tree_plugin.h 79 KB

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385138613871388138913901391139213931394139513961397139813991400140114021403140414051406140714081409141014111412141314141415141614171418141914201421142214231424142514261427142814291430143114321433143414351436143714381439144014411442144314441445144614471448144914501451145214531454145514561457145814591460146114621463146414651466146714681469147014711472147314741475147614771478147914801481148214831484148514861487148814891490149114921493149414951496149714981499150015011502150315041505150615071508150915101511151215131514151515161517151815191520152115221523152415251526152715281529153015311532153315341535153615371538153915401541154215431544154515461547154815491550155115521553155415551556155715581559156015611562156315641565156615671568156915701571157215731574157515761577157815791580158115821583158415851586158715881589159015911592159315941595159615971598159916001601160216031604160516061607160816091610161116121613161416151616161716181619162016211622162316241625162616271628162916301631163216331634163516361637163816391640164116421643164416451646164716481649165016511652165316541655165616571658165916601661166216631664166516661667166816691670167116721673167416751676167716781679168016811682168316841685168616871688168916901691169216931694169516961697169816991700170117021703170417051706170717081709171017111712171317141715171617171718171917201721172217231724172517261727172817291730173117321733173417351736173717381739174017411742174317441745174617471748174917501751175217531754175517561757175817591760176117621763176417651766176717681769177017711772177317741775177617771778177917801781178217831784178517861787178817891790179117921793179417951796179717981799180018011802180318041805180618071808180918101811181218131814181518161817181818191820182118221823182418251826182718281829183018311832183318341835183618371838183918401841184218431844184518461847184818491850185118521853185418551856185718581859186018611862186318641865186618671868186918701871187218731874187518761877187818791880188118821883188418851886188718881889189018911892189318941895189618971898189919001901190219031904190519061907190819091910191119121913191419151916191719181919192019211922192319241925192619271928192919301931193219331934193519361937193819391940194119421943194419451946194719481949195019511952195319541955195619571958195919601961196219631964196519661967196819691970197119721973197419751976197719781979198019811982198319841985198619871988198919901991199219931994199519961997199819992000200120022003200420052006200720082009201020112012201320142015201620172018201920202021202220232024202520262027202820292030203120322033203420352036203720382039204020412042204320442045204620472048204920502051205220532054205520562057205820592060206120622063206420652066206720682069207020712072207320742075207620772078207920802081208220832084208520862087208820892090209120922093209420952096209720982099210021012102210321042105210621072108210921102111211221132114211521162117211821192120212121222123212421252126212721282129213021312132213321342135213621372138213921402141214221432144214521462147214821492150215121522153215421552156215721582159216021612162216321642165216621672168216921702171217221732174217521762177217821792180218121822183218421852186218721882189219021912192219321942195219621972198219922002201220222032204220522062207220822092210221122122213221422152216221722182219222022212222222322242225222622272228222922302231223222332234223522362237223822392240224122422243224422452246224722482249225022512252225322542255225622572258225922602261226222632264226522662267226822692270227122722273227422752276227722782279228022812282228322842285228622872288228922902291229222932294229522962297229822992300230123022303230423052306230723082309231023112312231323142315231623172318231923202321232223232324232523262327232823292330233123322333233423352336233723382339234023412342234323442345234623472348234923502351235223532354235523562357235823592360236123622363236423652366236723682369237023712372237323742375237623772378237923802381238223832384238523862387238823892390239123922393239423952396239723982399240024012402240324042405240624072408240924102411241224132414241524162417241824192420242124222423242424252426242724282429243024312432243324342435243624372438243924402441244224432444244524462447244824492450245124522453245424552456245724582459246024612462246324642465246624672468246924702471247224732474247524762477247824792480248124822483248424852486248724882489249024912492249324942495249624972498249925002501250225032504250525062507250825092510251125122513251425152516251725182519252025212522252325242525252625272528252925302531253225332534253525362537253825392540254125422543254425452546254725482549255025512552255325542555255625572558255925602561256225632564256525662567256825692570257125722573257425752576257725782579258025812582258325842585258625872588258925902591259225932594259525962597259825992600260126022603
  1. /*
  2. * Read-Copy Update mechanism for mutual exclusion (tree-based version)
  3. * Internal non-public definitions that provide either classic
  4. * or preemptible semantics.
  5. *
  6. * This program is free software; you can redistribute it and/or modify
  7. * it under the terms of the GNU General Public License as published by
  8. * the Free Software Foundation; either version 2 of the License, or
  9. * (at your option) any later version.
  10. *
  11. * This program is distributed in the hope that it will be useful,
  12. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  14. * GNU General Public License for more details.
  15. *
  16. * You should have received a copy of the GNU General Public License
  17. * along with this program; if not, you can access it online at
  18. * http://www.gnu.org/licenses/gpl-2.0.html.
  19. *
  20. * Copyright Red Hat, 2009
  21. * Copyright IBM Corporation, 2009
  22. *
  23. * Author: Ingo Molnar <mingo@elte.hu>
  24. * Paul E. McKenney <paulmck@linux.vnet.ibm.com>
  25. */
  26. #include <linux/delay.h>
  27. #include <linux/gfp.h>
  28. #include <linux/oom.h>
  29. #include <linux/sched/debug.h>
  30. #include <linux/smpboot.h>
  31. #include <uapi/linux/sched/types.h>
  32. #include "../time/tick-internal.h"
  33. #ifdef CONFIG_RCU_BOOST
  34. #include "../locking/rtmutex_common.h"
  35. /*
  36. * Control variables for per-CPU and per-rcu_node kthreads. These
  37. * handle all flavors of RCU.
  38. */
  39. static DEFINE_PER_CPU(struct task_struct *, rcu_cpu_kthread_task);
  40. DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_status);
  41. DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_loops);
  42. DEFINE_PER_CPU(char, rcu_cpu_has_work);
  43. #else /* #ifdef CONFIG_RCU_BOOST */
  44. /*
  45. * Some architectures do not define rt_mutexes, but if !CONFIG_RCU_BOOST,
  46. * all uses are in dead code. Provide a definition to keep the compiler
  47. * happy, but add WARN_ON_ONCE() to complain if used in the wrong place.
  48. * This probably needs to be excluded from -rt builds.
  49. */
  50. #define rt_mutex_owner(a) ({ WARN_ON_ONCE(1); NULL; })
  51. #endif /* #else #ifdef CONFIG_RCU_BOOST */
  52. #ifdef CONFIG_RCU_NOCB_CPU
  53. static cpumask_var_t rcu_nocb_mask; /* CPUs to have callbacks offloaded. */
  54. static bool have_rcu_nocb_mask; /* Was rcu_nocb_mask allocated? */
  55. static bool __read_mostly rcu_nocb_poll; /* Offload kthread are to poll. */
  56. #endif /* #ifdef CONFIG_RCU_NOCB_CPU */
  57. /*
  58. * Check the RCU kernel configuration parameters and print informative
  59. * messages about anything out of the ordinary.
  60. */
  61. static void __init rcu_bootup_announce_oddness(void)
  62. {
  63. if (IS_ENABLED(CONFIG_RCU_TRACE))
  64. pr_info("\tRCU event tracing is enabled.\n");
  65. if ((IS_ENABLED(CONFIG_64BIT) && RCU_FANOUT != 64) ||
  66. (!IS_ENABLED(CONFIG_64BIT) && RCU_FANOUT != 32))
  67. pr_info("\tCONFIG_RCU_FANOUT set to non-default value of %d\n",
  68. RCU_FANOUT);
  69. if (rcu_fanout_exact)
  70. pr_info("\tHierarchical RCU autobalancing is disabled.\n");
  71. if (IS_ENABLED(CONFIG_RCU_FAST_NO_HZ))
  72. pr_info("\tRCU dyntick-idle grace-period acceleration is enabled.\n");
  73. if (IS_ENABLED(CONFIG_PROVE_RCU))
  74. pr_info("\tRCU lockdep checking is enabled.\n");
  75. if (RCU_NUM_LVLS >= 4)
  76. pr_info("\tFour(or more)-level hierarchy is enabled.\n");
  77. if (RCU_FANOUT_LEAF != 16)
  78. pr_info("\tBuild-time adjustment of leaf fanout to %d.\n",
  79. RCU_FANOUT_LEAF);
  80. if (rcu_fanout_leaf != RCU_FANOUT_LEAF)
  81. pr_info("\tBoot-time adjustment of leaf fanout to %d.\n", rcu_fanout_leaf);
  82. if (nr_cpu_ids != NR_CPUS)
  83. pr_info("\tRCU restricting CPUs from NR_CPUS=%d to nr_cpu_ids=%u.\n", NR_CPUS, nr_cpu_ids);
  84. #ifdef CONFIG_RCU_BOOST
  85. pr_info("\tRCU priority boosting: priority %d delay %d ms.\n", kthread_prio, CONFIG_RCU_BOOST_DELAY);
  86. #endif
  87. if (blimit != DEFAULT_RCU_BLIMIT)
  88. pr_info("\tBoot-time adjustment of callback invocation limit to %ld.\n", blimit);
  89. if (qhimark != DEFAULT_RCU_QHIMARK)
  90. pr_info("\tBoot-time adjustment of callback high-water mark to %ld.\n", qhimark);
  91. if (qlowmark != DEFAULT_RCU_QLOMARK)
  92. pr_info("\tBoot-time adjustment of callback low-water mark to %ld.\n", qlowmark);
  93. if (jiffies_till_first_fqs != ULONG_MAX)
  94. pr_info("\tBoot-time adjustment of first FQS scan delay to %ld jiffies.\n", jiffies_till_first_fqs);
  95. if (jiffies_till_next_fqs != ULONG_MAX)
  96. pr_info("\tBoot-time adjustment of subsequent FQS scan delay to %ld jiffies.\n", jiffies_till_next_fqs);
  97. if (rcu_kick_kthreads)
  98. pr_info("\tKick kthreads if too-long grace period.\n");
  99. if (IS_ENABLED(CONFIG_DEBUG_OBJECTS_RCU_HEAD))
  100. pr_info("\tRCU callback double-/use-after-free debug enabled.\n");
  101. if (gp_preinit_delay)
  102. pr_info("\tRCU debug GP pre-init slowdown %d jiffies.\n", gp_preinit_delay);
  103. if (gp_init_delay)
  104. pr_info("\tRCU debug GP init slowdown %d jiffies.\n", gp_init_delay);
  105. if (gp_cleanup_delay)
  106. pr_info("\tRCU debug GP init slowdown %d jiffies.\n", gp_cleanup_delay);
  107. if (IS_ENABLED(CONFIG_RCU_EQS_DEBUG))
  108. pr_info("\tRCU debug extended QS entry/exit.\n");
  109. rcupdate_announce_bootup_oddness();
  110. }
  111. #ifdef CONFIG_PREEMPT_RCU
  112. RCU_STATE_INITIALIZER(rcu_preempt, 'p', call_rcu);
  113. static struct rcu_state *const rcu_state_p = &rcu_preempt_state;
  114. static struct rcu_data __percpu *const rcu_data_p = &rcu_preempt_data;
  115. static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp,
  116. bool wake);
  117. /*
  118. * Tell them what RCU they are running.
  119. */
  120. static void __init rcu_bootup_announce(void)
  121. {
  122. pr_info("Preemptible hierarchical RCU implementation.\n");
  123. rcu_bootup_announce_oddness();
  124. }
  125. /* Flags for rcu_preempt_ctxt_queue() decision table. */
  126. #define RCU_GP_TASKS 0x8
  127. #define RCU_EXP_TASKS 0x4
  128. #define RCU_GP_BLKD 0x2
  129. #define RCU_EXP_BLKD 0x1
  130. /*
  131. * Queues a task preempted within an RCU-preempt read-side critical
  132. * section into the appropriate location within the ->blkd_tasks list,
  133. * depending on the states of any ongoing normal and expedited grace
  134. * periods. The ->gp_tasks pointer indicates which element the normal
  135. * grace period is waiting on (NULL if none), and the ->exp_tasks pointer
  136. * indicates which element the expedited grace period is waiting on (again,
  137. * NULL if none). If a grace period is waiting on a given element in the
  138. * ->blkd_tasks list, it also waits on all subsequent elements. Thus,
  139. * adding a task to the tail of the list blocks any grace period that is
  140. * already waiting on one of the elements. In contrast, adding a task
  141. * to the head of the list won't block any grace period that is already
  142. * waiting on one of the elements.
  143. *
  144. * This queuing is imprecise, and can sometimes make an ongoing grace
  145. * period wait for a task that is not strictly speaking blocking it.
  146. * Given the choice, we needlessly block a normal grace period rather than
  147. * blocking an expedited grace period.
  148. *
  149. * Note that an endless sequence of expedited grace periods still cannot
  150. * indefinitely postpone a normal grace period. Eventually, all of the
  151. * fixed number of preempted tasks blocking the normal grace period that are
  152. * not also blocking the expedited grace period will resume and complete
  153. * their RCU read-side critical sections. At that point, the ->gp_tasks
  154. * pointer will equal the ->exp_tasks pointer, at which point the end of
  155. * the corresponding expedited grace period will also be the end of the
  156. * normal grace period.
  157. */
  158. static void rcu_preempt_ctxt_queue(struct rcu_node *rnp, struct rcu_data *rdp)
  159. __releases(rnp->lock) /* But leaves rrupts disabled. */
  160. {
  161. int blkd_state = (rnp->gp_tasks ? RCU_GP_TASKS : 0) +
  162. (rnp->exp_tasks ? RCU_EXP_TASKS : 0) +
  163. (rnp->qsmask & rdp->grpmask ? RCU_GP_BLKD : 0) +
  164. (rnp->expmask & rdp->grpmask ? RCU_EXP_BLKD : 0);
  165. struct task_struct *t = current;
  166. lockdep_assert_held(&rnp->lock);
  167. WARN_ON_ONCE(rdp->mynode != rnp);
  168. WARN_ON_ONCE(rnp->level != rcu_num_lvls - 1);
  169. /*
  170. * Decide where to queue the newly blocked task. In theory,
  171. * this could be an if-statement. In practice, when I tried
  172. * that, it was quite messy.
  173. */
  174. switch (blkd_state) {
  175. case 0:
  176. case RCU_EXP_TASKS:
  177. case RCU_EXP_TASKS + RCU_GP_BLKD:
  178. case RCU_GP_TASKS:
  179. case RCU_GP_TASKS + RCU_EXP_TASKS:
  180. /*
  181. * Blocking neither GP, or first task blocking the normal
  182. * GP but not blocking the already-waiting expedited GP.
  183. * Queue at the head of the list to avoid unnecessarily
  184. * blocking the already-waiting GPs.
  185. */
  186. list_add(&t->rcu_node_entry, &rnp->blkd_tasks);
  187. break;
  188. case RCU_EXP_BLKD:
  189. case RCU_GP_BLKD:
  190. case RCU_GP_BLKD + RCU_EXP_BLKD:
  191. case RCU_GP_TASKS + RCU_EXP_BLKD:
  192. case RCU_GP_TASKS + RCU_GP_BLKD + RCU_EXP_BLKD:
  193. case RCU_GP_TASKS + RCU_EXP_TASKS + RCU_GP_BLKD + RCU_EXP_BLKD:
  194. /*
  195. * First task arriving that blocks either GP, or first task
  196. * arriving that blocks the expedited GP (with the normal
  197. * GP already waiting), or a task arriving that blocks
  198. * both GPs with both GPs already waiting. Queue at the
  199. * tail of the list to avoid any GP waiting on any of the
  200. * already queued tasks that are not blocking it.
  201. */
  202. list_add_tail(&t->rcu_node_entry, &rnp->blkd_tasks);
  203. break;
  204. case RCU_EXP_TASKS + RCU_EXP_BLKD:
  205. case RCU_EXP_TASKS + RCU_GP_BLKD + RCU_EXP_BLKD:
  206. case RCU_GP_TASKS + RCU_EXP_TASKS + RCU_EXP_BLKD:
  207. /*
  208. * Second or subsequent task blocking the expedited GP.
  209. * The task either does not block the normal GP, or is the
  210. * first task blocking the normal GP. Queue just after
  211. * the first task blocking the expedited GP.
  212. */
  213. list_add(&t->rcu_node_entry, rnp->exp_tasks);
  214. break;
  215. case RCU_GP_TASKS + RCU_GP_BLKD:
  216. case RCU_GP_TASKS + RCU_EXP_TASKS + RCU_GP_BLKD:
  217. /*
  218. * Second or subsequent task blocking the normal GP.
  219. * The task does not block the expedited GP. Queue just
  220. * after the first task blocking the normal GP.
  221. */
  222. list_add(&t->rcu_node_entry, rnp->gp_tasks);
  223. break;
  224. default:
  225. /* Yet another exercise in excessive paranoia. */
  226. WARN_ON_ONCE(1);
  227. break;
  228. }
  229. /*
  230. * We have now queued the task. If it was the first one to
  231. * block either grace period, update the ->gp_tasks and/or
  232. * ->exp_tasks pointers, respectively, to reference the newly
  233. * blocked tasks.
  234. */
  235. if (!rnp->gp_tasks && (blkd_state & RCU_GP_BLKD))
  236. rnp->gp_tasks = &t->rcu_node_entry;
  237. if (!rnp->exp_tasks && (blkd_state & RCU_EXP_BLKD))
  238. rnp->exp_tasks = &t->rcu_node_entry;
  239. WARN_ON_ONCE(!(blkd_state & RCU_GP_BLKD) !=
  240. !(rnp->qsmask & rdp->grpmask));
  241. WARN_ON_ONCE(!(blkd_state & RCU_EXP_BLKD) !=
  242. !(rnp->expmask & rdp->grpmask));
  243. raw_spin_unlock_rcu_node(rnp); /* interrupts remain disabled. */
  244. /*
  245. * Report the quiescent state for the expedited GP. This expedited
  246. * GP should not be able to end until we report, so there should be
  247. * no need to check for a subsequent expedited GP. (Though we are
  248. * still in a quiescent state in any case.)
  249. */
  250. if (blkd_state & RCU_EXP_BLKD &&
  251. t->rcu_read_unlock_special.b.exp_need_qs) {
  252. t->rcu_read_unlock_special.b.exp_need_qs = false;
  253. rcu_report_exp_rdp(rdp->rsp, rdp, true);
  254. } else {
  255. WARN_ON_ONCE(t->rcu_read_unlock_special.b.exp_need_qs);
  256. }
  257. }
  258. /*
  259. * Record a preemptible-RCU quiescent state for the specified CPU. Note
  260. * that this just means that the task currently running on the CPU is
  261. * not in a quiescent state. There might be any number of tasks blocked
  262. * while in an RCU read-side critical section.
  263. *
  264. * As with the other rcu_*_qs() functions, callers to this function
  265. * must disable preemption.
  266. */
  267. static void rcu_preempt_qs(void)
  268. {
  269. RCU_LOCKDEP_WARN(preemptible(), "rcu_preempt_qs() invoked with preemption enabled!!!\n");
  270. if (__this_cpu_read(rcu_data_p->cpu_no_qs.s)) {
  271. trace_rcu_grace_period(TPS("rcu_preempt"),
  272. __this_cpu_read(rcu_data_p->gpnum),
  273. TPS("cpuqs"));
  274. __this_cpu_write(rcu_data_p->cpu_no_qs.b.norm, false);
  275. barrier(); /* Coordinate with rcu_preempt_check_callbacks(). */
  276. current->rcu_read_unlock_special.b.need_qs = false;
  277. }
  278. }
  279. /*
  280. * We have entered the scheduler, and the current task might soon be
  281. * context-switched away from. If this task is in an RCU read-side
  282. * critical section, we will no longer be able to rely on the CPU to
  283. * record that fact, so we enqueue the task on the blkd_tasks list.
  284. * The task will dequeue itself when it exits the outermost enclosing
  285. * RCU read-side critical section. Therefore, the current grace period
  286. * cannot be permitted to complete until the blkd_tasks list entries
  287. * predating the current grace period drain, in other words, until
  288. * rnp->gp_tasks becomes NULL.
  289. *
  290. * Caller must disable interrupts.
  291. */
  292. static void rcu_preempt_note_context_switch(bool preempt)
  293. {
  294. struct task_struct *t = current;
  295. struct rcu_data *rdp;
  296. struct rcu_node *rnp;
  297. RCU_LOCKDEP_WARN(!irqs_disabled(), "rcu_preempt_note_context_switch() invoked with interrupts enabled!!!\n");
  298. WARN_ON_ONCE(!preempt && t->rcu_read_lock_nesting > 0);
  299. if (t->rcu_read_lock_nesting > 0 &&
  300. !t->rcu_read_unlock_special.b.blocked) {
  301. /* Possibly blocking in an RCU read-side critical section. */
  302. rdp = this_cpu_ptr(rcu_state_p->rda);
  303. rnp = rdp->mynode;
  304. raw_spin_lock_rcu_node(rnp);
  305. t->rcu_read_unlock_special.b.blocked = true;
  306. t->rcu_blocked_node = rnp;
  307. /*
  308. * Verify the CPU's sanity, trace the preemption, and
  309. * then queue the task as required based on the states
  310. * of any ongoing and expedited grace periods.
  311. */
  312. WARN_ON_ONCE((rdp->grpmask & rcu_rnp_online_cpus(rnp)) == 0);
  313. WARN_ON_ONCE(!list_empty(&t->rcu_node_entry));
  314. trace_rcu_preempt_task(rdp->rsp->name,
  315. t->pid,
  316. (rnp->qsmask & rdp->grpmask)
  317. ? rnp->gpnum
  318. : rnp->gpnum + 1);
  319. rcu_preempt_ctxt_queue(rnp, rdp);
  320. } else if (t->rcu_read_lock_nesting < 0 &&
  321. t->rcu_read_unlock_special.s) {
  322. /*
  323. * Complete exit from RCU read-side critical section on
  324. * behalf of preempted instance of __rcu_read_unlock().
  325. */
  326. rcu_read_unlock_special(t);
  327. }
  328. /*
  329. * Either we were not in an RCU read-side critical section to
  330. * begin with, or we have now recorded that critical section
  331. * globally. Either way, we can now note a quiescent state
  332. * for this CPU. Again, if we were in an RCU read-side critical
  333. * section, and if that critical section was blocking the current
  334. * grace period, then the fact that the task has been enqueued
  335. * means that we continue to block the current grace period.
  336. */
  337. rcu_preempt_qs();
  338. }
  339. /*
  340. * Check for preempted RCU readers blocking the current grace period
  341. * for the specified rcu_node structure. If the caller needs a reliable
  342. * answer, it must hold the rcu_node's ->lock.
  343. */
  344. static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp)
  345. {
  346. return rnp->gp_tasks != NULL;
  347. }
  348. /*
  349. * Advance a ->blkd_tasks-list pointer to the next entry, instead
  350. * returning NULL if at the end of the list.
  351. */
  352. static struct list_head *rcu_next_node_entry(struct task_struct *t,
  353. struct rcu_node *rnp)
  354. {
  355. struct list_head *np;
  356. np = t->rcu_node_entry.next;
  357. if (np == &rnp->blkd_tasks)
  358. np = NULL;
  359. return np;
  360. }
  361. /*
  362. * Return true if the specified rcu_node structure has tasks that were
  363. * preempted within an RCU read-side critical section.
  364. */
  365. static bool rcu_preempt_has_tasks(struct rcu_node *rnp)
  366. {
  367. return !list_empty(&rnp->blkd_tasks);
  368. }
  369. /*
  370. * Handle special cases during rcu_read_unlock(), such as needing to
  371. * notify RCU core processing or task having blocked during the RCU
  372. * read-side critical section.
  373. */
  374. void rcu_read_unlock_special(struct task_struct *t)
  375. {
  376. bool empty_exp;
  377. bool empty_norm;
  378. bool empty_exp_now;
  379. unsigned long flags;
  380. struct list_head *np;
  381. bool drop_boost_mutex = false;
  382. struct rcu_data *rdp;
  383. struct rcu_node *rnp;
  384. union rcu_special special;
  385. /* NMI handlers cannot block and cannot safely manipulate state. */
  386. if (in_nmi())
  387. return;
  388. local_irq_save(flags);
  389. /*
  390. * If RCU core is waiting for this CPU to exit its critical section,
  391. * report the fact that it has exited. Because irqs are disabled,
  392. * t->rcu_read_unlock_special cannot change.
  393. */
  394. special = t->rcu_read_unlock_special;
  395. if (special.b.need_qs) {
  396. rcu_preempt_qs();
  397. t->rcu_read_unlock_special.b.need_qs = false;
  398. if (!t->rcu_read_unlock_special.s) {
  399. local_irq_restore(flags);
  400. return;
  401. }
  402. }
  403. /*
  404. * Respond to a request for an expedited grace period, but only if
  405. * we were not preempted, meaning that we were running on the same
  406. * CPU throughout. If we were preempted, the exp_need_qs flag
  407. * would have been cleared at the time of the first preemption,
  408. * and the quiescent state would be reported when we were dequeued.
  409. */
  410. if (special.b.exp_need_qs) {
  411. WARN_ON_ONCE(special.b.blocked);
  412. t->rcu_read_unlock_special.b.exp_need_qs = false;
  413. rdp = this_cpu_ptr(rcu_state_p->rda);
  414. rcu_report_exp_rdp(rcu_state_p, rdp, true);
  415. if (!t->rcu_read_unlock_special.s) {
  416. local_irq_restore(flags);
  417. return;
  418. }
  419. }
  420. /* Hardware IRQ handlers cannot block, complain if they get here. */
  421. if (in_irq() || in_serving_softirq()) {
  422. lockdep_rcu_suspicious(__FILE__, __LINE__,
  423. "rcu_read_unlock() from irq or softirq with blocking in critical section!!!\n");
  424. pr_alert("->rcu_read_unlock_special: %#x (b: %d, enq: %d nq: %d)\n",
  425. t->rcu_read_unlock_special.s,
  426. t->rcu_read_unlock_special.b.blocked,
  427. t->rcu_read_unlock_special.b.exp_need_qs,
  428. t->rcu_read_unlock_special.b.need_qs);
  429. local_irq_restore(flags);
  430. return;
  431. }
  432. /* Clean up if blocked during RCU read-side critical section. */
  433. if (special.b.blocked) {
  434. t->rcu_read_unlock_special.b.blocked = false;
  435. /*
  436. * Remove this task from the list it blocked on. The task
  437. * now remains queued on the rcu_node corresponding to the
  438. * CPU it first blocked on, so there is no longer any need
  439. * to loop. Retain a WARN_ON_ONCE() out of sheer paranoia.
  440. */
  441. rnp = t->rcu_blocked_node;
  442. raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */
  443. WARN_ON_ONCE(rnp != t->rcu_blocked_node);
  444. WARN_ON_ONCE(rnp->level != rcu_num_lvls - 1);
  445. empty_norm = !rcu_preempt_blocked_readers_cgp(rnp);
  446. empty_exp = sync_rcu_preempt_exp_done(rnp);
  447. smp_mb(); /* ensure expedited fastpath sees end of RCU c-s. */
  448. np = rcu_next_node_entry(t, rnp);
  449. list_del_init(&t->rcu_node_entry);
  450. t->rcu_blocked_node = NULL;
  451. trace_rcu_unlock_preempted_task(TPS("rcu_preempt"),
  452. rnp->gpnum, t->pid);
  453. if (&t->rcu_node_entry == rnp->gp_tasks)
  454. rnp->gp_tasks = np;
  455. if (&t->rcu_node_entry == rnp->exp_tasks)
  456. rnp->exp_tasks = np;
  457. if (IS_ENABLED(CONFIG_RCU_BOOST)) {
  458. /* Snapshot ->boost_mtx ownership w/rnp->lock held. */
  459. drop_boost_mutex = rt_mutex_owner(&rnp->boost_mtx) == t;
  460. if (&t->rcu_node_entry == rnp->boost_tasks)
  461. rnp->boost_tasks = np;
  462. }
  463. /*
  464. * If this was the last task on the current list, and if
  465. * we aren't waiting on any CPUs, report the quiescent state.
  466. * Note that rcu_report_unblock_qs_rnp() releases rnp->lock,
  467. * so we must take a snapshot of the expedited state.
  468. */
  469. empty_exp_now = sync_rcu_preempt_exp_done(rnp);
  470. if (!empty_norm && !rcu_preempt_blocked_readers_cgp(rnp)) {
  471. trace_rcu_quiescent_state_report(TPS("preempt_rcu"),
  472. rnp->gpnum,
  473. 0, rnp->qsmask,
  474. rnp->level,
  475. rnp->grplo,
  476. rnp->grphi,
  477. !!rnp->gp_tasks);
  478. rcu_report_unblock_qs_rnp(rcu_state_p, rnp, flags);
  479. } else {
  480. raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
  481. }
  482. /* Unboost if we were boosted. */
  483. if (IS_ENABLED(CONFIG_RCU_BOOST) && drop_boost_mutex)
  484. rt_mutex_unlock(&rnp->boost_mtx);
  485. /*
  486. * If this was the last task on the expedited lists,
  487. * then we need to report up the rcu_node hierarchy.
  488. */
  489. if (!empty_exp && empty_exp_now)
  490. rcu_report_exp_rnp(rcu_state_p, rnp, true);
  491. } else {
  492. local_irq_restore(flags);
  493. }
  494. }
  495. /*
  496. * Dump detailed information for all tasks blocking the current RCU
  497. * grace period on the specified rcu_node structure.
  498. */
  499. static void rcu_print_detail_task_stall_rnp(struct rcu_node *rnp)
  500. {
  501. unsigned long flags;
  502. struct task_struct *t;
  503. raw_spin_lock_irqsave_rcu_node(rnp, flags);
  504. if (!rcu_preempt_blocked_readers_cgp(rnp)) {
  505. raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
  506. return;
  507. }
  508. t = list_entry(rnp->gp_tasks->prev,
  509. struct task_struct, rcu_node_entry);
  510. list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry)
  511. sched_show_task(t);
  512. raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
  513. }
  514. /*
  515. * Dump detailed information for all tasks blocking the current RCU
  516. * grace period.
  517. */
  518. static void rcu_print_detail_task_stall(struct rcu_state *rsp)
  519. {
  520. struct rcu_node *rnp = rcu_get_root(rsp);
  521. rcu_print_detail_task_stall_rnp(rnp);
  522. rcu_for_each_leaf_node(rsp, rnp)
  523. rcu_print_detail_task_stall_rnp(rnp);
  524. }
  525. static void rcu_print_task_stall_begin(struct rcu_node *rnp)
  526. {
  527. pr_err("\tTasks blocked on level-%d rcu_node (CPUs %d-%d):",
  528. rnp->level, rnp->grplo, rnp->grphi);
  529. }
  530. static void rcu_print_task_stall_end(void)
  531. {
  532. pr_cont("\n");
  533. }
  534. /*
  535. * Scan the current list of tasks blocked within RCU read-side critical
  536. * sections, printing out the tid of each.
  537. */
  538. static int rcu_print_task_stall(struct rcu_node *rnp)
  539. {
  540. struct task_struct *t;
  541. int ndetected = 0;
  542. if (!rcu_preempt_blocked_readers_cgp(rnp))
  543. return 0;
  544. rcu_print_task_stall_begin(rnp);
  545. t = list_entry(rnp->gp_tasks->prev,
  546. struct task_struct, rcu_node_entry);
  547. list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) {
  548. pr_cont(" P%d", t->pid);
  549. ndetected++;
  550. }
  551. rcu_print_task_stall_end();
  552. return ndetected;
  553. }
  554. /*
  555. * Scan the current list of tasks blocked within RCU read-side critical
  556. * sections, printing out the tid of each that is blocking the current
  557. * expedited grace period.
  558. */
  559. static int rcu_print_task_exp_stall(struct rcu_node *rnp)
  560. {
  561. struct task_struct *t;
  562. int ndetected = 0;
  563. if (!rnp->exp_tasks)
  564. return 0;
  565. t = list_entry(rnp->exp_tasks->prev,
  566. struct task_struct, rcu_node_entry);
  567. list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) {
  568. pr_cont(" P%d", t->pid);
  569. ndetected++;
  570. }
  571. return ndetected;
  572. }
  573. /*
  574. * Check that the list of blocked tasks for the newly completed grace
  575. * period is in fact empty. It is a serious bug to complete a grace
  576. * period that still has RCU readers blocked! This function must be
  577. * invoked -before- updating this rnp's ->gpnum, and the rnp's ->lock
  578. * must be held by the caller.
  579. *
  580. * Also, if there are blocked tasks on the list, they automatically
  581. * block the newly created grace period, so set up ->gp_tasks accordingly.
  582. */
  583. static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp)
  584. {
  585. struct task_struct *t;
  586. RCU_LOCKDEP_WARN(preemptible(), "rcu_preempt_check_blocked_tasks() invoked with preemption enabled!!!\n");
  587. WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp));
  588. if (rcu_preempt_has_tasks(rnp)) {
  589. rnp->gp_tasks = rnp->blkd_tasks.next;
  590. t = container_of(rnp->gp_tasks, struct task_struct,
  591. rcu_node_entry);
  592. trace_rcu_unlock_preempted_task(TPS("rcu_preempt-GPS"),
  593. rnp->gpnum, t->pid);
  594. }
  595. WARN_ON_ONCE(rnp->qsmask);
  596. }
  597. /*
  598. * Check for a quiescent state from the current CPU. When a task blocks,
  599. * the task is recorded in the corresponding CPU's rcu_node structure,
  600. * which is checked elsewhere.
  601. *
  602. * Caller must disable hard irqs.
  603. */
  604. static void rcu_preempt_check_callbacks(void)
  605. {
  606. struct task_struct *t = current;
  607. if (t->rcu_read_lock_nesting == 0) {
  608. rcu_preempt_qs();
  609. return;
  610. }
  611. if (t->rcu_read_lock_nesting > 0 &&
  612. __this_cpu_read(rcu_data_p->core_needs_qs) &&
  613. __this_cpu_read(rcu_data_p->cpu_no_qs.b.norm))
  614. t->rcu_read_unlock_special.b.need_qs = true;
  615. }
  616. #ifdef CONFIG_RCU_BOOST
  617. static void rcu_preempt_do_callbacks(void)
  618. {
  619. rcu_do_batch(rcu_state_p, this_cpu_ptr(rcu_data_p));
  620. }
  621. #endif /* #ifdef CONFIG_RCU_BOOST */
  622. /**
  623. * call_rcu() - Queue an RCU callback for invocation after a grace period.
  624. * @head: structure to be used for queueing the RCU updates.
  625. * @func: actual callback function to be invoked after the grace period
  626. *
  627. * The callback function will be invoked some time after a full grace
  628. * period elapses, in other words after all pre-existing RCU read-side
  629. * critical sections have completed. However, the callback function
  630. * might well execute concurrently with RCU read-side critical sections
  631. * that started after call_rcu() was invoked. RCU read-side critical
  632. * sections are delimited by rcu_read_lock() and rcu_read_unlock(),
  633. * and may be nested.
  634. *
  635. * Note that all CPUs must agree that the grace period extended beyond
  636. * all pre-existing RCU read-side critical section. On systems with more
  637. * than one CPU, this means that when "func()" is invoked, each CPU is
  638. * guaranteed to have executed a full memory barrier since the end of its
  639. * last RCU read-side critical section whose beginning preceded the call
  640. * to call_rcu(). It also means that each CPU executing an RCU read-side
  641. * critical section that continues beyond the start of "func()" must have
  642. * executed a memory barrier after the call_rcu() but before the beginning
  643. * of that RCU read-side critical section. Note that these guarantees
  644. * include CPUs that are offline, idle, or executing in user mode, as
  645. * well as CPUs that are executing in the kernel.
  646. *
  647. * Furthermore, if CPU A invoked call_rcu() and CPU B invoked the
  648. * resulting RCU callback function "func()", then both CPU A and CPU B are
  649. * guaranteed to execute a full memory barrier during the time interval
  650. * between the call to call_rcu() and the invocation of "func()" -- even
  651. * if CPU A and CPU B are the same CPU (but again only if the system has
  652. * more than one CPU).
  653. */
  654. void call_rcu(struct rcu_head *head, rcu_callback_t func)
  655. {
  656. __call_rcu(head, func, rcu_state_p, -1, 0);
  657. }
  658. EXPORT_SYMBOL_GPL(call_rcu);
  659. /**
  660. * synchronize_rcu - wait until a grace period has elapsed.
  661. *
  662. * Control will return to the caller some time after a full grace
  663. * period has elapsed, in other words after all currently executing RCU
  664. * read-side critical sections have completed. Note, however, that
  665. * upon return from synchronize_rcu(), the caller might well be executing
  666. * concurrently with new RCU read-side critical sections that began while
  667. * synchronize_rcu() was waiting. RCU read-side critical sections are
  668. * delimited by rcu_read_lock() and rcu_read_unlock(), and may be nested.
  669. *
  670. * See the description of synchronize_sched() for more detailed
  671. * information on memory-ordering guarantees. However, please note
  672. * that -only- the memory-ordering guarantees apply. For example,
  673. * synchronize_rcu() is -not- guaranteed to wait on things like code
  674. * protected by preempt_disable(), instead, synchronize_rcu() is -only-
  675. * guaranteed to wait on RCU read-side critical sections, that is, sections
  676. * of code protected by rcu_read_lock().
  677. */
  678. void synchronize_rcu(void)
  679. {
  680. RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map) ||
  681. lock_is_held(&rcu_lock_map) ||
  682. lock_is_held(&rcu_sched_lock_map),
  683. "Illegal synchronize_rcu() in RCU read-side critical section");
  684. if (rcu_scheduler_active == RCU_SCHEDULER_INACTIVE)
  685. return;
  686. if (rcu_gp_is_expedited())
  687. synchronize_rcu_expedited();
  688. else
  689. wait_rcu_gp(call_rcu);
  690. }
  691. EXPORT_SYMBOL_GPL(synchronize_rcu);
  692. /**
  693. * rcu_barrier - Wait until all in-flight call_rcu() callbacks complete.
  694. *
  695. * Note that this primitive does not necessarily wait for an RCU grace period
  696. * to complete. For example, if there are no RCU callbacks queued anywhere
  697. * in the system, then rcu_barrier() is within its rights to return
  698. * immediately, without waiting for anything, much less an RCU grace period.
  699. */
  700. void rcu_barrier(void)
  701. {
  702. _rcu_barrier(rcu_state_p);
  703. }
  704. EXPORT_SYMBOL_GPL(rcu_barrier);
  705. /*
  706. * Initialize preemptible RCU's state structures.
  707. */
  708. static void __init __rcu_init_preempt(void)
  709. {
  710. rcu_init_one(rcu_state_p);
  711. }
  712. /*
  713. * Check for a task exiting while in a preemptible-RCU read-side
  714. * critical section, clean up if so. No need to issue warnings,
  715. * as debug_check_no_locks_held() already does this if lockdep
  716. * is enabled.
  717. */
  718. void exit_rcu(void)
  719. {
  720. struct task_struct *t = current;
  721. if (likely(list_empty(&current->rcu_node_entry)))
  722. return;
  723. t->rcu_read_lock_nesting = 1;
  724. barrier();
  725. t->rcu_read_unlock_special.b.blocked = true;
  726. __rcu_read_unlock();
  727. }
  728. #else /* #ifdef CONFIG_PREEMPT_RCU */
  729. static struct rcu_state *const rcu_state_p = &rcu_sched_state;
  730. /*
  731. * Tell them what RCU they are running.
  732. */
  733. static void __init rcu_bootup_announce(void)
  734. {
  735. pr_info("Hierarchical RCU implementation.\n");
  736. rcu_bootup_announce_oddness();
  737. }
  738. /*
  739. * Because preemptible RCU does not exist, we never have to check for
  740. * CPUs being in quiescent states.
  741. */
  742. static void rcu_preempt_note_context_switch(bool preempt)
  743. {
  744. }
  745. /*
  746. * Because preemptible RCU does not exist, there are never any preempted
  747. * RCU readers.
  748. */
  749. static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp)
  750. {
  751. return 0;
  752. }
  753. /*
  754. * Because there is no preemptible RCU, there can be no readers blocked.
  755. */
  756. static bool rcu_preempt_has_tasks(struct rcu_node *rnp)
  757. {
  758. return false;
  759. }
  760. /*
  761. * Because preemptible RCU does not exist, we never have to check for
  762. * tasks blocked within RCU read-side critical sections.
  763. */
  764. static void rcu_print_detail_task_stall(struct rcu_state *rsp)
  765. {
  766. }
  767. /*
  768. * Because preemptible RCU does not exist, we never have to check for
  769. * tasks blocked within RCU read-side critical sections.
  770. */
  771. static int rcu_print_task_stall(struct rcu_node *rnp)
  772. {
  773. return 0;
  774. }
  775. /*
  776. * Because preemptible RCU does not exist, we never have to check for
  777. * tasks blocked within RCU read-side critical sections that are
  778. * blocking the current expedited grace period.
  779. */
  780. static int rcu_print_task_exp_stall(struct rcu_node *rnp)
  781. {
  782. return 0;
  783. }
  784. /*
  785. * Because there is no preemptible RCU, there can be no readers blocked,
  786. * so there is no need to check for blocked tasks. So check only for
  787. * bogus qsmask values.
  788. */
  789. static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp)
  790. {
  791. WARN_ON_ONCE(rnp->qsmask);
  792. }
  793. /*
  794. * Because preemptible RCU does not exist, it never has any callbacks
  795. * to check.
  796. */
  797. static void rcu_preempt_check_callbacks(void)
  798. {
  799. }
  800. /*
  801. * Because preemptible RCU does not exist, rcu_barrier() is just
  802. * another name for rcu_barrier_sched().
  803. */
  804. void rcu_barrier(void)
  805. {
  806. rcu_barrier_sched();
  807. }
  808. EXPORT_SYMBOL_GPL(rcu_barrier);
  809. /*
  810. * Because preemptible RCU does not exist, it need not be initialized.
  811. */
  812. static void __init __rcu_init_preempt(void)
  813. {
  814. }
  815. /*
  816. * Because preemptible RCU does not exist, tasks cannot possibly exit
  817. * while in preemptible RCU read-side critical sections.
  818. */
  819. void exit_rcu(void)
  820. {
  821. }
  822. #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
  823. #ifdef CONFIG_RCU_BOOST
  824. #include "../locking/rtmutex_common.h"
  825. static void rcu_wake_cond(struct task_struct *t, int status)
  826. {
  827. /*
  828. * If the thread is yielding, only wake it when this
  829. * is invoked from idle
  830. */
  831. if (status != RCU_KTHREAD_YIELDING || is_idle_task(current))
  832. wake_up_process(t);
  833. }
  834. /*
  835. * Carry out RCU priority boosting on the task indicated by ->exp_tasks
  836. * or ->boost_tasks, advancing the pointer to the next task in the
  837. * ->blkd_tasks list.
  838. *
  839. * Note that irqs must be enabled: boosting the task can block.
  840. * Returns 1 if there are more tasks needing to be boosted.
  841. */
  842. static int rcu_boost(struct rcu_node *rnp)
  843. {
  844. unsigned long flags;
  845. struct task_struct *t;
  846. struct list_head *tb;
  847. if (READ_ONCE(rnp->exp_tasks) == NULL &&
  848. READ_ONCE(rnp->boost_tasks) == NULL)
  849. return 0; /* Nothing left to boost. */
  850. raw_spin_lock_irqsave_rcu_node(rnp, flags);
  851. /*
  852. * Recheck under the lock: all tasks in need of boosting
  853. * might exit their RCU read-side critical sections on their own.
  854. */
  855. if (rnp->exp_tasks == NULL && rnp->boost_tasks == NULL) {
  856. raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
  857. return 0;
  858. }
  859. /*
  860. * Preferentially boost tasks blocking expedited grace periods.
  861. * This cannot starve the normal grace periods because a second
  862. * expedited grace period must boost all blocked tasks, including
  863. * those blocking the pre-existing normal grace period.
  864. */
  865. if (rnp->exp_tasks != NULL) {
  866. tb = rnp->exp_tasks;
  867. rnp->n_exp_boosts++;
  868. } else {
  869. tb = rnp->boost_tasks;
  870. rnp->n_normal_boosts++;
  871. }
  872. rnp->n_tasks_boosted++;
  873. /*
  874. * We boost task t by manufacturing an rt_mutex that appears to
  875. * be held by task t. We leave a pointer to that rt_mutex where
  876. * task t can find it, and task t will release the mutex when it
  877. * exits its outermost RCU read-side critical section. Then
  878. * simply acquiring this artificial rt_mutex will boost task
  879. * t's priority. (Thanks to tglx for suggesting this approach!)
  880. *
  881. * Note that task t must acquire rnp->lock to remove itself from
  882. * the ->blkd_tasks list, which it will do from exit() if from
  883. * nowhere else. We therefore are guaranteed that task t will
  884. * stay around at least until we drop rnp->lock. Note that
  885. * rnp->lock also resolves races between our priority boosting
  886. * and task t's exiting its outermost RCU read-side critical
  887. * section.
  888. */
  889. t = container_of(tb, struct task_struct, rcu_node_entry);
  890. rt_mutex_init_proxy_locked(&rnp->boost_mtx, t);
  891. raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
  892. /* Lock only for side effect: boosts task t's priority. */
  893. rt_mutex_lock(&rnp->boost_mtx);
  894. rt_mutex_unlock(&rnp->boost_mtx); /* Then keep lockdep happy. */
  895. return READ_ONCE(rnp->exp_tasks) != NULL ||
  896. READ_ONCE(rnp->boost_tasks) != NULL;
  897. }
  898. /*
  899. * Priority-boosting kthread, one per leaf rcu_node.
  900. */
  901. static int rcu_boost_kthread(void *arg)
  902. {
  903. struct rcu_node *rnp = (struct rcu_node *)arg;
  904. int spincnt = 0;
  905. int more2boost;
  906. trace_rcu_utilization(TPS("Start boost kthread@init"));
  907. for (;;) {
  908. rnp->boost_kthread_status = RCU_KTHREAD_WAITING;
  909. trace_rcu_utilization(TPS("End boost kthread@rcu_wait"));
  910. rcu_wait(rnp->boost_tasks || rnp->exp_tasks);
  911. trace_rcu_utilization(TPS("Start boost kthread@rcu_wait"));
  912. rnp->boost_kthread_status = RCU_KTHREAD_RUNNING;
  913. more2boost = rcu_boost(rnp);
  914. if (more2boost)
  915. spincnt++;
  916. else
  917. spincnt = 0;
  918. if (spincnt > 10) {
  919. rnp->boost_kthread_status = RCU_KTHREAD_YIELDING;
  920. trace_rcu_utilization(TPS("End boost kthread@rcu_yield"));
  921. schedule_timeout_interruptible(2);
  922. trace_rcu_utilization(TPS("Start boost kthread@rcu_yield"));
  923. spincnt = 0;
  924. }
  925. }
  926. /* NOTREACHED */
  927. trace_rcu_utilization(TPS("End boost kthread@notreached"));
  928. return 0;
  929. }
  930. /*
  931. * Check to see if it is time to start boosting RCU readers that are
  932. * blocking the current grace period, and, if so, tell the per-rcu_node
  933. * kthread to start boosting them. If there is an expedited grace
  934. * period in progress, it is always time to boost.
  935. *
  936. * The caller must hold rnp->lock, which this function releases.
  937. * The ->boost_kthread_task is immortal, so we don't need to worry
  938. * about it going away.
  939. */
  940. static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags)
  941. __releases(rnp->lock)
  942. {
  943. struct task_struct *t;
  944. lockdep_assert_held(&rnp->lock);
  945. if (!rcu_preempt_blocked_readers_cgp(rnp) && rnp->exp_tasks == NULL) {
  946. raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
  947. return;
  948. }
  949. if (rnp->exp_tasks != NULL ||
  950. (rnp->gp_tasks != NULL &&
  951. rnp->boost_tasks == NULL &&
  952. rnp->qsmask == 0 &&
  953. ULONG_CMP_GE(jiffies, rnp->boost_time))) {
  954. if (rnp->exp_tasks == NULL)
  955. rnp->boost_tasks = rnp->gp_tasks;
  956. raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
  957. t = rnp->boost_kthread_task;
  958. if (t)
  959. rcu_wake_cond(t, rnp->boost_kthread_status);
  960. } else {
  961. raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
  962. }
  963. }
  964. /*
  965. * Wake up the per-CPU kthread to invoke RCU callbacks.
  966. */
  967. static void invoke_rcu_callbacks_kthread(void)
  968. {
  969. unsigned long flags;
  970. local_irq_save(flags);
  971. __this_cpu_write(rcu_cpu_has_work, 1);
  972. if (__this_cpu_read(rcu_cpu_kthread_task) != NULL &&
  973. current != __this_cpu_read(rcu_cpu_kthread_task)) {
  974. rcu_wake_cond(__this_cpu_read(rcu_cpu_kthread_task),
  975. __this_cpu_read(rcu_cpu_kthread_status));
  976. }
  977. local_irq_restore(flags);
  978. }
  979. /*
  980. * Is the current CPU running the RCU-callbacks kthread?
  981. * Caller must have preemption disabled.
  982. */
  983. static bool rcu_is_callbacks_kthread(void)
  984. {
  985. return __this_cpu_read(rcu_cpu_kthread_task) == current;
  986. }
  987. #define RCU_BOOST_DELAY_JIFFIES DIV_ROUND_UP(CONFIG_RCU_BOOST_DELAY * HZ, 1000)
  988. /*
  989. * Do priority-boost accounting for the start of a new grace period.
  990. */
  991. static void rcu_preempt_boost_start_gp(struct rcu_node *rnp)
  992. {
  993. rnp->boost_time = jiffies + RCU_BOOST_DELAY_JIFFIES;
  994. }
  995. /*
  996. * Create an RCU-boost kthread for the specified node if one does not
  997. * already exist. We only create this kthread for preemptible RCU.
  998. * Returns zero if all is well, a negated errno otherwise.
  999. */
  1000. static int rcu_spawn_one_boost_kthread(struct rcu_state *rsp,
  1001. struct rcu_node *rnp)
  1002. {
  1003. int rnp_index = rnp - &rsp->node[0];
  1004. unsigned long flags;
  1005. struct sched_param sp;
  1006. struct task_struct *t;
  1007. if (rcu_state_p != rsp)
  1008. return 0;
  1009. if (!rcu_scheduler_fully_active || rcu_rnp_online_cpus(rnp) == 0)
  1010. return 0;
  1011. rsp->boost = 1;
  1012. if (rnp->boost_kthread_task != NULL)
  1013. return 0;
  1014. t = kthread_create(rcu_boost_kthread, (void *)rnp,
  1015. "rcub/%d", rnp_index);
  1016. if (IS_ERR(t))
  1017. return PTR_ERR(t);
  1018. raw_spin_lock_irqsave_rcu_node(rnp, flags);
  1019. rnp->boost_kthread_task = t;
  1020. raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
  1021. sp.sched_priority = kthread_prio;
  1022. sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
  1023. wake_up_process(t); /* get to TASK_INTERRUPTIBLE quickly. */
  1024. return 0;
  1025. }
  1026. static void rcu_kthread_do_work(void)
  1027. {
  1028. rcu_do_batch(&rcu_sched_state, this_cpu_ptr(&rcu_sched_data));
  1029. rcu_do_batch(&rcu_bh_state, this_cpu_ptr(&rcu_bh_data));
  1030. rcu_preempt_do_callbacks();
  1031. }
  1032. static void rcu_cpu_kthread_setup(unsigned int cpu)
  1033. {
  1034. struct sched_param sp;
  1035. sp.sched_priority = kthread_prio;
  1036. sched_setscheduler_nocheck(current, SCHED_FIFO, &sp);
  1037. }
  1038. static void rcu_cpu_kthread_park(unsigned int cpu)
  1039. {
  1040. per_cpu(rcu_cpu_kthread_status, cpu) = RCU_KTHREAD_OFFCPU;
  1041. }
  1042. static int rcu_cpu_kthread_should_run(unsigned int cpu)
  1043. {
  1044. return __this_cpu_read(rcu_cpu_has_work);
  1045. }
  1046. /*
  1047. * Per-CPU kernel thread that invokes RCU callbacks. This replaces the
  1048. * RCU softirq used in flavors and configurations of RCU that do not
  1049. * support RCU priority boosting.
  1050. */
  1051. static void rcu_cpu_kthread(unsigned int cpu)
  1052. {
  1053. unsigned int *statusp = this_cpu_ptr(&rcu_cpu_kthread_status);
  1054. char work, *workp = this_cpu_ptr(&rcu_cpu_has_work);
  1055. int spincnt;
  1056. for (spincnt = 0; spincnt < 10; spincnt++) {
  1057. trace_rcu_utilization(TPS("Start CPU kthread@rcu_wait"));
  1058. local_bh_disable();
  1059. *statusp = RCU_KTHREAD_RUNNING;
  1060. this_cpu_inc(rcu_cpu_kthread_loops);
  1061. local_irq_disable();
  1062. work = *workp;
  1063. *workp = 0;
  1064. local_irq_enable();
  1065. if (work)
  1066. rcu_kthread_do_work();
  1067. local_bh_enable();
  1068. if (*workp == 0) {
  1069. trace_rcu_utilization(TPS("End CPU kthread@rcu_wait"));
  1070. *statusp = RCU_KTHREAD_WAITING;
  1071. return;
  1072. }
  1073. }
  1074. *statusp = RCU_KTHREAD_YIELDING;
  1075. trace_rcu_utilization(TPS("Start CPU kthread@rcu_yield"));
  1076. schedule_timeout_interruptible(2);
  1077. trace_rcu_utilization(TPS("End CPU kthread@rcu_yield"));
  1078. *statusp = RCU_KTHREAD_WAITING;
  1079. }
  1080. /*
  1081. * Set the per-rcu_node kthread's affinity to cover all CPUs that are
  1082. * served by the rcu_node in question. The CPU hotplug lock is still
  1083. * held, so the value of rnp->qsmaskinit will be stable.
  1084. *
  1085. * We don't include outgoingcpu in the affinity set, use -1 if there is
  1086. * no outgoing CPU. If there are no CPUs left in the affinity set,
  1087. * this function allows the kthread to execute on any CPU.
  1088. */
  1089. static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu)
  1090. {
  1091. struct task_struct *t = rnp->boost_kthread_task;
  1092. unsigned long mask = rcu_rnp_online_cpus(rnp);
  1093. cpumask_var_t cm;
  1094. int cpu;
  1095. if (!t)
  1096. return;
  1097. if (!zalloc_cpumask_var(&cm, GFP_KERNEL))
  1098. return;
  1099. for_each_leaf_node_possible_cpu(rnp, cpu)
  1100. if ((mask & leaf_node_cpu_bit(rnp, cpu)) &&
  1101. cpu != outgoingcpu)
  1102. cpumask_set_cpu(cpu, cm);
  1103. if (cpumask_weight(cm) == 0)
  1104. cpumask_setall(cm);
  1105. set_cpus_allowed_ptr(t, cm);
  1106. free_cpumask_var(cm);
  1107. }
  1108. static struct smp_hotplug_thread rcu_cpu_thread_spec = {
  1109. .store = &rcu_cpu_kthread_task,
  1110. .thread_should_run = rcu_cpu_kthread_should_run,
  1111. .thread_fn = rcu_cpu_kthread,
  1112. .thread_comm = "rcuc/%u",
  1113. .setup = rcu_cpu_kthread_setup,
  1114. .park = rcu_cpu_kthread_park,
  1115. };
  1116. /*
  1117. * Spawn boost kthreads -- called as soon as the scheduler is running.
  1118. */
  1119. static void __init rcu_spawn_boost_kthreads(void)
  1120. {
  1121. struct rcu_node *rnp;
  1122. int cpu;
  1123. for_each_possible_cpu(cpu)
  1124. per_cpu(rcu_cpu_has_work, cpu) = 0;
  1125. BUG_ON(smpboot_register_percpu_thread(&rcu_cpu_thread_spec));
  1126. rcu_for_each_leaf_node(rcu_state_p, rnp)
  1127. (void)rcu_spawn_one_boost_kthread(rcu_state_p, rnp);
  1128. }
  1129. static void rcu_prepare_kthreads(int cpu)
  1130. {
  1131. struct rcu_data *rdp = per_cpu_ptr(rcu_state_p->rda, cpu);
  1132. struct rcu_node *rnp = rdp->mynode;
  1133. /* Fire up the incoming CPU's kthread and leaf rcu_node kthread. */
  1134. if (rcu_scheduler_fully_active)
  1135. (void)rcu_spawn_one_boost_kthread(rcu_state_p, rnp);
  1136. }
  1137. #else /* #ifdef CONFIG_RCU_BOOST */
  1138. static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags)
  1139. __releases(rnp->lock)
  1140. {
  1141. raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
  1142. }
  1143. static void invoke_rcu_callbacks_kthread(void)
  1144. {
  1145. WARN_ON_ONCE(1);
  1146. }
  1147. static bool rcu_is_callbacks_kthread(void)
  1148. {
  1149. return false;
  1150. }
  1151. static void rcu_preempt_boost_start_gp(struct rcu_node *rnp)
  1152. {
  1153. }
  1154. static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu)
  1155. {
  1156. }
  1157. static void __init rcu_spawn_boost_kthreads(void)
  1158. {
  1159. }
  1160. static void rcu_prepare_kthreads(int cpu)
  1161. {
  1162. }
  1163. #endif /* #else #ifdef CONFIG_RCU_BOOST */
  1164. #if !defined(CONFIG_RCU_FAST_NO_HZ)
  1165. /*
  1166. * Check to see if any future RCU-related work will need to be done
  1167. * by the current CPU, even if none need be done immediately, returning
  1168. * 1 if so. This function is part of the RCU implementation; it is -not-
  1169. * an exported member of the RCU API.
  1170. *
  1171. * Because we not have RCU_FAST_NO_HZ, just check whether this CPU needs
  1172. * any flavor of RCU.
  1173. */
  1174. int rcu_needs_cpu(u64 basemono, u64 *nextevt)
  1175. {
  1176. *nextevt = KTIME_MAX;
  1177. return rcu_cpu_has_callbacks(NULL);
  1178. }
  1179. /*
  1180. * Because we do not have RCU_FAST_NO_HZ, don't bother cleaning up
  1181. * after it.
  1182. */
  1183. static void rcu_cleanup_after_idle(void)
  1184. {
  1185. }
  1186. /*
  1187. * Do the idle-entry grace-period work, which, because CONFIG_RCU_FAST_NO_HZ=n,
  1188. * is nothing.
  1189. */
  1190. static void rcu_prepare_for_idle(void)
  1191. {
  1192. }
  1193. /*
  1194. * Don't bother keeping a running count of the number of RCU callbacks
  1195. * posted because CONFIG_RCU_FAST_NO_HZ=n.
  1196. */
  1197. static void rcu_idle_count_callbacks_posted(void)
  1198. {
  1199. }
  1200. #else /* #if !defined(CONFIG_RCU_FAST_NO_HZ) */
  1201. /*
  1202. * This code is invoked when a CPU goes idle, at which point we want
  1203. * to have the CPU do everything required for RCU so that it can enter
  1204. * the energy-efficient dyntick-idle mode. This is handled by a
  1205. * state machine implemented by rcu_prepare_for_idle() below.
  1206. *
  1207. * The following three proprocessor symbols control this state machine:
  1208. *
  1209. * RCU_IDLE_GP_DELAY gives the number of jiffies that a CPU is permitted
  1210. * to sleep in dyntick-idle mode with RCU callbacks pending. This
  1211. * is sized to be roughly one RCU grace period. Those energy-efficiency
  1212. * benchmarkers who might otherwise be tempted to set this to a large
  1213. * number, be warned: Setting RCU_IDLE_GP_DELAY too high can hang your
  1214. * system. And if you are -that- concerned about energy efficiency,
  1215. * just power the system down and be done with it!
  1216. * RCU_IDLE_LAZY_GP_DELAY gives the number of jiffies that a CPU is
  1217. * permitted to sleep in dyntick-idle mode with only lazy RCU
  1218. * callbacks pending. Setting this too high can OOM your system.
  1219. *
  1220. * The values below work well in practice. If future workloads require
  1221. * adjustment, they can be converted into kernel config parameters, though
  1222. * making the state machine smarter might be a better option.
  1223. */
  1224. #define RCU_IDLE_GP_DELAY 4 /* Roughly one grace period. */
  1225. #define RCU_IDLE_LAZY_GP_DELAY (6 * HZ) /* Roughly six seconds. */
  1226. static int rcu_idle_gp_delay = RCU_IDLE_GP_DELAY;
  1227. module_param(rcu_idle_gp_delay, int, 0644);
  1228. static int rcu_idle_lazy_gp_delay = RCU_IDLE_LAZY_GP_DELAY;
  1229. module_param(rcu_idle_lazy_gp_delay, int, 0644);
  1230. /*
  1231. * Try to advance callbacks for all flavors of RCU on the current CPU, but
  1232. * only if it has been awhile since the last time we did so. Afterwards,
  1233. * if there are any callbacks ready for immediate invocation, return true.
  1234. */
  1235. static bool __maybe_unused rcu_try_advance_all_cbs(void)
  1236. {
  1237. bool cbs_ready = false;
  1238. struct rcu_data *rdp;
  1239. struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks);
  1240. struct rcu_node *rnp;
  1241. struct rcu_state *rsp;
  1242. /* Exit early if we advanced recently. */
  1243. if (jiffies == rdtp->last_advance_all)
  1244. return false;
  1245. rdtp->last_advance_all = jiffies;
  1246. for_each_rcu_flavor(rsp) {
  1247. rdp = this_cpu_ptr(rsp->rda);
  1248. rnp = rdp->mynode;
  1249. /*
  1250. * Don't bother checking unless a grace period has
  1251. * completed since we last checked and there are
  1252. * callbacks not yet ready to invoke.
  1253. */
  1254. if ((rdp->completed != rnp->completed ||
  1255. unlikely(READ_ONCE(rdp->gpwrap))) &&
  1256. rcu_segcblist_pend_cbs(&rdp->cblist))
  1257. note_gp_changes(rsp, rdp);
  1258. if (rcu_segcblist_ready_cbs(&rdp->cblist))
  1259. cbs_ready = true;
  1260. }
  1261. return cbs_ready;
  1262. }
  1263. /*
  1264. * Allow the CPU to enter dyntick-idle mode unless it has callbacks ready
  1265. * to invoke. If the CPU has callbacks, try to advance them. Tell the
  1266. * caller to set the timeout based on whether or not there are non-lazy
  1267. * callbacks.
  1268. *
  1269. * The caller must have disabled interrupts.
  1270. */
  1271. int rcu_needs_cpu(u64 basemono, u64 *nextevt)
  1272. {
  1273. struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks);
  1274. unsigned long dj;
  1275. RCU_LOCKDEP_WARN(!irqs_disabled(), "rcu_needs_cpu() invoked with irqs enabled!!!");
  1276. /* Snapshot to detect later posting of non-lazy callback. */
  1277. rdtp->nonlazy_posted_snap = rdtp->nonlazy_posted;
  1278. /* If no callbacks, RCU doesn't need the CPU. */
  1279. if (!rcu_cpu_has_callbacks(&rdtp->all_lazy)) {
  1280. *nextevt = KTIME_MAX;
  1281. return 0;
  1282. }
  1283. /* Attempt to advance callbacks. */
  1284. if (rcu_try_advance_all_cbs()) {
  1285. /* Some ready to invoke, so initiate later invocation. */
  1286. invoke_rcu_core();
  1287. return 1;
  1288. }
  1289. rdtp->last_accelerate = jiffies;
  1290. /* Request timer delay depending on laziness, and round. */
  1291. if (!rdtp->all_lazy) {
  1292. dj = round_up(rcu_idle_gp_delay + jiffies,
  1293. rcu_idle_gp_delay) - jiffies;
  1294. } else {
  1295. dj = round_jiffies(rcu_idle_lazy_gp_delay + jiffies) - jiffies;
  1296. }
  1297. *nextevt = basemono + dj * TICK_NSEC;
  1298. return 0;
  1299. }
  1300. /*
  1301. * Prepare a CPU for idle from an RCU perspective. The first major task
  1302. * is to sense whether nohz mode has been enabled or disabled via sysfs.
  1303. * The second major task is to check to see if a non-lazy callback has
  1304. * arrived at a CPU that previously had only lazy callbacks. The third
  1305. * major task is to accelerate (that is, assign grace-period numbers to)
  1306. * any recently arrived callbacks.
  1307. *
  1308. * The caller must have disabled interrupts.
  1309. */
  1310. static void rcu_prepare_for_idle(void)
  1311. {
  1312. bool needwake;
  1313. struct rcu_data *rdp;
  1314. struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks);
  1315. struct rcu_node *rnp;
  1316. struct rcu_state *rsp;
  1317. int tne;
  1318. RCU_LOCKDEP_WARN(!irqs_disabled(), "rcu_prepare_for_idle() invoked with irqs enabled!!!");
  1319. if (rcu_is_nocb_cpu(smp_processor_id()))
  1320. return;
  1321. /* Handle nohz enablement switches conservatively. */
  1322. tne = READ_ONCE(tick_nohz_active);
  1323. if (tne != rdtp->tick_nohz_enabled_snap) {
  1324. if (rcu_cpu_has_callbacks(NULL))
  1325. invoke_rcu_core(); /* force nohz to see update. */
  1326. rdtp->tick_nohz_enabled_snap = tne;
  1327. return;
  1328. }
  1329. if (!tne)
  1330. return;
  1331. /*
  1332. * If a non-lazy callback arrived at a CPU having only lazy
  1333. * callbacks, invoke RCU core for the side-effect of recalculating
  1334. * idle duration on re-entry to idle.
  1335. */
  1336. if (rdtp->all_lazy &&
  1337. rdtp->nonlazy_posted != rdtp->nonlazy_posted_snap) {
  1338. rdtp->all_lazy = false;
  1339. rdtp->nonlazy_posted_snap = rdtp->nonlazy_posted;
  1340. invoke_rcu_core();
  1341. return;
  1342. }
  1343. /*
  1344. * If we have not yet accelerated this jiffy, accelerate all
  1345. * callbacks on this CPU.
  1346. */
  1347. if (rdtp->last_accelerate == jiffies)
  1348. return;
  1349. rdtp->last_accelerate = jiffies;
  1350. for_each_rcu_flavor(rsp) {
  1351. rdp = this_cpu_ptr(rsp->rda);
  1352. if (rcu_segcblist_pend_cbs(&rdp->cblist))
  1353. continue;
  1354. rnp = rdp->mynode;
  1355. raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */
  1356. needwake = rcu_accelerate_cbs(rsp, rnp, rdp);
  1357. raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */
  1358. if (needwake)
  1359. rcu_gp_kthread_wake(rsp);
  1360. }
  1361. }
  1362. /*
  1363. * Clean up for exit from idle. Attempt to advance callbacks based on
  1364. * any grace periods that elapsed while the CPU was idle, and if any
  1365. * callbacks are now ready to invoke, initiate invocation.
  1366. */
  1367. static void rcu_cleanup_after_idle(void)
  1368. {
  1369. RCU_LOCKDEP_WARN(!irqs_disabled(), "rcu_cleanup_after_idle() invoked with irqs enabled!!!");
  1370. if (rcu_is_nocb_cpu(smp_processor_id()))
  1371. return;
  1372. if (rcu_try_advance_all_cbs())
  1373. invoke_rcu_core();
  1374. }
  1375. /*
  1376. * Keep a running count of the number of non-lazy callbacks posted
  1377. * on this CPU. This running counter (which is never decremented) allows
  1378. * rcu_prepare_for_idle() to detect when something out of the idle loop
  1379. * posts a callback, even if an equal number of callbacks are invoked.
  1380. * Of course, callbacks should only be posted from within a trace event
  1381. * designed to be called from idle or from within RCU_NONIDLE().
  1382. */
  1383. static void rcu_idle_count_callbacks_posted(void)
  1384. {
  1385. __this_cpu_add(rcu_dynticks.nonlazy_posted, 1);
  1386. }
  1387. /*
  1388. * Data for flushing lazy RCU callbacks at OOM time.
  1389. */
  1390. static atomic_t oom_callback_count;
  1391. static DECLARE_WAIT_QUEUE_HEAD(oom_callback_wq);
  1392. /*
  1393. * RCU OOM callback -- decrement the outstanding count and deliver the
  1394. * wake-up if we are the last one.
  1395. */
  1396. static void rcu_oom_callback(struct rcu_head *rhp)
  1397. {
  1398. if (atomic_dec_and_test(&oom_callback_count))
  1399. wake_up(&oom_callback_wq);
  1400. }
  1401. /*
  1402. * Post an rcu_oom_notify callback on the current CPU if it has at
  1403. * least one lazy callback. This will unnecessarily post callbacks
  1404. * to CPUs that already have a non-lazy callback at the end of their
  1405. * callback list, but this is an infrequent operation, so accept some
  1406. * extra overhead to keep things simple.
  1407. */
  1408. static void rcu_oom_notify_cpu(void *unused)
  1409. {
  1410. struct rcu_state *rsp;
  1411. struct rcu_data *rdp;
  1412. for_each_rcu_flavor(rsp) {
  1413. rdp = raw_cpu_ptr(rsp->rda);
  1414. if (rcu_segcblist_n_lazy_cbs(&rdp->cblist)) {
  1415. atomic_inc(&oom_callback_count);
  1416. rsp->call(&rdp->oom_head, rcu_oom_callback);
  1417. }
  1418. }
  1419. }
  1420. /*
  1421. * If low on memory, ensure that each CPU has a non-lazy callback.
  1422. * This will wake up CPUs that have only lazy callbacks, in turn
  1423. * ensuring that they free up the corresponding memory in a timely manner.
  1424. * Because an uncertain amount of memory will be freed in some uncertain
  1425. * timeframe, we do not claim to have freed anything.
  1426. */
  1427. static int rcu_oom_notify(struct notifier_block *self,
  1428. unsigned long notused, void *nfreed)
  1429. {
  1430. int cpu;
  1431. /* Wait for callbacks from earlier instance to complete. */
  1432. wait_event(oom_callback_wq, atomic_read(&oom_callback_count) == 0);
  1433. smp_mb(); /* Ensure callback reuse happens after callback invocation. */
  1434. /*
  1435. * Prevent premature wakeup: ensure that all increments happen
  1436. * before there is a chance of the counter reaching zero.
  1437. */
  1438. atomic_set(&oom_callback_count, 1);
  1439. for_each_online_cpu(cpu) {
  1440. smp_call_function_single(cpu, rcu_oom_notify_cpu, NULL, 1);
  1441. cond_resched_rcu_qs();
  1442. }
  1443. /* Unconditionally decrement: no need to wake ourselves up. */
  1444. atomic_dec(&oom_callback_count);
  1445. return NOTIFY_OK;
  1446. }
  1447. static struct notifier_block rcu_oom_nb = {
  1448. .notifier_call = rcu_oom_notify
  1449. };
  1450. static int __init rcu_register_oom_notifier(void)
  1451. {
  1452. register_oom_notifier(&rcu_oom_nb);
  1453. return 0;
  1454. }
  1455. early_initcall(rcu_register_oom_notifier);
  1456. #endif /* #else #if !defined(CONFIG_RCU_FAST_NO_HZ) */
  1457. #ifdef CONFIG_RCU_FAST_NO_HZ
  1458. static void print_cpu_stall_fast_no_hz(char *cp, int cpu)
  1459. {
  1460. struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu);
  1461. unsigned long nlpd = rdtp->nonlazy_posted - rdtp->nonlazy_posted_snap;
  1462. sprintf(cp, "last_accelerate: %04lx/%04lx, nonlazy_posted: %ld, %c%c",
  1463. rdtp->last_accelerate & 0xffff, jiffies & 0xffff,
  1464. ulong2long(nlpd),
  1465. rdtp->all_lazy ? 'L' : '.',
  1466. rdtp->tick_nohz_enabled_snap ? '.' : 'D');
  1467. }
  1468. #else /* #ifdef CONFIG_RCU_FAST_NO_HZ */
  1469. static void print_cpu_stall_fast_no_hz(char *cp, int cpu)
  1470. {
  1471. *cp = '\0';
  1472. }
  1473. #endif /* #else #ifdef CONFIG_RCU_FAST_NO_HZ */
  1474. /* Initiate the stall-info list. */
  1475. static void print_cpu_stall_info_begin(void)
  1476. {
  1477. pr_cont("\n");
  1478. }
  1479. /*
  1480. * Print out diagnostic information for the specified stalled CPU.
  1481. *
  1482. * If the specified CPU is aware of the current RCU grace period
  1483. * (flavor specified by rsp), then print the number of scheduling
  1484. * clock interrupts the CPU has taken during the time that it has
  1485. * been aware. Otherwise, print the number of RCU grace periods
  1486. * that this CPU is ignorant of, for example, "1" if the CPU was
  1487. * aware of the previous grace period.
  1488. *
  1489. * Also print out idle and (if CONFIG_RCU_FAST_NO_HZ) idle-entry info.
  1490. */
  1491. static void print_cpu_stall_info(struct rcu_state *rsp, int cpu)
  1492. {
  1493. char fast_no_hz[72];
  1494. struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
  1495. struct rcu_dynticks *rdtp = rdp->dynticks;
  1496. char *ticks_title;
  1497. unsigned long ticks_value;
  1498. if (rsp->gpnum == rdp->gpnum) {
  1499. ticks_title = "ticks this GP";
  1500. ticks_value = rdp->ticks_this_gp;
  1501. } else {
  1502. ticks_title = "GPs behind";
  1503. ticks_value = rsp->gpnum - rdp->gpnum;
  1504. }
  1505. print_cpu_stall_fast_no_hz(fast_no_hz, cpu);
  1506. pr_err("\t%d-%c%c%c: (%lu %s) idle=%03x/%llx/%d softirq=%u/%u fqs=%ld %s\n",
  1507. cpu,
  1508. "O."[!!cpu_online(cpu)],
  1509. "o."[!!(rdp->grpmask & rdp->mynode->qsmaskinit)],
  1510. "N."[!!(rdp->grpmask & rdp->mynode->qsmaskinitnext)],
  1511. ticks_value, ticks_title,
  1512. rcu_dynticks_snap(rdtp) & 0xfff,
  1513. rdtp->dynticks_nesting, rdtp->dynticks_nmi_nesting,
  1514. rdp->softirq_snap, kstat_softirqs_cpu(RCU_SOFTIRQ, cpu),
  1515. READ_ONCE(rsp->n_force_qs) - rsp->n_force_qs_gpstart,
  1516. fast_no_hz);
  1517. }
  1518. /* Terminate the stall-info list. */
  1519. static void print_cpu_stall_info_end(void)
  1520. {
  1521. pr_err("\t");
  1522. }
  1523. /* Zero ->ticks_this_gp for all flavors of RCU. */
  1524. static void zero_cpu_stall_ticks(struct rcu_data *rdp)
  1525. {
  1526. rdp->ticks_this_gp = 0;
  1527. rdp->softirq_snap = kstat_softirqs_cpu(RCU_SOFTIRQ, smp_processor_id());
  1528. }
  1529. /* Increment ->ticks_this_gp for all flavors of RCU. */
  1530. static void increment_cpu_stall_ticks(void)
  1531. {
  1532. struct rcu_state *rsp;
  1533. for_each_rcu_flavor(rsp)
  1534. raw_cpu_inc(rsp->rda->ticks_this_gp);
  1535. }
  1536. #ifdef CONFIG_RCU_NOCB_CPU
  1537. /*
  1538. * Offload callback processing from the boot-time-specified set of CPUs
  1539. * specified by rcu_nocb_mask. For each CPU in the set, there is a
  1540. * kthread created that pulls the callbacks from the corresponding CPU,
  1541. * waits for a grace period to elapse, and invokes the callbacks.
  1542. * The no-CBs CPUs do a wake_up() on their kthread when they insert
  1543. * a callback into any empty list, unless the rcu_nocb_poll boot parameter
  1544. * has been specified, in which case each kthread actively polls its
  1545. * CPU. (Which isn't so great for energy efficiency, but which does
  1546. * reduce RCU's overhead on that CPU.)
  1547. *
  1548. * This is intended to be used in conjunction with Frederic Weisbecker's
  1549. * adaptive-idle work, which would seriously reduce OS jitter on CPUs
  1550. * running CPU-bound user-mode computations.
  1551. *
  1552. * Offloading of callback processing could also in theory be used as
  1553. * an energy-efficiency measure because CPUs with no RCU callbacks
  1554. * queued are more aggressive about entering dyntick-idle mode.
  1555. */
  1556. /* Parse the boot-time rcu_nocb_mask CPU list from the kernel parameters. */
  1557. static int __init rcu_nocb_setup(char *str)
  1558. {
  1559. alloc_bootmem_cpumask_var(&rcu_nocb_mask);
  1560. have_rcu_nocb_mask = true;
  1561. cpulist_parse(str, rcu_nocb_mask);
  1562. return 1;
  1563. }
  1564. __setup("rcu_nocbs=", rcu_nocb_setup);
  1565. static int __init parse_rcu_nocb_poll(char *arg)
  1566. {
  1567. rcu_nocb_poll = true;
  1568. return 0;
  1569. }
  1570. early_param("rcu_nocb_poll", parse_rcu_nocb_poll);
  1571. /*
  1572. * Wake up any no-CBs CPUs' kthreads that were waiting on the just-ended
  1573. * grace period.
  1574. */
  1575. static void rcu_nocb_gp_cleanup(struct swait_queue_head *sq)
  1576. {
  1577. swake_up_all(sq);
  1578. }
  1579. /*
  1580. * Set the root rcu_node structure's ->need_future_gp field
  1581. * based on the sum of those of all rcu_node structures. This does
  1582. * double-count the root rcu_node structure's requests, but this
  1583. * is necessary to handle the possibility of a rcu_nocb_kthread()
  1584. * having awakened during the time that the rcu_node structures
  1585. * were being updated for the end of the previous grace period.
  1586. */
  1587. static void rcu_nocb_gp_set(struct rcu_node *rnp, int nrq)
  1588. {
  1589. rnp->need_future_gp[(rnp->completed + 1) & 0x1] += nrq;
  1590. }
  1591. static struct swait_queue_head *rcu_nocb_gp_get(struct rcu_node *rnp)
  1592. {
  1593. return &rnp->nocb_gp_wq[rnp->completed & 0x1];
  1594. }
  1595. static void rcu_init_one_nocb(struct rcu_node *rnp)
  1596. {
  1597. init_swait_queue_head(&rnp->nocb_gp_wq[0]);
  1598. init_swait_queue_head(&rnp->nocb_gp_wq[1]);
  1599. }
  1600. /* Is the specified CPU a no-CBs CPU? */
  1601. bool rcu_is_nocb_cpu(int cpu)
  1602. {
  1603. if (have_rcu_nocb_mask)
  1604. return cpumask_test_cpu(cpu, rcu_nocb_mask);
  1605. return false;
  1606. }
  1607. /*
  1608. * Kick the leader kthread for this NOCB group. Caller holds ->nocb_lock
  1609. * and this function releases it.
  1610. */
  1611. static void __wake_nocb_leader(struct rcu_data *rdp, bool force,
  1612. unsigned long flags)
  1613. __releases(rdp->nocb_lock)
  1614. {
  1615. struct rcu_data *rdp_leader = rdp->nocb_leader;
  1616. lockdep_assert_held(&rdp->nocb_lock);
  1617. if (!READ_ONCE(rdp_leader->nocb_kthread)) {
  1618. raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags);
  1619. return;
  1620. }
  1621. if (rdp_leader->nocb_leader_sleep || force) {
  1622. /* Prior smp_mb__after_atomic() orders against prior enqueue. */
  1623. WRITE_ONCE(rdp_leader->nocb_leader_sleep, false);
  1624. del_timer(&rdp->nocb_timer);
  1625. raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags);
  1626. smp_mb(); /* ->nocb_leader_sleep before swake_up(). */
  1627. swake_up(&rdp_leader->nocb_wq);
  1628. } else {
  1629. raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags);
  1630. }
  1631. }
  1632. /*
  1633. * Kick the leader kthread for this NOCB group, but caller has not
  1634. * acquired locks.
  1635. */
  1636. static void wake_nocb_leader(struct rcu_data *rdp, bool force)
  1637. {
  1638. unsigned long flags;
  1639. raw_spin_lock_irqsave(&rdp->nocb_lock, flags);
  1640. __wake_nocb_leader(rdp, force, flags);
  1641. }
  1642. /*
  1643. * Arrange to wake the leader kthread for this NOCB group at some
  1644. * future time when it is safe to do so.
  1645. */
  1646. static void wake_nocb_leader_defer(struct rcu_data *rdp, int waketype,
  1647. const char *reason)
  1648. {
  1649. unsigned long flags;
  1650. raw_spin_lock_irqsave(&rdp->nocb_lock, flags);
  1651. if (rdp->nocb_defer_wakeup == RCU_NOCB_WAKE_NOT)
  1652. mod_timer(&rdp->nocb_timer, jiffies + 1);
  1653. WRITE_ONCE(rdp->nocb_defer_wakeup, waketype);
  1654. trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, reason);
  1655. raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags);
  1656. }
  1657. /*
  1658. * Does the specified CPU need an RCU callback for the specified flavor
  1659. * of rcu_barrier()?
  1660. */
  1661. static bool rcu_nocb_cpu_needs_barrier(struct rcu_state *rsp, int cpu)
  1662. {
  1663. struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
  1664. unsigned long ret;
  1665. #ifdef CONFIG_PROVE_RCU
  1666. struct rcu_head *rhp;
  1667. #endif /* #ifdef CONFIG_PROVE_RCU */
  1668. /*
  1669. * Check count of all no-CBs callbacks awaiting invocation.
  1670. * There needs to be a barrier before this function is called,
  1671. * but associated with a prior determination that no more
  1672. * callbacks would be posted. In the worst case, the first
  1673. * barrier in _rcu_barrier() suffices (but the caller cannot
  1674. * necessarily rely on this, not a substitute for the caller
  1675. * getting the concurrency design right!). There must also be
  1676. * a barrier between the following load an posting of a callback
  1677. * (if a callback is in fact needed). This is associated with an
  1678. * atomic_inc() in the caller.
  1679. */
  1680. ret = atomic_long_read(&rdp->nocb_q_count);
  1681. #ifdef CONFIG_PROVE_RCU
  1682. rhp = READ_ONCE(rdp->nocb_head);
  1683. if (!rhp)
  1684. rhp = READ_ONCE(rdp->nocb_gp_head);
  1685. if (!rhp)
  1686. rhp = READ_ONCE(rdp->nocb_follower_head);
  1687. /* Having no rcuo kthread but CBs after scheduler starts is bad! */
  1688. if (!READ_ONCE(rdp->nocb_kthread) && rhp &&
  1689. rcu_scheduler_fully_active) {
  1690. /* RCU callback enqueued before CPU first came online??? */
  1691. pr_err("RCU: Never-onlined no-CBs CPU %d has CB %p\n",
  1692. cpu, rhp->func);
  1693. WARN_ON_ONCE(1);
  1694. }
  1695. #endif /* #ifdef CONFIG_PROVE_RCU */
  1696. return !!ret;
  1697. }
  1698. /*
  1699. * Enqueue the specified string of rcu_head structures onto the specified
  1700. * CPU's no-CBs lists. The CPU is specified by rdp, the head of the
  1701. * string by rhp, and the tail of the string by rhtp. The non-lazy/lazy
  1702. * counts are supplied by rhcount and rhcount_lazy.
  1703. *
  1704. * If warranted, also wake up the kthread servicing this CPUs queues.
  1705. */
  1706. static void __call_rcu_nocb_enqueue(struct rcu_data *rdp,
  1707. struct rcu_head *rhp,
  1708. struct rcu_head **rhtp,
  1709. int rhcount, int rhcount_lazy,
  1710. unsigned long flags)
  1711. {
  1712. int len;
  1713. struct rcu_head **old_rhpp;
  1714. struct task_struct *t;
  1715. /* Enqueue the callback on the nocb list and update counts. */
  1716. atomic_long_add(rhcount, &rdp->nocb_q_count);
  1717. /* rcu_barrier() relies on ->nocb_q_count add before xchg. */
  1718. old_rhpp = xchg(&rdp->nocb_tail, rhtp);
  1719. WRITE_ONCE(*old_rhpp, rhp);
  1720. atomic_long_add(rhcount_lazy, &rdp->nocb_q_count_lazy);
  1721. smp_mb__after_atomic(); /* Store *old_rhpp before _wake test. */
  1722. /* If we are not being polled and there is a kthread, awaken it ... */
  1723. t = READ_ONCE(rdp->nocb_kthread);
  1724. if (rcu_nocb_poll || !t) {
  1725. trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu,
  1726. TPS("WakeNotPoll"));
  1727. return;
  1728. }
  1729. len = atomic_long_read(&rdp->nocb_q_count);
  1730. if (old_rhpp == &rdp->nocb_head) {
  1731. if (!irqs_disabled_flags(flags)) {
  1732. /* ... if queue was empty ... */
  1733. wake_nocb_leader(rdp, false);
  1734. trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu,
  1735. TPS("WakeEmpty"));
  1736. } else {
  1737. wake_nocb_leader_defer(rdp, RCU_NOCB_WAKE,
  1738. TPS("WakeEmptyIsDeferred"));
  1739. }
  1740. rdp->qlen_last_fqs_check = 0;
  1741. } else if (len > rdp->qlen_last_fqs_check + qhimark) {
  1742. /* ... or if many callbacks queued. */
  1743. if (!irqs_disabled_flags(flags)) {
  1744. wake_nocb_leader(rdp, true);
  1745. trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu,
  1746. TPS("WakeOvf"));
  1747. } else {
  1748. wake_nocb_leader_defer(rdp, RCU_NOCB_WAKE,
  1749. TPS("WakeOvfIsDeferred"));
  1750. }
  1751. rdp->qlen_last_fqs_check = LONG_MAX / 2;
  1752. } else {
  1753. trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, TPS("WakeNot"));
  1754. }
  1755. return;
  1756. }
  1757. /*
  1758. * This is a helper for __call_rcu(), which invokes this when the normal
  1759. * callback queue is inoperable. If this is not a no-CBs CPU, this
  1760. * function returns failure back to __call_rcu(), which can complain
  1761. * appropriately.
  1762. *
  1763. * Otherwise, this function queues the callback where the corresponding
  1764. * "rcuo" kthread can find it.
  1765. */
  1766. static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp,
  1767. bool lazy, unsigned long flags)
  1768. {
  1769. if (!rcu_is_nocb_cpu(rdp->cpu))
  1770. return false;
  1771. __call_rcu_nocb_enqueue(rdp, rhp, &rhp->next, 1, lazy, flags);
  1772. if (__is_kfree_rcu_offset((unsigned long)rhp->func))
  1773. trace_rcu_kfree_callback(rdp->rsp->name, rhp,
  1774. (unsigned long)rhp->func,
  1775. -atomic_long_read(&rdp->nocb_q_count_lazy),
  1776. -atomic_long_read(&rdp->nocb_q_count));
  1777. else
  1778. trace_rcu_callback(rdp->rsp->name, rhp,
  1779. -atomic_long_read(&rdp->nocb_q_count_lazy),
  1780. -atomic_long_read(&rdp->nocb_q_count));
  1781. /*
  1782. * If called from an extended quiescent state with interrupts
  1783. * disabled, invoke the RCU core in order to allow the idle-entry
  1784. * deferred-wakeup check to function.
  1785. */
  1786. if (irqs_disabled_flags(flags) &&
  1787. !rcu_is_watching() &&
  1788. cpu_online(smp_processor_id()))
  1789. invoke_rcu_core();
  1790. return true;
  1791. }
  1792. /*
  1793. * Adopt orphaned callbacks on a no-CBs CPU, or return 0 if this is
  1794. * not a no-CBs CPU.
  1795. */
  1796. static bool __maybe_unused rcu_nocb_adopt_orphan_cbs(struct rcu_data *my_rdp,
  1797. struct rcu_data *rdp,
  1798. unsigned long flags)
  1799. {
  1800. RCU_LOCKDEP_WARN(!irqs_disabled(), "rcu_nocb_adopt_orphan_cbs() invoked with irqs enabled!!!");
  1801. if (!rcu_is_nocb_cpu(smp_processor_id()))
  1802. return false; /* Not NOCBs CPU, caller must migrate CBs. */
  1803. __call_rcu_nocb_enqueue(my_rdp, rcu_segcblist_head(&rdp->cblist),
  1804. rcu_segcblist_tail(&rdp->cblist),
  1805. rcu_segcblist_n_cbs(&rdp->cblist),
  1806. rcu_segcblist_n_lazy_cbs(&rdp->cblist), flags);
  1807. rcu_segcblist_init(&rdp->cblist);
  1808. rcu_segcblist_disable(&rdp->cblist);
  1809. return true;
  1810. }
  1811. /*
  1812. * If necessary, kick off a new grace period, and either way wait
  1813. * for a subsequent grace period to complete.
  1814. */
  1815. static void rcu_nocb_wait_gp(struct rcu_data *rdp)
  1816. {
  1817. unsigned long c;
  1818. bool d;
  1819. unsigned long flags;
  1820. bool needwake;
  1821. struct rcu_node *rnp = rdp->mynode;
  1822. raw_spin_lock_irqsave_rcu_node(rnp, flags);
  1823. needwake = rcu_start_future_gp(rnp, rdp, &c);
  1824. raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
  1825. if (needwake)
  1826. rcu_gp_kthread_wake(rdp->rsp);
  1827. /*
  1828. * Wait for the grace period. Do so interruptibly to avoid messing
  1829. * up the load average.
  1830. */
  1831. trace_rcu_future_gp(rnp, rdp, c, TPS("StartWait"));
  1832. for (;;) {
  1833. swait_event_interruptible(
  1834. rnp->nocb_gp_wq[c & 0x1],
  1835. (d = ULONG_CMP_GE(READ_ONCE(rnp->completed), c)));
  1836. if (likely(d))
  1837. break;
  1838. WARN_ON(signal_pending(current));
  1839. trace_rcu_future_gp(rnp, rdp, c, TPS("ResumeWait"));
  1840. }
  1841. trace_rcu_future_gp(rnp, rdp, c, TPS("EndWait"));
  1842. smp_mb(); /* Ensure that CB invocation happens after GP end. */
  1843. }
  1844. /*
  1845. * Leaders come here to wait for additional callbacks to show up.
  1846. * This function does not return until callbacks appear.
  1847. */
  1848. static void nocb_leader_wait(struct rcu_data *my_rdp)
  1849. {
  1850. bool firsttime = true;
  1851. unsigned long flags;
  1852. bool gotcbs;
  1853. struct rcu_data *rdp;
  1854. struct rcu_head **tail;
  1855. wait_again:
  1856. /* Wait for callbacks to appear. */
  1857. if (!rcu_nocb_poll) {
  1858. trace_rcu_nocb_wake(my_rdp->rsp->name, my_rdp->cpu, TPS("Sleep"));
  1859. swait_event_interruptible(my_rdp->nocb_wq,
  1860. !READ_ONCE(my_rdp->nocb_leader_sleep));
  1861. raw_spin_lock_irqsave(&my_rdp->nocb_lock, flags);
  1862. my_rdp->nocb_leader_sleep = true;
  1863. WRITE_ONCE(my_rdp->nocb_defer_wakeup, RCU_NOCB_WAKE_NOT);
  1864. del_timer(&my_rdp->nocb_timer);
  1865. raw_spin_unlock_irqrestore(&my_rdp->nocb_lock, flags);
  1866. } else if (firsttime) {
  1867. firsttime = false; /* Don't drown trace log with "Poll"! */
  1868. trace_rcu_nocb_wake(my_rdp->rsp->name, my_rdp->cpu, TPS("Poll"));
  1869. }
  1870. /*
  1871. * Each pass through the following loop checks a follower for CBs.
  1872. * We are our own first follower. Any CBs found are moved to
  1873. * nocb_gp_head, where they await a grace period.
  1874. */
  1875. gotcbs = false;
  1876. smp_mb(); /* wakeup and _sleep before ->nocb_head reads. */
  1877. for (rdp = my_rdp; rdp; rdp = rdp->nocb_next_follower) {
  1878. rdp->nocb_gp_head = READ_ONCE(rdp->nocb_head);
  1879. if (!rdp->nocb_gp_head)
  1880. continue; /* No CBs here, try next follower. */
  1881. /* Move callbacks to wait-for-GP list, which is empty. */
  1882. WRITE_ONCE(rdp->nocb_head, NULL);
  1883. rdp->nocb_gp_tail = xchg(&rdp->nocb_tail, &rdp->nocb_head);
  1884. gotcbs = true;
  1885. }
  1886. /* No callbacks? Sleep a bit if polling, and go retry. */
  1887. if (unlikely(!gotcbs)) {
  1888. WARN_ON(signal_pending(current));
  1889. if (rcu_nocb_poll) {
  1890. schedule_timeout_interruptible(1);
  1891. } else {
  1892. trace_rcu_nocb_wake(my_rdp->rsp->name, my_rdp->cpu,
  1893. TPS("WokeEmpty"));
  1894. }
  1895. goto wait_again;
  1896. }
  1897. /* Wait for one grace period. */
  1898. rcu_nocb_wait_gp(my_rdp);
  1899. /* Each pass through the following loop wakes a follower, if needed. */
  1900. for (rdp = my_rdp; rdp; rdp = rdp->nocb_next_follower) {
  1901. if (!rcu_nocb_poll &&
  1902. READ_ONCE(rdp->nocb_head) &&
  1903. READ_ONCE(my_rdp->nocb_leader_sleep)) {
  1904. raw_spin_lock_irqsave(&my_rdp->nocb_lock, flags);
  1905. my_rdp->nocb_leader_sleep = false;/* No need to sleep.*/
  1906. raw_spin_unlock_irqrestore(&my_rdp->nocb_lock, flags);
  1907. }
  1908. if (!rdp->nocb_gp_head)
  1909. continue; /* No CBs, so no need to wake follower. */
  1910. /* Append callbacks to follower's "done" list. */
  1911. raw_spin_lock_irqsave(&rdp->nocb_lock, flags);
  1912. tail = rdp->nocb_follower_tail;
  1913. rdp->nocb_follower_tail = rdp->nocb_gp_tail;
  1914. *tail = rdp->nocb_gp_head;
  1915. raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags);
  1916. if (rdp != my_rdp && tail == &rdp->nocb_follower_head) {
  1917. /* List was empty, so wake up the follower. */
  1918. swake_up(&rdp->nocb_wq);
  1919. }
  1920. }
  1921. /* If we (the leader) don't have CBs, go wait some more. */
  1922. if (!my_rdp->nocb_follower_head)
  1923. goto wait_again;
  1924. }
  1925. /*
  1926. * Followers come here to wait for additional callbacks to show up.
  1927. * This function does not return until callbacks appear.
  1928. */
  1929. static void nocb_follower_wait(struct rcu_data *rdp)
  1930. {
  1931. for (;;) {
  1932. trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, TPS("FollowerSleep"));
  1933. swait_event_interruptible(rdp->nocb_wq,
  1934. READ_ONCE(rdp->nocb_follower_head));
  1935. if (smp_load_acquire(&rdp->nocb_follower_head)) {
  1936. /* ^^^ Ensure CB invocation follows _head test. */
  1937. return;
  1938. }
  1939. WARN_ON(signal_pending(current));
  1940. trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, TPS("WokeEmpty"));
  1941. }
  1942. }
  1943. /*
  1944. * Per-rcu_data kthread, but only for no-CBs CPUs. Each kthread invokes
  1945. * callbacks queued by the corresponding no-CBs CPU, however, there is
  1946. * an optional leader-follower relationship so that the grace-period
  1947. * kthreads don't have to do quite so many wakeups.
  1948. */
  1949. static int rcu_nocb_kthread(void *arg)
  1950. {
  1951. int c, cl;
  1952. unsigned long flags;
  1953. struct rcu_head *list;
  1954. struct rcu_head *next;
  1955. struct rcu_head **tail;
  1956. struct rcu_data *rdp = arg;
  1957. /* Each pass through this loop invokes one batch of callbacks */
  1958. for (;;) {
  1959. /* Wait for callbacks. */
  1960. if (rdp->nocb_leader == rdp)
  1961. nocb_leader_wait(rdp);
  1962. else
  1963. nocb_follower_wait(rdp);
  1964. /* Pull the ready-to-invoke callbacks onto local list. */
  1965. raw_spin_lock_irqsave(&rdp->nocb_lock, flags);
  1966. list = rdp->nocb_follower_head;
  1967. rdp->nocb_follower_head = NULL;
  1968. tail = rdp->nocb_follower_tail;
  1969. rdp->nocb_follower_tail = &rdp->nocb_follower_head;
  1970. raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags);
  1971. BUG_ON(!list);
  1972. trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, TPS("WokeNonEmpty"));
  1973. /* Each pass through the following loop invokes a callback. */
  1974. trace_rcu_batch_start(rdp->rsp->name,
  1975. atomic_long_read(&rdp->nocb_q_count_lazy),
  1976. atomic_long_read(&rdp->nocb_q_count), -1);
  1977. c = cl = 0;
  1978. while (list) {
  1979. next = list->next;
  1980. /* Wait for enqueuing to complete, if needed. */
  1981. while (next == NULL && &list->next != tail) {
  1982. trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu,
  1983. TPS("WaitQueue"));
  1984. schedule_timeout_interruptible(1);
  1985. trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu,
  1986. TPS("WokeQueue"));
  1987. next = list->next;
  1988. }
  1989. debug_rcu_head_unqueue(list);
  1990. local_bh_disable();
  1991. if (__rcu_reclaim(rdp->rsp->name, list))
  1992. cl++;
  1993. c++;
  1994. local_bh_enable();
  1995. cond_resched_rcu_qs();
  1996. list = next;
  1997. }
  1998. trace_rcu_batch_end(rdp->rsp->name, c, !!list, 0, 0, 1);
  1999. smp_mb__before_atomic(); /* _add after CB invocation. */
  2000. atomic_long_add(-c, &rdp->nocb_q_count);
  2001. atomic_long_add(-cl, &rdp->nocb_q_count_lazy);
  2002. rdp->n_nocbs_invoked += c;
  2003. }
  2004. return 0;
  2005. }
  2006. /* Is a deferred wakeup of rcu_nocb_kthread() required? */
  2007. static int rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp)
  2008. {
  2009. return READ_ONCE(rdp->nocb_defer_wakeup);
  2010. }
  2011. /* Do a deferred wakeup of rcu_nocb_kthread(). */
  2012. static void do_nocb_deferred_wakeup_common(struct rcu_data *rdp)
  2013. {
  2014. unsigned long flags;
  2015. int ndw;
  2016. raw_spin_lock_irqsave(&rdp->nocb_lock, flags);
  2017. if (!rcu_nocb_need_deferred_wakeup(rdp)) {
  2018. raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags);
  2019. return;
  2020. }
  2021. ndw = READ_ONCE(rdp->nocb_defer_wakeup);
  2022. WRITE_ONCE(rdp->nocb_defer_wakeup, RCU_NOCB_WAKE_NOT);
  2023. __wake_nocb_leader(rdp, ndw == RCU_NOCB_WAKE_FORCE, flags);
  2024. trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, TPS("DeferredWake"));
  2025. }
  2026. /* Do a deferred wakeup of rcu_nocb_kthread() from a timer handler. */
  2027. static void do_nocb_deferred_wakeup_timer(unsigned long x)
  2028. {
  2029. do_nocb_deferred_wakeup_common((struct rcu_data *)x);
  2030. }
  2031. /*
  2032. * Do a deferred wakeup of rcu_nocb_kthread() from fastpath.
  2033. * This means we do an inexact common-case check. Note that if
  2034. * we miss, ->nocb_timer will eventually clean things up.
  2035. */
  2036. static void do_nocb_deferred_wakeup(struct rcu_data *rdp)
  2037. {
  2038. if (rcu_nocb_need_deferred_wakeup(rdp))
  2039. do_nocb_deferred_wakeup_common(rdp);
  2040. }
  2041. void __init rcu_init_nohz(void)
  2042. {
  2043. int cpu;
  2044. bool need_rcu_nocb_mask = true;
  2045. struct rcu_state *rsp;
  2046. #if defined(CONFIG_NO_HZ_FULL)
  2047. if (tick_nohz_full_running && cpumask_weight(tick_nohz_full_mask))
  2048. need_rcu_nocb_mask = true;
  2049. #endif /* #if defined(CONFIG_NO_HZ_FULL) */
  2050. if (!have_rcu_nocb_mask && need_rcu_nocb_mask) {
  2051. if (!zalloc_cpumask_var(&rcu_nocb_mask, GFP_KERNEL)) {
  2052. pr_info("rcu_nocb_mask allocation failed, callback offloading disabled.\n");
  2053. return;
  2054. }
  2055. have_rcu_nocb_mask = true;
  2056. }
  2057. if (!have_rcu_nocb_mask)
  2058. return;
  2059. #if defined(CONFIG_NO_HZ_FULL)
  2060. if (tick_nohz_full_running)
  2061. cpumask_or(rcu_nocb_mask, rcu_nocb_mask, tick_nohz_full_mask);
  2062. #endif /* #if defined(CONFIG_NO_HZ_FULL) */
  2063. if (!cpumask_subset(rcu_nocb_mask, cpu_possible_mask)) {
  2064. pr_info("\tNote: kernel parameter 'rcu_nocbs=' contains nonexistent CPUs.\n");
  2065. cpumask_and(rcu_nocb_mask, cpu_possible_mask,
  2066. rcu_nocb_mask);
  2067. }
  2068. pr_info("\tOffload RCU callbacks from CPUs: %*pbl.\n",
  2069. cpumask_pr_args(rcu_nocb_mask));
  2070. if (rcu_nocb_poll)
  2071. pr_info("\tPoll for callbacks from no-CBs CPUs.\n");
  2072. for_each_rcu_flavor(rsp) {
  2073. for_each_cpu(cpu, rcu_nocb_mask)
  2074. init_nocb_callback_list(per_cpu_ptr(rsp->rda, cpu));
  2075. rcu_organize_nocb_kthreads(rsp);
  2076. }
  2077. }
  2078. /* Initialize per-rcu_data variables for no-CBs CPUs. */
  2079. static void __init rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp)
  2080. {
  2081. rdp->nocb_tail = &rdp->nocb_head;
  2082. init_swait_queue_head(&rdp->nocb_wq);
  2083. rdp->nocb_follower_tail = &rdp->nocb_follower_head;
  2084. raw_spin_lock_init(&rdp->nocb_lock);
  2085. setup_timer(&rdp->nocb_timer, do_nocb_deferred_wakeup_timer,
  2086. (unsigned long)rdp);
  2087. }
  2088. /*
  2089. * If the specified CPU is a no-CBs CPU that does not already have its
  2090. * rcuo kthread for the specified RCU flavor, spawn it. If the CPUs are
  2091. * brought online out of order, this can require re-organizing the
  2092. * leader-follower relationships.
  2093. */
  2094. static void rcu_spawn_one_nocb_kthread(struct rcu_state *rsp, int cpu)
  2095. {
  2096. struct rcu_data *rdp;
  2097. struct rcu_data *rdp_last;
  2098. struct rcu_data *rdp_old_leader;
  2099. struct rcu_data *rdp_spawn = per_cpu_ptr(rsp->rda, cpu);
  2100. struct task_struct *t;
  2101. /*
  2102. * If this isn't a no-CBs CPU or if it already has an rcuo kthread,
  2103. * then nothing to do.
  2104. */
  2105. if (!rcu_is_nocb_cpu(cpu) || rdp_spawn->nocb_kthread)
  2106. return;
  2107. /* If we didn't spawn the leader first, reorganize! */
  2108. rdp_old_leader = rdp_spawn->nocb_leader;
  2109. if (rdp_old_leader != rdp_spawn && !rdp_old_leader->nocb_kthread) {
  2110. rdp_last = NULL;
  2111. rdp = rdp_old_leader;
  2112. do {
  2113. rdp->nocb_leader = rdp_spawn;
  2114. if (rdp_last && rdp != rdp_spawn)
  2115. rdp_last->nocb_next_follower = rdp;
  2116. if (rdp == rdp_spawn) {
  2117. rdp = rdp->nocb_next_follower;
  2118. } else {
  2119. rdp_last = rdp;
  2120. rdp = rdp->nocb_next_follower;
  2121. rdp_last->nocb_next_follower = NULL;
  2122. }
  2123. } while (rdp);
  2124. rdp_spawn->nocb_next_follower = rdp_old_leader;
  2125. }
  2126. /* Spawn the kthread for this CPU and RCU flavor. */
  2127. t = kthread_run(rcu_nocb_kthread, rdp_spawn,
  2128. "rcuo%c/%d", rsp->abbr, cpu);
  2129. BUG_ON(IS_ERR(t));
  2130. WRITE_ONCE(rdp_spawn->nocb_kthread, t);
  2131. }
  2132. /*
  2133. * If the specified CPU is a no-CBs CPU that does not already have its
  2134. * rcuo kthreads, spawn them.
  2135. */
  2136. static void rcu_spawn_all_nocb_kthreads(int cpu)
  2137. {
  2138. struct rcu_state *rsp;
  2139. if (rcu_scheduler_fully_active)
  2140. for_each_rcu_flavor(rsp)
  2141. rcu_spawn_one_nocb_kthread(rsp, cpu);
  2142. }
  2143. /*
  2144. * Once the scheduler is running, spawn rcuo kthreads for all online
  2145. * no-CBs CPUs. This assumes that the early_initcall()s happen before
  2146. * non-boot CPUs come online -- if this changes, we will need to add
  2147. * some mutual exclusion.
  2148. */
  2149. static void __init rcu_spawn_nocb_kthreads(void)
  2150. {
  2151. int cpu;
  2152. for_each_online_cpu(cpu)
  2153. rcu_spawn_all_nocb_kthreads(cpu);
  2154. }
  2155. /* How many follower CPU IDs per leader? Default of -1 for sqrt(nr_cpu_ids). */
  2156. static int rcu_nocb_leader_stride = -1;
  2157. module_param(rcu_nocb_leader_stride, int, 0444);
  2158. /*
  2159. * Initialize leader-follower relationships for all no-CBs CPU.
  2160. */
  2161. static void __init rcu_organize_nocb_kthreads(struct rcu_state *rsp)
  2162. {
  2163. int cpu;
  2164. int ls = rcu_nocb_leader_stride;
  2165. int nl = 0; /* Next leader. */
  2166. struct rcu_data *rdp;
  2167. struct rcu_data *rdp_leader = NULL; /* Suppress misguided gcc warn. */
  2168. struct rcu_data *rdp_prev = NULL;
  2169. if (!have_rcu_nocb_mask)
  2170. return;
  2171. if (ls == -1) {
  2172. ls = int_sqrt(nr_cpu_ids);
  2173. rcu_nocb_leader_stride = ls;
  2174. }
  2175. /*
  2176. * Each pass through this loop sets up one rcu_data structure.
  2177. * Should the corresponding CPU come online in the future, then
  2178. * we will spawn the needed set of rcu_nocb_kthread() kthreads.
  2179. */
  2180. for_each_cpu(cpu, rcu_nocb_mask) {
  2181. rdp = per_cpu_ptr(rsp->rda, cpu);
  2182. if (rdp->cpu >= nl) {
  2183. /* New leader, set up for followers & next leader. */
  2184. nl = DIV_ROUND_UP(rdp->cpu + 1, ls) * ls;
  2185. rdp->nocb_leader = rdp;
  2186. rdp_leader = rdp;
  2187. } else {
  2188. /* Another follower, link to previous leader. */
  2189. rdp->nocb_leader = rdp_leader;
  2190. rdp_prev->nocb_next_follower = rdp;
  2191. }
  2192. rdp_prev = rdp;
  2193. }
  2194. }
  2195. /* Prevent __call_rcu() from enqueuing callbacks on no-CBs CPUs */
  2196. static bool init_nocb_callback_list(struct rcu_data *rdp)
  2197. {
  2198. if (!rcu_is_nocb_cpu(rdp->cpu))
  2199. return false;
  2200. /* If there are early-boot callbacks, move them to nocb lists. */
  2201. if (!rcu_segcblist_empty(&rdp->cblist)) {
  2202. rdp->nocb_head = rcu_segcblist_head(&rdp->cblist);
  2203. rdp->nocb_tail = rcu_segcblist_tail(&rdp->cblist);
  2204. atomic_long_set(&rdp->nocb_q_count,
  2205. rcu_segcblist_n_cbs(&rdp->cblist));
  2206. atomic_long_set(&rdp->nocb_q_count_lazy,
  2207. rcu_segcblist_n_lazy_cbs(&rdp->cblist));
  2208. rcu_segcblist_init(&rdp->cblist);
  2209. }
  2210. rcu_segcblist_disable(&rdp->cblist);
  2211. return true;
  2212. }
  2213. #else /* #ifdef CONFIG_RCU_NOCB_CPU */
  2214. static bool rcu_nocb_cpu_needs_barrier(struct rcu_state *rsp, int cpu)
  2215. {
  2216. WARN_ON_ONCE(1); /* Should be dead code. */
  2217. return false;
  2218. }
  2219. static void rcu_nocb_gp_cleanup(struct swait_queue_head *sq)
  2220. {
  2221. }
  2222. static void rcu_nocb_gp_set(struct rcu_node *rnp, int nrq)
  2223. {
  2224. }
  2225. static struct swait_queue_head *rcu_nocb_gp_get(struct rcu_node *rnp)
  2226. {
  2227. return NULL;
  2228. }
  2229. static void rcu_init_one_nocb(struct rcu_node *rnp)
  2230. {
  2231. }
  2232. static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp,
  2233. bool lazy, unsigned long flags)
  2234. {
  2235. return false;
  2236. }
  2237. static bool __maybe_unused rcu_nocb_adopt_orphan_cbs(struct rcu_data *my_rdp,
  2238. struct rcu_data *rdp,
  2239. unsigned long flags)
  2240. {
  2241. return false;
  2242. }
  2243. static void __init rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp)
  2244. {
  2245. }
  2246. static int rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp)
  2247. {
  2248. return false;
  2249. }
  2250. static void do_nocb_deferred_wakeup(struct rcu_data *rdp)
  2251. {
  2252. }
  2253. static void rcu_spawn_all_nocb_kthreads(int cpu)
  2254. {
  2255. }
  2256. static void __init rcu_spawn_nocb_kthreads(void)
  2257. {
  2258. }
  2259. static bool init_nocb_callback_list(struct rcu_data *rdp)
  2260. {
  2261. return false;
  2262. }
  2263. #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */
  2264. /*
  2265. * An adaptive-ticks CPU can potentially execute in kernel mode for an
  2266. * arbitrarily long period of time with the scheduling-clock tick turned
  2267. * off. RCU will be paying attention to this CPU because it is in the
  2268. * kernel, but the CPU cannot be guaranteed to be executing the RCU state
  2269. * machine because the scheduling-clock tick has been disabled. Therefore,
  2270. * if an adaptive-ticks CPU is failing to respond to the current grace
  2271. * period and has not be idle from an RCU perspective, kick it.
  2272. */
  2273. static void __maybe_unused rcu_kick_nohz_cpu(int cpu)
  2274. {
  2275. #ifdef CONFIG_NO_HZ_FULL
  2276. if (tick_nohz_full_cpu(cpu))
  2277. smp_send_reschedule(cpu);
  2278. #endif /* #ifdef CONFIG_NO_HZ_FULL */
  2279. }
  2280. /*
  2281. * Is this CPU a NO_HZ_FULL CPU that should ignore RCU so that the
  2282. * grace-period kthread will do force_quiescent_state() processing?
  2283. * The idea is to avoid waking up RCU core processing on such a
  2284. * CPU unless the grace period has extended for too long.
  2285. *
  2286. * This code relies on the fact that all NO_HZ_FULL CPUs are also
  2287. * CONFIG_RCU_NOCB_CPU CPUs.
  2288. */
  2289. static bool rcu_nohz_full_cpu(struct rcu_state *rsp)
  2290. {
  2291. #ifdef CONFIG_NO_HZ_FULL
  2292. if (tick_nohz_full_cpu(smp_processor_id()) &&
  2293. (!rcu_gp_in_progress(rsp) ||
  2294. ULONG_CMP_LT(jiffies, READ_ONCE(rsp->gp_start) + HZ)))
  2295. return true;
  2296. #endif /* #ifdef CONFIG_NO_HZ_FULL */
  2297. return false;
  2298. }
  2299. /*
  2300. * Bind the grace-period kthread for the sysidle flavor of RCU to the
  2301. * timekeeping CPU.
  2302. */
  2303. static void rcu_bind_gp_kthread(void)
  2304. {
  2305. int __maybe_unused cpu;
  2306. if (!tick_nohz_full_enabled())
  2307. return;
  2308. housekeeping_affine(current);
  2309. }
  2310. /* Record the current task on dyntick-idle entry. */
  2311. static void rcu_dynticks_task_enter(void)
  2312. {
  2313. #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL)
  2314. WRITE_ONCE(current->rcu_tasks_idle_cpu, smp_processor_id());
  2315. #endif /* #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) */
  2316. }
  2317. /* Record no current task on dyntick-idle exit. */
  2318. static void rcu_dynticks_task_exit(void)
  2319. {
  2320. #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL)
  2321. WRITE_ONCE(current->rcu_tasks_idle_cpu, -1);
  2322. #endif /* #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) */
  2323. }