  2359. return early_node_map[i].nid;
  2360. }
  2361. return 0;
  2362. }
  2363. #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
  2364. /* Basic iterator support to walk early_node_map[] */
  2365. #define for_each_active_range_index_in_nid(i, nid) \
  2366. for (i = first_active_region_index_in_nid(nid); i != -1; \
  2367. i = next_active_region_index_in_nid(i, nid))
  2368. /**
  2369. * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
  2370. * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
  2371. * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
  2372. *
  2373. * If an architecture guarantees that all ranges registered with
  2374. * add_active_ranges() contain no holes and may be freed, this
  2375. * this function may be used instead of calling free_bootmem() manually.
  2376. */
  2377. void __init free_bootmem_with_active_regions(int nid,
  2378. unsigned long max_low_pfn)
  2379. {
  2380. int i;
  2381. for_each_active_range_index_in_nid(i, nid) {
  2382. unsigned long size_pages = 0;
  2383. unsigned long end_pfn = early_node_map[i].end_pfn;
  2384. if (early_node_map[i].start_pfn >= max_low_pfn)
  2385. continue;
  2386. if (end_pfn > max_low_pfn)
  2387. end_pfn = max_low_pfn;
  2388. size_pages = end_pfn - early_node_map[i].start_pfn;
  2389. free_bootmem_node(NODE_DATA(early_node_map[i].nid),
  2390. PFN_PHYS(early_node_map[i].start_pfn),
  2391. size_pages << PAGE_SHIFT);
  2392. }
  2393. }
  2394. /**
  2395. * sparse_memory_present_with_active_regions - Call memory_present for each active range
  2396. * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
  2397. *
  2398. * If an architecture guarantees that all ranges registered with
  2399. * add_active_ranges() contain no holes and may be freed, this
  2400. * function may be used instead of calling memory_present() manually.
  2401. */
  2402. void __init sparse_memory_present_with_active_regions(int nid)
  2403. {
  2404. int i;
  2405. for_each_active_range_index_in_nid(i, nid)
  2406. memory_present(early_node_map[i].nid,
  2407. early_node_map[i].start_pfn,
  2408. early_node_map[i].end_pfn);
  2409. }
  2410. /**
  2411. * push_node_boundaries - Push node boundaries to at least the requested boundary
  2412. * @nid: The nid of the node to push the boundary for
  2413. * @start_pfn: The start pfn of the node
  2414. * @end_pfn: The end pfn of the node
  2415. *
  2416. * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
  2417. * time. Specifically, on x86_64, SRAT will report ranges that can potentially
  2418. * be hotplugged even though no physical memory exists. This function allows
  2419. * an arch to push out the node boundaries so mem_map is allocated that can
  2420. * be used later.
  2421. */
  2422. #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
  2423. void __init push_node_boundaries(unsigned int nid,
  2424. unsigned long start_pfn, unsigned long end_pfn)
  2425. {
  2426. printk(KERN_DEBUG "Entering push_node_boundaries(%u, %lu, %lu)\n",
  2427. nid, start_pfn, end_pfn);
  2428. /* Initialise the boundary for this node if necessary */
  2429. if (node_boundary_end_pfn[nid] == 0)
  2430. node_boundary_start_pfn[nid] = -1UL;
  2431. /* Update the boundaries */
  2432. if (node_boundary_start_pfn[nid] > start_pfn)
  2433. node_boundary_start_pfn[nid] = start_pfn;
  2434. if (node_boundary_end_pfn[nid] < end_pfn)
  2435. node_boundary_end_pfn[nid] = end_pfn;
  2436. }
  2437. /* If necessary, push the node boundary out for reserve hotadd */
  2438. static void __meminit account_node_boundary(unsigned int nid,
  2439. unsigned long *start_pfn, unsigned long *end_pfn)
  2440. {
  2441. printk(KERN_DEBUG "Entering account_node_boundary(%u, %lu, %lu)\n",
  2442. nid, *start_pfn, *end_pfn);
  2443. /* Return if boundary information has not been provided */
  2444. if (node_boundary_end_pfn[nid] == 0)
  2445. return;
  2446. /* Check the boundaries and update if necessary */
  2447. if (node_boundary_start_pfn[nid] < *start_pfn)
  2448. *start_pfn = node_boundary_start_pfn[nid];
  2449. if (node_boundary_end_pfn[nid] > *end_pfn)
  2450. *end_pfn = node_boundary_end_pfn[nid];
  2451. }
  2452. #else
  2453. void __init push_node_boundaries(unsigned int nid,
  2454. unsigned long start_pfn, unsigned long end_pfn) {}
  2455. static void __meminit account_node_boundary(unsigned int nid,
  2456. unsigned long *start_pfn, unsigned long *end_pfn) {}
  2457. #endif
  2458. /**
  2459. * get_pfn_range_for_nid - Return the start and end page frames for a node
  2460. * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
  2461. * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
  2462. * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
  2463. *
  2464. * It returns the start and end page frame of a node based on information
  2465. * provided by an arch calling add_active_range(). If called for a node
  2466. * with no available memory, a warning is printed and the start and end
  2467. * PFNs will be 0.
  2468. */
  2469. void __meminit get_pfn_range_for_nid(unsigned int nid,
  2470. unsigned long *start_pfn, unsigned long *end_pfn)
  2471. {
  2472. int i;
  2473. *start_pfn = -1UL;
  2474. *end_pfn = 0;
  2475. for_each_active_range_index_in_nid(i, nid) {
  2476. *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
  2477. *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
  2478. }
  2479. if (*start_pfn == -1UL)
  2480. *start_pfn = 0;
  2481. /* Push the node boundaries out if requested */
  2482. account_node_boundary(nid, start_pfn, end_pfn);
  2483. }
  2484. /*
  2485. * This finds a zone that can be used for ZONE_MOVABLE pages. The
  2486. * assumption is made that zones within a node are ordered in monotonic
  2487. * increasing memory addresses so that the "highest" populated zone is used
  2488. */
  2489. void __init find_usable_zone_for_movable(void)
  2490. {
  2491. int zone_index;
  2492. for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
  2493. if (zone_index == ZONE_MOVABLE)
  2494. continue;
  2495. if (arch_zone_highest_possible_pfn[zone_index] >
  2496. arch_zone_lowest_possible_pfn[zone_index])
  2497. break;
  2498. }
  2499. VM_BUG_ON(zone_index == -1);
  2500. movable_zone = zone_index;
  2501. }
  2502. /*
  2503. * The zone ranges provided by the architecture do not include ZONE_MOVABLE
  2504. * because it is sized independant of architecture. Unlike the other zones,
  2505. * the starting point for ZONE_MOVABLE is not fixed. It may be different
  2506. * in each node depending on the size of each node and how evenly kernelcore
  2507. * is distributed. This helper function adjusts the zone ranges
  2508. * provided by the architecture for a given node by using the end of the
  2509. * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
  2510. * zones within a node are in order of monotonic increases memory addresses
  2511. */
  2512. void __meminit adjust_zone_range_for_zone_movable(int nid,
  2513. unsigned long zone_type,
  2514. unsigned long node_start_pfn,
  2515. unsigned long node_end_pfn,
  2516. unsigned long *zone_start_pfn,
  2517. unsigned long *zone_end_pfn)
  2518. {
  2519. /* Only adjust if ZONE_MOVABLE is on this node */
  2520. if (zone_movable_pfn[nid]) {
  2521. /* Size ZONE_MOVABLE */
  2522. if (zone_type == ZONE_MOVABLE) {
  2523. *zone_start_pfn = zone_movable_pfn[nid];
  2524. *zone_end_pfn = min(node_end_pfn,
  2525. arch_zone_highest_possible_pfn[movable_zone]);
  2526. /* Adjust for ZONE_MOVABLE starting within this range */
  2527. } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
  2528. *zone_end_pfn > zone_movable_pfn[nid]) {
  2529. *zone_end_pfn = zone_movable_pfn[nid];
  2530. /* Check if this whole range is within ZONE_MOVABLE */
  2531. } else if (*zone_start_pfn >= zone_movable_pfn[nid])
  2532. *zone_start_pfn = *zone_end_pfn;
  2533. }
  2534. }
  2535. /*
  2536. * Return the number of pages a zone spans in a node, including holes
  2537. * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
  2538. */
  2539. static unsigned long __meminit zone_spanned_pages_in_node(int nid,
  2540. unsigned long zone_type,
  2541. unsigned long *ignored)
  2542. {
  2543. unsigned long node_start_pfn, node_end_pfn;
  2544. unsigned long zone_start_pfn, zone_end_pfn;
  2545. /* Get the start and end of the node and zone */
  2546. get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
  2547. zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
  2548. zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
  2549. adjust_zone_range_for_zone_movable(nid, zone_type,
  2550. node_start_pfn, node_end_pfn,
  2551. &zone_start_pfn, &zone_end_pfn);
  2552. /* Check that this node has pages within the zone's required range */
  2553. if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
  2554. return 0;
  2555. /* Move the zone boundaries inside the node if necessary */
  2556. zone_end_pfn = min(zone_end_pfn, node_end_pfn);
  2557. zone_start_pfn = max(zone_start_pfn, node_start_pfn);
  2558. /* Return the spanned pages */
  2559. return zone_end_pfn - zone_start_pfn;
  2560. }
  2561. /*
  2562. * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
  2563. * then all holes in the requested range will be accounted for.
  2564. */
  2565. unsigned long __meminit __absent_pages_in_range(int nid,
  2566. unsigned long range_start_pfn,
  2567. unsigned long range_end_pfn)
  2568. {
  2569. int i = 0;
  2570. unsigned long prev_end_pfn = 0, hole_pages = 0;
  2571. unsigned long start_pfn;
  2572. /* Find the end_pfn of the first active range of pfns in the node */
  2573. i = first_active_region_index_in_nid(nid);
  2574. if (i == -1)
  2575. return 0;
  2576. prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
  2577. /* Account for ranges before physical memory on this node */
  2578. if (early_node_map[i].start_pfn > range_start_pfn)
  2579. hole_pages = prev_end_pfn - range_start_pfn;
  2580. /* Find all holes for the zone within the node */
  2581. for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
  2582. /* No need to continue if prev_end_pfn is outside the zone */
  2583. if (prev_end_pfn >= range_end_pfn)
  2584. break;
  2585. /* Make sure the end of the zone is not within the hole */
  2586. start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
  2587. prev_end_pfn = max(prev_end_pfn, range_start_pfn);
  2588. /* Update the hole size cound and move on */
  2589. if (start_pfn > range_start_pfn) {
  2590. BUG_ON(prev_end_pfn > start_pfn);
  2591. hole_pages += start_pfn - prev_end_pfn;
  2592. }
  2593. prev_end_pfn = early_node_map[i].end_pfn;
  2594. }
  2595. /* Account for ranges past physical memory on this node */
  2596. if (range_end_pfn > prev_end_pfn)
  2597. hole_pages += range_end_pfn -
  2598. max(range_start_pfn, prev_end_pfn);
  2599. return hole_pages;
  2600. }
  2601. /**
  2602. * absent_pages_in_range - Return number of page frames in holes within a range
  2603. * @start_pfn: The start PFN to start searching for holes
  2604. * @end_pfn: The end PFN to stop searching for holes
  2605. *
  2606. * It returns the number of pages frames in memory holes within a range.
  2607. */
  2608. unsigned long __init absent_pages_in_range(unsigned long start_pfn,
  2609. unsigned long end_pfn)
  2610. {
  2611. return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
  2612. }
  2613. /* Return the number of page frames in holes in a zone on a node */
  2614. static unsigned long __meminit zone_absent_pages_in_node(int nid,
  2615. unsigned long zone_type,
  2616. unsigned long *ignored)
  2617. {
  2618. unsigned long node_start_pfn, node_end_pfn;
  2619. unsigned long zone_start_pfn, zone_end_pfn;
  2620. get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
  2621. zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
  2622. node_start_pfn);
  2623. zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
  2624. node_end_pfn);
  2625. adjust_zone_range_for_zone_movable(nid, zone_type,
  2626. node_start_pfn, node_end_pfn,
  2627. &zone_start_pfn, &zone_end_pfn);
  2628. return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
  2629. }
  2630. #else
  2631. static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
  2632. unsigned long zone_type,
  2633. unsigned long *zones_size)
  2634. {
  2635. return zones_size[zone_type];
  2636. }
  2637. static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
  2638. unsigned long zone_type,
  2639. unsigned long *zholes_size)
  2640. {
  2641. if (!zholes_size)
  2642. return 0;
  2643. return zholes_size[zone_type];
  2644. }
  2645. #endif
  2646. static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
  2647. unsigned long *zones_size, unsigned long *zholes_size)
  2648. {
  2649. unsigned long realtotalpages, totalpages = 0;
  2650. enum zone_type i;
  2651. for (i = 0; i < MAX_NR_ZONES; i++)
  2652. totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
  2653. zones_size);
  2654. pgdat->node_spanned_pages = totalpages;
  2655. realtotalpages = totalpages;
  2656. for (i = 0; i < MAX_NR_ZONES; i++)
  2657. realtotalpages -=
  2658. zone_absent_pages_in_node(pgdat->node_id, i,
  2659. zholes_size);
  2660. pgdat->node_present_pages = realtotalpages;
  2661. printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
  2662. realtotalpages);
  2663. }
  2664. #ifndef CONFIG_SPARSEMEM
  2665. /*
  2666. * Calculate the size of the zone->blockflags rounded to an unsigned long
  2667. * Start by making sure zonesize is a multiple of MAX_ORDER-1 by rounding up
  2668. * Then figure 1 NR_PAGEBLOCK_BITS worth of bits per MAX_ORDER-1, finally
  2669. * round what is now in bits to nearest long in bits, then return it in
  2670. * bytes.
  2671. */
  2672. static unsigned long __init usemap_size(unsigned long zonesize)
  2673. {
  2674. unsigned long usemapsize;
  2675. usemapsize = roundup(zonesize, MAX_ORDER_NR_PAGES);
  2676. usemapsize = usemapsize >> (MAX_ORDER-1);
  2677. usemapsize *= NR_PAGEBLOCK_BITS;
  2678. usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
  2679. return usemapsize / 8;
  2680. }
  2681. static void __init setup_usemap(struct pglist_data *pgdat,
  2682. struct zone *zone, unsigned long zonesize)
  2683. {
  2684. unsigned long usemapsize = usemap_size(zonesize);
  2685. zone->pageblock_flags = NULL;
  2686. if (usemapsize) {
  2687. zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
  2688. memset(zone->pageblock_flags, 0, usemapsize);
  2689. }
  2690. }
  2691. #else
  2692. static void inline setup_usemap(struct pglist_data *pgdat,
  2693. struct zone *zone, unsigned long zonesize) {}
  2694. #endif /* CONFIG_SPARSEMEM */
  2695. /*
  2696. * Set up the zone data structures:
  2697. * - mark all pages reserved
  2698. * - mark all memory queues empty
  2699. * - clear the memory bitmaps
  2700. */
  2701. static void __meminit free_area_init_core(struct pglist_data *pgdat,
  2702. unsigned long *zones_size, unsigned long *zholes_size)
  2703. {
  2704. enum zone_type j;
  2705. int nid = pgdat->node_id;
  2706. unsigned long zone_start_pfn = pgdat->node_start_pfn;
  2707. int ret;
  2708. pgdat_resize_init(pgdat);
  2709. pgdat->nr_zones = 0;
  2710. init_waitqueue_head(&pgdat->kswapd_wait);
  2711. pgdat->kswapd_max_order = 0;
  2712. for (j = 0; j < MAX_NR_ZONES; j++) {
  2713. struct zone *zone = pgdat->node_zones + j;
  2714. unsigned long size, realsize, memmap_pages;
  2715. size = zone_spanned_pages_in_node(nid, j, zones_size);
  2716. realsize = size - zone_absent_pages_in_node(nid, j,
  2717. zholes_size);
  2718. /*
  2719. * Adjust realsize so that it accounts for how much memory
  2720. * is used by this zone for memmap. This affects the watermark
  2721. * and per-cpu initialisations
  2722. */
  2723. memmap_pages = (size * sizeof(struct page)) >> PAGE_SHIFT;
  2724. if (realsize >= memmap_pages) {
  2725. realsize -= memmap_pages;
  2726. printk(KERN_DEBUG
  2727. " %s zone: %lu pages used for memmap\n",
  2728. zone_names[j], memmap_pages);
  2729. } else
  2730. printk(KERN_WARNING
  2731. " %s zone: %lu pages exceeds realsize %lu\n",
  2732. zone_names[j], memmap_pages, realsize);
  2733. /* Account for reserved pages */
  2734. if (j == 0 && realsize > dma_reserve) {
  2735. realsize -= dma_reserve;
  2736. printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
  2737. zone_names[0], dma_reserve);
  2738. }
  2739. if (!is_highmem_idx(j))
  2740. nr_kernel_pages += realsize;
  2741. nr_all_pages += realsize;
  2742. zone->spanned_pages = size;
  2743. zone->present_pages = realsize;
  2744. #ifdef CONFIG_NUMA
  2745. zone->node = nid;
  2746. zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
  2747. / 100;
  2748. zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
  2749. #endif
  2750. zone->name = zone_names[j];
  2751. spin_lock_init(&zone->lock);
  2752. spin_lock_init(&zone->lru_lock);
  2753. zone_seqlock_init(zone);
  2754. zone->zone_pgdat = pgdat;
  2755. zone->prev_priority = DEF_PRIORITY;
  2756. zone_pcp_init(zone);
  2757. INIT_LIST_HEAD(&zone->active_list);
  2758. INIT_LIST_HEAD(&zone->inactive_list);
  2759. zone->nr_scan_active = 0;
  2760. zone->nr_scan_inactive = 0;
  2761. zap_zone_vm_stats(zone);
  2762. atomic_set(&zone->reclaim_in_progress, 0);
  2763. if (!size)
  2764. continue;
  2765. setup_usemap(pgdat, zone, size);
  2766. ret = init_currently_empty_zone(zone, zone_start_pfn,
  2767. size, MEMMAP_EARLY);
  2768. BUG_ON(ret);
  2769. zone_start_pfn += size;
  2770. }
  2771. }
  2772. static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
  2773. {
  2774. /* Skip empty nodes */
  2775. if (!pgdat->node_spanned_pages)
  2776. return;
  2777. #ifdef CONFIG_FLAT_NODE_MEM_MAP
  2778. /* ia64 gets its own node_mem_map, before this, without bootmem */
  2779. if (!pgdat->node_mem_map) {
  2780. unsigned long size, start, end;
  2781. struct page *map;
  2782. /*
  2783. * The zone's endpoints aren't required to be MAX_ORDER
  2784. * aligned but the node_mem_map endpoints must be in order
  2785. * for the buddy allocator to function correctly.
  2786. */
  2787. start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
  2788. end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
  2789. end = ALIGN(end, MAX_ORDER_NR_PAGES);
  2790. size = (end - start) * sizeof(struct page);
  2791. map = alloc_remap(pgdat->node_id, size);
  2792. if (!map)
  2793. map = alloc_bootmem_node(pgdat, size);
  2794. pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
  2795. }
  2796. #ifndef CONFIG_NEED_MULTIPLE_NODES
  2797. /*
  2798. * With no DISCONTIG, the global mem_map is just set as node 0's
  2799. */
  2800. if (pgdat == NODE_DATA(0)) {
  2801. mem_map = NODE_DATA(0)->node_mem_map;
  2802. #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
  2803. if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
  2804. mem_map -= pgdat->node_start_pfn;
  2805. #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
  2806. }
  2807. #endif
  2808. #endif /* CONFIG_FLAT_NODE_MEM_MAP */
  2809. }
  2810. void __meminit free_area_init_node(int nid, struct pglist_data *pgdat,
  2811. unsigned long *zones_size, unsigned long node_start_pfn,
  2812. unsigned long *zholes_size)
  2813. {
  2814. pgdat->node_id = nid;
  2815. pgdat->node_start_pfn = node_start_pfn;
  2816. calculate_node_totalpages(pgdat, zones_size, zholes_size);
  2817. alloc_node_mem_map(pgdat);
  2818. free_area_init_core(pgdat, zones_size, zholes_size);
  2819. }
  2820. #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
  2821. #if MAX_NUMNODES > 1
  2822. /*
  2823. * Figure out the number of possible node ids.
  2824. */
  2825. static void __init setup_nr_node_ids(void)
  2826. {
  2827. unsigned int node;
  2828. unsigned int highest = 0;
  2829. for_each_node_mask(node, node_possible_map)
  2830. highest = node;
  2831. nr_node_ids = highest + 1;
  2832. }
  2833. #else
  2834. static inline void setup_nr_node_ids(void)
  2835. {
  2836. }
  2837. #endif
  2838. /**
  2839. * add_active_range - Register a range of PFNs backed by physical memory
  2840. * @nid: The node ID the range resides on
  2841. * @start_pfn: The start PFN of the available physical memory
  2842. * @end_pfn: The end PFN of the available physical memory
  2843. *
  2844. * These ranges are stored in an early_node_map[] and later used by
  2845. * free_area_init_nodes() to calculate zone sizes and holes. If the
  2846. * range spans a memory hole, it is up to the architecture to ensure
  2847. * the memory is not freed by the bootmem allocator. If possible
  2848. * the range being registered will be merged with existing ranges.
  2849. */
  2850. void __init add_active_range(unsigned int nid, unsigned long start_pfn,
  2851. unsigned long end_pfn)
  2852. {
  2853. int i;
  2854. printk(KERN_DEBUG "Entering add_active_range(%d, %lu, %lu) "
  2855. "%d entries of %d used\n",
  2856. nid, start_pfn, end_pfn,
  2857. nr_nodemap_entries, MAX_ACTIVE_REGIONS);
  2858. /* Merge with existing active regions if possible */
  2859. for (i = 0; i < nr_nodemap_entries; i++) {
  2860. if (early_node_map[i].nid != nid)
  2861. continue;
  2862. /* Skip if an existing region covers this new one */
  2863. if (start_pfn >= early_node_map[i].start_pfn &&
  2864. end_pfn <= early_node_map[i].end_pfn)
  2865. return;
  2866. /* Merge forward if suitable */
  2867. if (start_pfn <= early_node_map[i].end_pfn &&
  2868. end_pfn > early_node_map[i].end_pfn) {
  2869. early_node_map[i].end_pfn = end_pfn;
  2870. return;
  2871. }
  2872. /* Merge backward if suitable */
  2873. if (start_pfn < early_node_map[i].end_pfn &&
  2874. end_pfn >= early_node_map[i].start_pfn) {
  2875. early_node_map[i].start_pfn = start_pfn;
  2876. return;
  2877. }
  2878. }
  2879. /* Check that early_node_map is large enough */
  2880. if (i >= MAX_ACTIVE_REGIONS) {
  2881. printk(KERN_CRIT "More than %d memory regions, truncating\n",
  2882. MAX_ACTIVE_REGIONS);
  2883. return;
  2884. }
  2885. early_node_map[i].nid = nid;
  2886. early_node_map[i].start_pfn = start_pfn;
  2887. early_node_map[i].end_pfn = end_pfn;
  2888. nr_nodemap_entries = i + 1;
  2889. }
  2890. /**
  2891. * shrink_active_range - Shrink an existing registered range of PFNs
  2892. * @nid: The node id the range is on that should be shrunk
  2893. * @old_end_pfn: The old end PFN of the range
  2894. * @new_end_pfn: The new PFN of the range
  2895. *
  2896. * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
  2897. * The map is kept at the end physical page range that has already been
  2898. * registered with add_active_range(). This function allows an arch to shrink
  2899. * an existing registered range.
  2900. */
  2901. void __init shrink_active_range(unsigned int nid, unsigned long old_end_pfn,
  2902. unsigned long new_end_pfn)
  2903. {
  2904. int i;
  2905. /* Find the old active region end and shrink */
  2906. for_each_active_range_index_in_nid(i, nid)
  2907. if (early_node_map[i].end_pfn == old_end_pfn) {
  2908. early_node_map[i].end_pfn = new_end_pfn;
  2909. break;
  2910. }
  2911. }
  2912. /**
  2913. * remove_all_active_ranges - Remove all currently registered regions
  2914. *
  2915. * During discovery, it may be found that a table like SRAT is invalid
  2916. * and an alternative discovery method must be used. This function removes
  2917. * all currently registered regions.
  2918. */
  2919. void __init remove_all_active_ranges(void)
  2920. {
  2921. memset(early_node_map, 0, sizeof(early_node_map));
  2922. nr_nodemap_entries = 0;
  2923. #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
  2924. memset(node_boundary_start_pfn, 0, sizeof(node_boundary_start_pfn));
  2925. memset(node_boundary_end_pfn, 0, sizeof(node_boundary_end_pfn));
  2926. #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
  2927. }
  2928. /* Compare two active node_active_regions */
  2929. static int __init cmp_node_active_region(const void *a, const void *b)
  2930. {
  2931. struct node_active_region *arange = (struct node_active_region *)a;
  2932. struct node_active_region *brange = (struct node_active_region *)b;
  2933. /* Done this way to avoid overflows */
  2934. if (arange->start_pfn > brange->start_pfn)
  2935. return 1;
  2936. if (arange->start_pfn < brange->start_pfn)
  2937. return -1;
  2938. return 0;
  2939. }
  2940. /* sort the node_map by start_pfn */
  2941. static void __init sort_node_map(void)
  2942. {
  2943. sort(early_node_map, (size_t)nr_nodemap_entries,
  2944. sizeof(struct node_active_region),
  2945. cmp_node_active_region, NULL);
  2946. }
  2947. /* Find the lowest pfn for a node */
  2948. unsigned long __init find_min_pfn_for_node(unsigned long nid)
  2949. {
  2950. int i;
  2951. unsigned long min_pfn = ULONG_MAX;
  2952. /* Assuming a sorted map, the first range found has the starting pfn */
  2953. for_each_active_range_index_in_nid(i, nid)
  2954. min_pfn = min(min_pfn, early_node_map[i].start_pfn);
  2955. if (min_pfn == ULONG_MAX) {
  2956. printk(KERN_WARNING
  2957. "Could not find start_pfn for node %lu\n", nid);
  2958. return 0;
  2959. }
  2960. return min_pfn;
  2961. }
  2962. /**
  2963. * find_min_pfn_with_active_regions - Find the minimum PFN registered
  2964. *
  2965. * It returns the minimum PFN based on information provided via
  2966. * add_active_range().
  2967. */
  2968. unsigned long __init find_min_pfn_with_active_regions(void)
  2969. {
  2970. return find_min_pfn_for_node(MAX_NUMNODES);
  2971. }
  2972. /**
  2973. * find_max_pfn_with_active_regions - Find the maximum PFN registered
  2974. *
  2975. * It returns the maximum PFN based on information provided via
  2976. * add_active_range().
  2977. */
  2978. unsigned long __init find_max_pfn_with_active_regions(void)
  2979. {
  2980. int i;
  2981. unsigned long max_pfn = 0;
  2982. for (i = 0; i < nr_nodemap_entries; i++)
  2983. max_pfn = max(max_pfn, early_node_map[i].end_pfn);
  2984. return max_pfn;
  2985. }
  2986. /*
  2987. * early_calculate_totalpages()
  2988. * Sum pages in active regions for movable zone.
  2989. * Populate N_HIGH_MEMORY for calculating usable_nodes.
  2990. */
  2991. unsigned long __init early_calculate_totalpages(void)
  2992. {
  2993. int i;
  2994. unsigned long totalpages = 0;
  2995. for (i = 0; i < nr_nodemap_entries; i++) {
  2996. unsigned long pages = early_node_map[i].end_pfn -
  2997. early_node_map[i].start_pfn;
  2998. totalpages += pages;
  2999. if (pages)
  3000. node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
  3001. }
  3002. return totalpages;
  3003. }
  3004. /*
  3005. * Find the PFN the Movable zone begins in each node. Kernel memory
  3006. * is spread evenly between nodes as long as the nodes have enough
  3007. * memory. When they don't, some nodes will have more kernelcore than
  3008. * others
  3009. */
  3010. void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
  3011. {
  3012. int i, nid;
  3013. unsigned long usable_startpfn;
  3014. unsigned long kernelcore_node, kernelcore_remaining;
  3015. unsigned long totalpages = early_calculate_totalpages();
  3016. int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
  3017. /*
  3018. * If movablecore was specified, calculate what size of
  3019. * kernelcore that corresponds so that memory usable for
  3020. * any allocation type is evenly spread. If both kernelcore
  3021. * and movablecore are specified, then the value of kernelcore
  3022. * will be used for required_kernelcore if it's greater than
  3023. * what movablecore would have allowed.
  3024. */
  3025. if (required_movablecore) {
  3026. unsigned long corepages;
  3027. /*
  3028. * Round-up so that ZONE_MOVABLE is at least as large as what
  3029. * was requested by the user
  3030. */
  3031. required_movablecore =
  3032. roundup(required_movablecore, MAX_ORDER_NR_PAGES);
  3033. corepages = totalpages - required_movablecore;
  3034. required_kernelcore = max(required_kernelcore, corepages);
  3035. }
  3036. /* If kernelcore was not specified, there is no ZONE_MOVABLE */
  3037. if (!required_kernelcore)
  3038. return;
  3039. /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
  3040. find_usable_zone_for_movable();
  3041. usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
  3042. restart:
  3043. /* Spread kernelcore memory as evenly as possible throughout nodes */
  3044. kernelcore_node = required_kernelcore / usable_nodes;
  3045. for_each_node_state(nid, N_HIGH_MEMORY) {
  3046. /*
  3047. * Recalculate kernelcore_node if the division per node
  3048. * now exceeds what is necessary to satisfy the requested
  3049. * amount of memory for the kernel
  3050. */
  3051. if (required_kernelcore < kernelcore_node)
  3052. kernelcore_node = required_kernelcore / usable_nodes;
  3053. /*
  3054. * As the map is walked, we track how much memory is usable
  3055. * by the kernel using kernelcore_remaining. When it is
  3056. * 0, the rest of the node is usable by ZONE_MOVABLE
  3057. */
  3058. kernelcore_remaining = kernelcore_node;
  3059. /* Go through each range of PFNs within this node */
  3060. for_each_active_range_index_in_nid(i, nid) {
  3061. unsigned long start_pfn, end_pfn;
  3062. unsigned long size_pages;
  3063. start_pfn = max(early_node_map[i].start_pfn,
  3064. zone_movable_pfn[nid]);
  3065. end_pfn = early_node_map[i].end_pfn;
  3066. if (start_pfn >= end_pfn)
  3067. continue;
  3068. /* Account for what is only usable for kernelcore */
  3069. if (start_pfn < usable_startpfn) {
  3070. unsigned long kernel_pages;
  3071. kernel_pages = min(end_pfn, usable_startpfn)
  3072. - start_pfn;
  3073. kernelcore_remaining -= min(kernel_pages,
  3074. kernelcore_remaining);
  3075. required_kernelcore -= min(kernel_pages,
  3076. required_kernelcore);
  3077. /* Continue if range is now fully accounted */
  3078. if (end_pfn <= usable_startpfn) {
  3079. /*
  3080. * Push zone_movable_pfn to the end so
  3081. * that if we have to rebalance
  3082. * kernelcore across nodes, we will
  3083. * not double account here
  3084. */
  3085. zone_movable_pfn[nid] = end_pfn;
  3086. continue;
  3087. }
  3088. start_pfn = usable_startpfn;
  3089. }
  3090. /*
  3091. * The usable PFN range for ZONE_MOVABLE is from
  3092. * start_pfn->end_pfn. Calculate size_pages as the
  3093. * number of pages used as kernelcore
  3094. */
  3095. size_pages = end_pfn - start_pfn;
  3096. if (size_pages > kernelcore_remaining)
  3097. size_pages = kernelcore_remaining;
  3098. zone_movable_pfn[nid] = start_pfn + size_pages;
  3099. /*
  3100. * Some kernelcore has been met, update counts and
  3101. * break if the kernelcore for this node has been
  3102. * satisified
  3103. */
  3104. required_kernelcore -= min(required_kernelcore,
  3105. size_pages);
  3106. kernelcore_remaining -= size_pages;
  3107. if (!kernelcore_remaining)
  3108. break;
  3109. }
  3110. }
  3111. /*
  3112. * If there is still required_kernelcore, we do another pass with one
  3113. * less node in the count. This will push zone_movable_pfn[nid] further
  3114. * along on the nodes that still have memory until kernelcore is
  3115. * satisified
  3116. */
  3117. usable_nodes--;
  3118. if (usable_nodes && required_kernelcore > usable_nodes)
  3119. goto restart;
  3120. /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
  3121. for (nid = 0; nid < MAX_NUMNODES; nid++)
  3122. zone_movable_pfn[nid] =
  3123. roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
  3124. }
  3125. /* Any regular memory on that node ? */
  3126. static void check_for_regular_memory(pg_data_t *pgdat)
  3127. {
  3128. #ifdef CONFIG_HIGHMEM
  3129. enum zone_type zone_type;
  3130. for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
  3131. struct zone *zone = &pgdat->node_zones[zone_type];
  3132. if (zone->present_pages)
  3133. node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
  3134. }
  3135. #endif
  3136. }
  3137. /**
  3138. * free_area_init_nodes - Initialise all pg_data_t and zone data
  3139. * @max_zone_pfn: an array of max PFNs for each zone
  3140. *
  3141. * This will call free_area_init_node() for each active node in the system.
  3142. * Using the page ranges provided by add_active_range(), the size of each
  3143. * zone in each node and their holes is calculated. If the maximum PFN
  3144. * between two adjacent zones match, it is assumed that the zone is empty.
  3145. * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
  3146. * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
  3147. * starts where the previous one ended. For example, ZONE_DMA32 starts
  3148. * at arch_max_dma_pfn.
  3149. */
  3150. void __init free_area_init_nodes(unsigned long *max_zone_pfn)
  3151. {
  3152. unsigned long nid;
  3153. enum zone_type i;
  3154. /* Sort early_node_map as initialisation assumes it is sorted */
  3155. sort_node_map();
  3156. /* Record where the zone boundaries are */
  3157. memset(arch_zone_lowest_possible_pfn, 0,
  3158. sizeof(arch_zone_lowest_possible_pfn));
  3159. memset(arch_zone_highest_possible_pfn, 0,
  3160. sizeof(arch_zone_highest_possible_pfn));
  3161. arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
  3162. arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
  3163. for (i = 1; i < MAX_NR_ZONES; i++) {
  3164. if (i == ZONE_MOVABLE)
  3165. continue;
  3166. arch_zone_lowest_possible_pfn[i] =
  3167. arch_zone_highest_possible_pfn[i-1];
  3168. arch_zone_highest_possible_pfn[i] =
  3169. max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
  3170. }
  3171. arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
  3172. arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
  3173. /* Find the PFNs that ZONE_MOVABLE begins at in each node */
  3174. memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
  3175. find_zone_movable_pfns_for_nodes(zone_movable_pfn);
  3176. /* Print out the zone ranges */
  3177. printk("Zone PFN ranges:\n");
  3178. for (i = 0; i < MAX_NR_ZONES; i++) {
  3179. if (i == ZONE_MOVABLE)
  3180. continue;
  3181. printk(" %-8s %8lu -> %8lu\n",
  3182. zone_names[i],
  3183. arch_zone_lowest_possible_pfn[i],
  3184. arch_zone_highest_possible_pfn[i]);
  3185. }
  3186. /* Print out the PFNs ZONE_MOVABLE begins at in each node */
  3187. printk("Movable zone start PFN for each node\n");
  3188. for (i = 0; i < MAX_NUMNODES; i++) {
  3189. if (zone_movable_pfn[i])
  3190. printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
  3191. }
  3192. /* Print out the early_node_map[] */
  3193. printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
  3194. for (i = 0; i < nr_nodemap_entries; i++)
  3195. printk(" %3d: %8lu -> %8lu\n", early_node_map[i].nid,
  3196. early_node_map[i].start_pfn,
  3197. early_node_map[i].end_pfn);
  3198. /* Initialise every node */
  3199. setup_nr_node_ids();
  3200. for_each_online_node(nid) {
  3201. pg_data_t *pgdat = NODE_DATA(nid);
  3202. free_area_init_node(nid, pgdat, NULL,
  3203. find_min_pfn_for_node(nid), NULL);
  3204. /* Any memory on that node */
  3205. if (pgdat->node_present_pages)
  3206. node_set_state(nid, N_HIGH_MEMORY);
  3207. check_for_regular_memory(pgdat);
  3208. }
  3209. }
  3210. static int __init cmdline_parse_core(char *p, unsigned long *core)
  3211. {
  3212. unsigned long long coremem;
  3213. if (!p)
  3214. return -EINVAL;
  3215. coremem = memparse(p, &p);
  3216. *core = coremem >> PAGE_SHIFT;
  3217. /* Paranoid check that UL is enough for the coremem value */
  3218. WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
  3219. return 0;
  3220. }
  3221. /*
  3222. * kernelcore=size sets the amount of memory for use for allocations that
  3223. * cannot be reclaimed or migrated.
  3224. */
  3225. static int __init cmdline_parse_kernelcore(char *p)
  3226. {
  3227. return cmdline_parse_core(p, &required_kernelcore);
  3228. }
  3229. /*
  3230. * movablecore=size sets the amount of memory for use for allocations that
  3231. * can be reclaimed or migrated.
  3232. */
  3233. static int __init cmdline_parse_movablecore(char *p)
  3234. {
  3235. return cmdline_parse_core(p, &required_movablecore);
  3236. }
  3237. early_param("kernelcore", cmdline_parse_kernelcore);
  3238. early_param("movablecore", cmdline_parse_movablecore);
  3239. #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
  3240. /**
  3241. * set_dma_reserve - set the specified number of pages reserved in the first zone
  3242. * @new_dma_reserve: The number of pages to mark reserved
  3243. *
  3244. * The per-cpu batchsize and zone watermarks are determined by present_pages.
  3245. * In the DMA zone, a significant percentage may be consumed by kernel image
  3246. * and other unfreeable allocations which can skew the watermarks badly. This
  3247. * function may optionally be used to account for unfreeable pages in the
  3248. * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
  3249. * smaller per-cpu batchsize.
  3250. */
  3251. void __init set_dma_reserve(unsigned long new_dma_reserve)
  3252. {
  3253. dma_reserve = new_dma_reserve;
  3254. }
  3255. #ifndef CONFIG_NEED_MULTIPLE_NODES
  3256. static bootmem_data_t contig_bootmem_data;
  3257. struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
  3258. EXPORT_SYMBOL(contig_page_data);
  3259. #endif
  3260. void __init free_area_init(unsigned long *zones_size)
  3261. {
  3262. free_area_init_node(0, NODE_DATA(0), zones_size,
  3263. __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
  3264. }
  3265. static int page_alloc_cpu_notify(struct notifier_block *self,
  3266. unsigned long action, void *hcpu)
  3267. {
  3268. int cpu = (unsigned long)hcpu;
  3269. if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
  3270. local_irq_disable();
  3271. __drain_pages(cpu);
  3272. vm_events_fold_cpu(cpu);
  3273. local_irq_enable();
  3274. refresh_cpu_vm_stats(cpu);
  3275. }
  3276. return NOTIFY_OK;
  3277. }
  3278. void __init page_alloc_init(void)
  3279. {
  3280. hotcpu_notifier(page_alloc_cpu_notify, 0);
  3281. }
  3282. /*
  3283. * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
  3284. * or min_free_kbytes changes.
  3285. */
  3286. static void calculate_totalreserve_pages(void)
  3287. {
  3288. struct pglist_data *pgdat;
  3289. unsigned long reserve_pages = 0;
  3290. enum zone_type i, j;
  3291. for_each_online_pgdat(pgdat) {
  3292. for (i = 0; i < MAX_NR_ZONES; i++) {
  3293. struct zone *zone = pgdat->node_zones + i;
  3294. unsigned long max = 0;
  3295. /* Find valid and maximum lowmem_reserve in the zone */
  3296. for (j = i; j < MAX_NR_ZONES; j++) {
  3297. if (zone->lowmem_reserve[j] > max)
  3298. max = zone->lowmem_reserve[j];
  3299. }
  3300. /* we treat pages_high as reserved pages. */
  3301. max += zone->pages_high;
  3302. if (max > zone->present_pages)
  3303. max = zone->present_pages;
  3304. reserve_pages += max;
  3305. }
  3306. }
  3307. totalreserve_pages = reserve_pages;
  3308. }
  3309. /*
  3310. * setup_per_zone_lowmem_reserve - called whenever
  3311. * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
  3312. * has a correct pages reserved value, so an adequate number of
  3313. * pages are left in the zone after a successful __alloc_pages().
  3314. */
  3315. static void setup_per_zone_lowmem_reserve(void)
  3316. {
  3317. struct pglist_data *pgdat;
  3318. enum zone_type j, idx;
  3319. for_each_online_pgdat(pgdat) {
  3320. for (j = 0; j < MAX_NR_ZONES; j++) {
  3321. struct zone *zone = pgdat->node_zones + j;
  3322. unsigned long present_pages = zone->present_pages;
  3323. zone->lowmem_reserve[j] = 0;
  3324. idx = j;
  3325. while (idx) {
  3326. struct zone *lower_zone;
  3327. idx--;
  3328. if (sysctl_lowmem_reserve_ratio[idx] < 1)
  3329. sysctl_lowmem_reserve_ratio[idx] = 1;
  3330. lower_zone = pgdat->node_zones + idx;
  3331. lower_zone->lowmem_reserve[j] = present_pages /
  3332. sysctl_lowmem_reserve_ratio[idx];
  3333. present_pages += lower_zone->present_pages;
  3334. }
  3335. }
  3336. }
  3337. /* update totalreserve_pages */
  3338. calculate_totalreserve_pages();
  3339. }
  3340. /**
  3341. * setup_per_zone_pages_min - called when min_free_kbytes changes.
  3342. *
  3343. * Ensures that the pages_{min,low,high} values for each zone are set correctly
  3344. * with respect to min_free_kbytes.
  3345. */
  3346. void setup_per_zone_pages_min(void)
  3347. {
  3348. unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
  3349. unsigned long lowmem_pages = 0;
  3350. struct zone *zone;
  3351. unsigned long flags;
  3352. /* Calculate total number of !ZONE_HIGHMEM pages */
  3353. for_each_zone(zone) {
  3354. if (!is_highmem(zone))
  3355. lowmem_pages += zone->present_pages;
  3356. }
  3357. for_each_zone(zone) {
  3358. u64 tmp;
  3359. spin_lock_irqsave(&zone->lru_lock, flags);
  3360. tmp = (u64)pages_min * zone->present_pages;
  3361. do_div(tmp, lowmem_pages);
  3362. if (is_highmem(zone)) {
  3363. /*
  3364. * __GFP_HIGH and PF_MEMALLOC allocations usually don't
  3365. * need highmem pages, so cap pages_min to a small
  3366. * value here.
  3367. *
  3368. * The (pages_high-pages_low) and (pages_low-pages_min)
  3369. * deltas controls asynch page reclaim, and so should
  3370. * not be capped for highmem.
  3371. */
  3372. int min_pages;
  3373. min_pages = zone->present_pages / 1024;
  3374. if (min_pages < SWAP_CLUSTER_MAX)
  3375. min_pages = SWAP_CLUSTER_MAX;
  3376. if (min_pages > 128)
  3377. min_pages = 128;
  3378. zone->pages_min = min_pages;
  3379. } else {
  3380. /*
  3381. * If it's a lowmem zone, reserve a number of pages
  3382. * proportionate to the zone's size.
  3383. */
  3384. zone->pages_min = tmp;
  3385. }
  3386. zone->pages_low = zone->pages_min + (tmp >> 2);
  3387. zone->pages_high = zone->pages_min + (tmp >> 1);
  3388. spin_unlock_irqrestore(&zone->lru_lock, flags);
  3389. }
  3390. /* update totalreserve_pages */
  3391. calculate_totalreserve_pages();
  3392. }
  3393. /*
  3394. * Initialise min_free_kbytes.
  3395. *
  3396. * For small machines we want it small (128k min). For large machines
  3397. * we want it large (64MB max). But it is not linear, because network
  3398. * bandwidth does not increase linearly with machine size. We use
  3399. *
  3400. * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
  3401. * min_free_kbytes = sqrt(lowmem_kbytes * 16)
  3402. *
  3403. * which yields
  3404. *
  3405. * 16MB: 512k
  3406. * 32MB: 724k
  3407. * 64MB: 1024k
  3408. * 128MB: 1448k
  3409. * 256MB: 2048k
  3410. * 512MB: 2896k
  3411. * 1024MB: 4096k
  3412. * 2048MB: 5792k
  3413. * 4096MB: 8192k
  3414. * 8192MB: 11584k
  3415. * 16384MB: 16384k
  3416. */
  3417. static int __init init_per_zone_pages_min(void)
  3418. {
  3419. unsigned long lowmem_kbytes;
  3420. lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
  3421. min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
  3422. if (min_free_kbytes < 128)
  3423. min_free_kbytes = 128;
  3424. if (min_free_kbytes > 65536)
  3425. min_free_kbytes = 65536;
  3426. setup_per_zone_pages_min();
  3427. setup_per_zone_lowmem_reserve();
  3428. return 0;
  3429. }
  3430. module_init(init_per_zone_pages_min)
  3431. /*
  3432. * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
  3433. * that we can call two helper functions whenever min_free_kbytes
  3434. * changes.
  3435. */
  3436. int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
  3437. struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
  3438. {
  3439. proc_dointvec(table, write, file, buffer, length, ppos);
  3440. if (write)
  3441. setup_per_zone_pages_min();
  3442. return 0;
  3443. }
  3444. #ifdef CONFIG_NUMA
  3445. int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
  3446. struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
  3447. {
  3448. struct zone *zone;
  3449. int rc;
  3450. rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
  3451. if (rc)
  3452. return rc;
  3453. for_each_zone(zone)
  3454. zone->min_unmapped_pages = (zone->present_pages *
  3455. sysctl_min_unmapped_ratio) / 100;
  3456. return 0;
  3457. }
  3458. int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
  3459. struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
  3460. {
  3461. struct zone *zone;
  3462. int rc;
  3463. rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
  3464. if (rc)
  3465. return rc;
  3466. for_each_zone(zone)
  3467. zone->min_slab_pages = (zone->present_pages *
  3468. sysctl_min_slab_ratio) / 100;
  3469. return 0;
  3470. }
  3471. #endif
  3472. /*
  3473. * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
  3474. * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
  3475. * whenever sysctl_lowmem_reserve_ratio changes.
  3476. *
  3477. * The reserve ratio obviously has absolutely no relation with the
  3478. * pages_min watermarks. The lowmem reserve ratio can only make sense
  3479. * if in function of the boot time zone sizes.
  3480. */
  3481. int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
  3482. struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
  3483. {
  3484. proc_dointvec_minmax(table, write, file, buffer, length, ppos);
  3485. setup_per_zone_lowmem_reserve();
  3486. return 0;
  3487. }
  3488. /*
  3489. * percpu_pagelist_fraction - changes the pcp->high for each zone on each
  3490. * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
  3491. * can have before it gets flushed back to buddy allocator.
  3492. */
  3493. int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
  3494. struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
  3495. {
  3496. struct zone *zone;
  3497. unsigned int cpu;
  3498. int ret;
  3499. ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
  3500. if (!write || (ret == -EINVAL))
  3501. return ret;
  3502. for_each_zone(zone) {
  3503. for_each_online_cpu(cpu) {
  3504. unsigned long high;
  3505. high = zone->present_pages / percpu_pagelist_fraction;
  3506. setup_pagelist_highmark(zone_pcp(zone, cpu), high);
  3507. }
  3508. }
  3509. return 0;
  3510. }
  3511. int hashdist = HASHDIST_DEFAULT;
  3512. #ifdef CONFIG_NUMA
  3513. static int __init set_hashdist(char *str)
  3514. {
  3515. if (!str)
  3516. return 0;
  3517. hashdist = simple_strtoul(str, &str, 0);
  3518. return 1;
  3519. }
  3520. __setup("hashdist=", set_hashdist);
  3521. #endif
  3522. /*
  3523. * allocate a large system hash table from bootmem
  3524. * - it is assumed that the hash table must contain an exact power-of-2
  3525. * quantity of entries
  3526. * - limit is the number of hash buckets, not the total allocation size
  3527. */
  3528. void *__init alloc_large_system_hash(const char *tablename,
  3529. unsigned long bucketsize,
  3530. unsigned long numentries,
  3531. int scale,
  3532. int flags,
  3533. unsigned int *_hash_shift,
  3534. unsigned int *_hash_mask,
  3535. unsigned long limit)
  3536. {
  3537. unsigned long long max = limit;
  3538. unsigned long log2qty, size;
  3539. void *table = NULL;
  3540. /* allow the kernel cmdline to have a say */
  3541. if (!numentries) {
  3542. /* round applicable memory size up to nearest megabyte */
  3543. numentries = nr_kernel_pages;
  3544. numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
  3545. numentries >>= 20 - PAGE_SHIFT;
  3546. numentries <<= 20 - PAGE_SHIFT;
  3547. /* limit to 1 bucket per 2^scale bytes of low memory */
  3548. if (scale > PAGE_SHIFT)
  3549. numentries >>= (scale - PAGE_SHIFT);
  3550. else
  3551. numentries <<= (PAGE_SHIFT - scale);
  3552. /* Make sure we've got at least a 0-order allocation.. */
  3553. if (unlikely((numentries * bucketsize) < PAGE_SIZE))
  3554. numentries = PAGE_SIZE / bucketsize;
  3555. }
  3556. numentries = roundup_pow_of_two(numentries);
  3557. /* limit allocation size to 1/16 total memory by default */
  3558. if (max == 0) {
  3559. max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
  3560. do_div(max, bucketsize);
  3561. }
  3562. if (numentries > max)
  3563. numentries = max;
  3564. log2qty = ilog2(numentries);
  3565. do {
  3566. size = bucketsize << log2qty;
  3567. if (flags & HASH_EARLY)
  3568. table = alloc_bootmem(size);
  3569. else if (hashdist)
  3570. table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
  3571. else {
  3572. unsigned long order;
  3573. for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
  3574. ;
  3575. table = (void*) __get_free_pages(GFP_ATOMIC, order);
  3576. /*
  3577. * If bucketsize is not a power-of-two, we may free
  3578. * some pages at the end of hash table.
  3579. */
  3580. if (table) {
  3581. unsigned long alloc_end = (unsigned long)table +
  3582. (PAGE_SIZE << order);
  3583. unsigned long used = (unsigned long)table +
  3584. PAGE_ALIGN(size);
  3585. split_page(virt_to_page(table), order);
  3586. while (used < alloc_end) {
  3587. free_page(used);
  3588. used += PAGE_SIZE;
  3589. }
  3590. }
  3591. }
  3592. } while (!table && size > PAGE_SIZE && --log2qty);
  3593. if (!table)
  3594. panic("Failed to allocate %s hash table\n", tablename);
  3595. printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n",
  3596. tablename,
  3597. (1U << log2qty),
  3598. ilog2(size) - PAGE_SHIFT,
  3599. size);
  3600. if (_hash_shift)
  3601. *_hash_shift = log2qty;
  3602. if (_hash_mask)
  3603. *_hash_mask = (1 << log2qty) - 1;
  3604. return table;
  3605. }
  3606. #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
  3607. struct page *pfn_to_page(unsigned long pfn)
  3608. {
  3609. return __pfn_to_page(pfn);
  3610. }
  3611. unsigned long page_to_pfn(struct page *page)
  3612. {
  3613. return __page_to_pfn(page);
  3614. }
  3615. EXPORT_SYMBOL(pfn_to_page);
  3616. EXPORT_SYMBOL(page_to_pfn);
  3617. #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
  3618. /* Return a pointer to the bitmap storing bits affecting a block of pages */
  3619. static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
  3620. unsigned long pfn)
  3621. {
  3622. #ifdef CONFIG_SPARSEMEM
  3623. return __pfn_to_section(pfn)->pageblock_flags;
  3624. #else
  3625. return zone->pageblock_flags;
  3626. #endif /* CONFIG_SPARSEMEM */
  3627. }
  3628. static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
  3629. {
  3630. #ifdef CONFIG_SPARSEMEM
  3631. pfn &= (PAGES_PER_SECTION-1);
  3632. return (pfn >> (MAX_ORDER-1)) * NR_PAGEBLOCK_BITS;
  3633. #else
  3634. pfn = pfn - zone->zone_start_pfn;
  3635. return (pfn >> (MAX_ORDER-1)) * NR_PAGEBLOCK_BITS;
  3636. #endif /* CONFIG_SPARSEMEM */
  3637. }
  3638. /**
  3639. * get_pageblock_flags_group - Return the requested group of flags for the MAX_ORDER_NR_PAGES block of pages
  3640. * @page: The page within the block of interest
  3641. * @start_bitidx: The first bit of interest to retrieve
  3642. * @end_bitidx: The last bit of interest
  3643. * returns pageblock_bits flags
  3644. */
  3645. unsigned long get_pageblock_flags_group(struct page *page,
  3646. int start_bitidx, int end_bitidx)
  3647. {
  3648. struct zone *zone;
  3649. unsigned long *bitmap;
  3650. unsigned long pfn, bitidx;
  3651. unsigned long flags = 0;
  3652. unsigned long value = 1;
  3653. zone = page_zone(page);
  3654. pfn = page_to_pfn(page);
  3655. bitmap = get_pageblock_bitmap(zone, pfn);
  3656. bitidx = pfn_to_bitidx(zone, pfn);
  3657. for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
  3658. if (test_bit(bitidx + start_bitidx, bitmap))
  3659. flags |= value;
  3660. return flags;
  3661. }
  3662. /**
  3663. * set_pageblock_flags_group - Set the requested group of flags for a MAX_ORDER_NR_PAGES block of pages
  3664. * @page: The page within the block of interest
  3665. * @start_bitidx: The first bit of interest
  3666. * @end_bitidx: The last bit of interest
  3667. * @flags: The flags to set
  3668. */
  3669. void set_pageblock_flags_group(struct page *page, unsigned long flags,
  3670. int start_bitidx, int end_bitidx)
  3671. {
  3672. struct zone *zone;
  3673. unsigned long *bitmap;
  3674. unsigned long pfn, bitidx;
  3675. unsigned long value = 1;
  3676. zone = page_zone(page);
  3677. pfn = page_to_pfn(page);
  3678. bitmap = get_pageblock_bitmap(zone, pfn);
  3679. bitidx = pfn_to_bitidx(zone, pfn);
  3680. for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
  3681. if (flags & value)
  3682. __set_bit(bitidx + start_bitidx, bitmap);
  3683. else
  3684. __clear_bit(bitidx + start_bitidx, bitmap);
  3685. }