vector.c 25 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955
  1. /*
  2. * Local APIC related interfaces to support IOAPIC, MSI, HT_IRQ etc.
  3. *
  4. * Copyright (C) 1997, 1998, 1999, 2000, 2009 Ingo Molnar, Hajnalka Szabo
  5. * Moved from arch/x86/kernel/apic/io_apic.c.
  6. * Jiang Liu <jiang.liu@linux.intel.com>
  7. * Enable support of hierarchical irqdomains
  8. *
  9. * This program is free software; you can redistribute it and/or modify
  10. * it under the terms of the GNU General Public License version 2 as
  11. * published by the Free Software Foundation.
  12. */
  13. #include <linux/interrupt.h>
  14. #include <linux/init.h>
  15. #include <linux/compiler.h>
  16. #include <linux/slab.h>
  17. #include <asm/irqdomain.h>
  18. #include <asm/hw_irq.h>
  19. #include <asm/apic.h>
  20. #include <asm/i8259.h>
  21. #include <asm/desc.h>
  22. #include <asm/irq_remapping.h>
  23. struct apic_chip_data {
  24. struct irq_cfg cfg;
  25. cpumask_var_t domain;
  26. cpumask_var_t old_domain;
  27. u8 move_in_progress : 1;
  28. };
  29. struct irq_domain *x86_vector_domain;
  30. EXPORT_SYMBOL_GPL(x86_vector_domain);
  31. static DEFINE_RAW_SPINLOCK(vector_lock);
  32. static cpumask_var_t vector_cpumask, vector_searchmask, searched_cpumask;
  33. static struct irq_chip lapic_controller;
  34. #ifdef CONFIG_X86_IO_APIC
  35. static struct apic_chip_data *legacy_irq_data[NR_IRQS_LEGACY];
  36. #endif
  37. void lock_vector_lock(void)
  38. {
  39. /* Used to the online set of cpus does not change
  40. * during assign_irq_vector.
  41. */
  42. raw_spin_lock(&vector_lock);
  43. }
  44. void unlock_vector_lock(void)
  45. {
  46. raw_spin_unlock(&vector_lock);
  47. }
  48. static struct apic_chip_data *apic_chip_data(struct irq_data *irq_data)
  49. {
  50. if (!irq_data)
  51. return NULL;
  52. while (irq_data->parent_data)
  53. irq_data = irq_data->parent_data;
  54. return irq_data->chip_data;
  55. }
  56. struct irq_cfg *irqd_cfg(struct irq_data *irq_data)
  57. {
  58. struct apic_chip_data *data = apic_chip_data(irq_data);
  59. return data ? &data->cfg : NULL;
  60. }
  61. EXPORT_SYMBOL_GPL(irqd_cfg);
  62. struct irq_cfg *irq_cfg(unsigned int irq)
  63. {
  64. return irqd_cfg(irq_get_irq_data(irq));
  65. }
  66. static struct apic_chip_data *alloc_apic_chip_data(int node)
  67. {
  68. struct apic_chip_data *data;
  69. data = kzalloc_node(sizeof(*data), GFP_KERNEL, node);
  70. if (!data)
  71. return NULL;
  72. if (!zalloc_cpumask_var_node(&data->domain, GFP_KERNEL, node))
  73. goto out_data;
  74. if (!zalloc_cpumask_var_node(&data->old_domain, GFP_KERNEL, node))
  75. goto out_domain;
  76. return data;
  77. out_domain:
  78. free_cpumask_var(data->domain);
  79. out_data:
  80. kfree(data);
  81. return NULL;
  82. }
  83. static void free_apic_chip_data(struct apic_chip_data *data)
  84. {
  85. if (data) {
  86. free_cpumask_var(data->domain);
  87. free_cpumask_var(data->old_domain);
  88. kfree(data);
  89. }
  90. }
  91. static int __assign_irq_vector(int irq, struct apic_chip_data *d,
  92. const struct cpumask *mask)
  93. {
  94. /*
  95. * NOTE! The local APIC isn't very good at handling
  96. * multiple interrupts at the same interrupt level.
  97. * As the interrupt level is determined by taking the
  98. * vector number and shifting that right by 4, we
  99. * want to spread these out a bit so that they don't
  100. * all fall in the same interrupt level.
  101. *
  102. * Also, we've got to be careful not to trash gate
  103. * 0x80, because int 0x80 is hm, kind of importantish. ;)
  104. */
  105. static int current_vector = FIRST_EXTERNAL_VECTOR + VECTOR_OFFSET_START;
  106. static int current_offset = VECTOR_OFFSET_START % 16;
  107. int cpu, vector;
  108. /*
  109. * If there is still a move in progress or the previous move has not
  110. * been cleaned up completely, tell the caller to come back later.
  111. */
  112. if (d->move_in_progress ||
  113. cpumask_intersects(d->old_domain, cpu_online_mask))
  114. return -EBUSY;
  115. /* Only try and allocate irqs on cpus that are present */
  116. cpumask_clear(d->old_domain);
  117. cpumask_clear(searched_cpumask);
  118. cpu = cpumask_first_and(mask, cpu_online_mask);
  119. while (cpu < nr_cpu_ids) {
  120. int new_cpu, offset;
  121. /* Get the possible target cpus for @mask/@cpu from the apic */
  122. apic->vector_allocation_domain(cpu, vector_cpumask, mask);
  123. /*
  124. * Clear the offline cpus from @vector_cpumask for searching
  125. * and verify whether the result overlaps with @mask. If true,
  126. * then the call to apic->cpu_mask_to_apicid_and() will
  127. * succeed as well. If not, no point in trying to find a
  128. * vector in this mask.
  129. */
  130. cpumask_and(vector_searchmask, vector_cpumask, cpu_online_mask);
  131. if (!cpumask_intersects(vector_searchmask, mask))
  132. goto next_cpu;
  133. if (cpumask_subset(vector_cpumask, d->domain)) {
  134. if (cpumask_equal(vector_cpumask, d->domain))
  135. goto success;
  136. /*
  137. * Mark the cpus which are not longer in the mask for
  138. * cleanup.
  139. */
  140. cpumask_andnot(d->old_domain, d->domain, vector_cpumask);
  141. vector = d->cfg.vector;
  142. goto update;
  143. }
  144. vector = current_vector;
  145. offset = current_offset;
  146. next:
  147. vector += 16;
  148. if (vector >= first_system_vector) {
  149. offset = (offset + 1) % 16;
  150. vector = FIRST_EXTERNAL_VECTOR + offset;
  151. }
  152. /* If the search wrapped around, try the next cpu */
  153. if (unlikely(current_vector == vector))
  154. goto next_cpu;
  155. if (test_bit(vector, used_vectors))
  156. goto next;
  157. for_each_cpu(new_cpu, vector_searchmask) {
  158. if (!IS_ERR_OR_NULL(per_cpu(vector_irq, new_cpu)[vector]))
  159. goto next;
  160. }
  161. /* Found one! */
  162. current_vector = vector;
  163. current_offset = offset;
  164. /* Schedule the old vector for cleanup on all cpus */
  165. if (d->cfg.vector)
  166. cpumask_copy(d->old_domain, d->domain);
  167. for_each_cpu(new_cpu, vector_searchmask)
  168. per_cpu(vector_irq, new_cpu)[vector] = irq_to_desc(irq);
  169. goto update;
  170. next_cpu:
  171. /*
  172. * We exclude the current @vector_cpumask from the requested
  173. * @mask and try again with the next online cpu in the
  174. * result. We cannot modify @mask, so we use @vector_cpumask
  175. * as a temporary buffer here as it will be reassigned when
  176. * calling apic->vector_allocation_domain() above.
  177. */
  178. cpumask_or(searched_cpumask, searched_cpumask, vector_cpumask);
  179. cpumask_andnot(vector_cpumask, mask, searched_cpumask);
  180. cpu = cpumask_first_and(vector_cpumask, cpu_online_mask);
  181. continue;
  182. }
  183. return -ENOSPC;
  184. update:
  185. /*
  186. * Exclude offline cpus from the cleanup mask and set the
  187. * move_in_progress flag when the result is not empty.
  188. */
  189. cpumask_and(d->old_domain, d->old_domain, cpu_online_mask);
  190. d->move_in_progress = !cpumask_empty(d->old_domain);
  191. d->cfg.old_vector = d->move_in_progress ? d->cfg.vector : 0;
  192. d->cfg.vector = vector;
  193. cpumask_copy(d->domain, vector_cpumask);
  194. success:
  195. /*
  196. * Cache destination APIC IDs into cfg->dest_apicid. This cannot fail
  197. * as we already established, that mask & d->domain & cpu_online_mask
  198. * is not empty.
  199. */
  200. BUG_ON(apic->cpu_mask_to_apicid_and(mask, d->domain,
  201. &d->cfg.dest_apicid));
  202. return 0;
  203. }
  204. static int assign_irq_vector(int irq, struct apic_chip_data *data,
  205. const struct cpumask *mask)
  206. {
  207. int err;
  208. unsigned long flags;
  209. raw_spin_lock_irqsave(&vector_lock, flags);
  210. err = __assign_irq_vector(irq, data, mask);
  211. raw_spin_unlock_irqrestore(&vector_lock, flags);
  212. return err;
  213. }
  214. static int assign_irq_vector_policy(int irq, int node,
  215. struct apic_chip_data *data,
  216. struct irq_alloc_info *info)
  217. {
  218. if (info && info->mask)
  219. return assign_irq_vector(irq, data, info->mask);
  220. if (node != NUMA_NO_NODE &&
  221. assign_irq_vector(irq, data, cpumask_of_node(node)) == 0)
  222. return 0;
  223. return assign_irq_vector(irq, data, apic->target_cpus());
  224. }
  225. static void clear_irq_vector(int irq, struct apic_chip_data *data)
  226. {
  227. struct irq_desc *desc;
  228. int cpu, vector;
  229. BUG_ON(!data->cfg.vector);
  230. vector = data->cfg.vector;
  231. for_each_cpu_and(cpu, data->domain, cpu_online_mask)
  232. per_cpu(vector_irq, cpu)[vector] = VECTOR_UNUSED;
  233. data->cfg.vector = 0;
  234. cpumask_clear(data->domain);
  235. /*
  236. * If move is in progress or the old_domain mask is not empty,
  237. * i.e. the cleanup IPI has not been processed yet, we need to remove
  238. * the old references to desc from all cpus vector tables.
  239. */
  240. if (!data->move_in_progress && cpumask_empty(data->old_domain))
  241. return;
  242. desc = irq_to_desc(irq);
  243. for_each_cpu_and(cpu, data->old_domain, cpu_online_mask) {
  244. for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS;
  245. vector++) {
  246. if (per_cpu(vector_irq, cpu)[vector] != desc)
  247. continue;
  248. per_cpu(vector_irq, cpu)[vector] = VECTOR_UNUSED;
  249. break;
  250. }
  251. }
  252. data->move_in_progress = 0;
  253. }
  254. void init_irq_alloc_info(struct irq_alloc_info *info,
  255. const struct cpumask *mask)
  256. {
  257. memset(info, 0, sizeof(*info));
  258. info->mask = mask;
  259. }
  260. void copy_irq_alloc_info(struct irq_alloc_info *dst, struct irq_alloc_info *src)
  261. {
  262. if (src)
  263. *dst = *src;
  264. else
  265. memset(dst, 0, sizeof(*dst));
  266. }
  267. static void x86_vector_free_irqs(struct irq_domain *domain,
  268. unsigned int virq, unsigned int nr_irqs)
  269. {
  270. struct apic_chip_data *apic_data;
  271. struct irq_data *irq_data;
  272. unsigned long flags;
  273. int i;
  274. for (i = 0; i < nr_irqs; i++) {
  275. irq_data = irq_domain_get_irq_data(x86_vector_domain, virq + i);
  276. if (irq_data && irq_data->chip_data) {
  277. raw_spin_lock_irqsave(&vector_lock, flags);
  278. clear_irq_vector(virq + i, irq_data->chip_data);
  279. apic_data = irq_data->chip_data;
  280. irq_domain_reset_irq_data(irq_data);
  281. raw_spin_unlock_irqrestore(&vector_lock, flags);
  282. free_apic_chip_data(apic_data);
  283. #ifdef CONFIG_X86_IO_APIC
  284. if (virq + i < nr_legacy_irqs())
  285. legacy_irq_data[virq + i] = NULL;
  286. #endif
  287. }
  288. }
  289. }
  290. static int x86_vector_alloc_irqs(struct irq_domain *domain, unsigned int virq,
  291. unsigned int nr_irqs, void *arg)
  292. {
  293. struct irq_alloc_info *info = arg;
  294. struct apic_chip_data *data;
  295. struct irq_data *irq_data;
  296. int i, err, node;
  297. if (disable_apic)
  298. return -ENXIO;
  299. /* Currently vector allocator can't guarantee contiguous allocations */
  300. if ((info->flags & X86_IRQ_ALLOC_CONTIGUOUS_VECTORS) && nr_irqs > 1)
  301. return -ENOSYS;
  302. for (i = 0; i < nr_irqs; i++) {
  303. irq_data = irq_domain_get_irq_data(domain, virq + i);
  304. BUG_ON(!irq_data);
  305. node = irq_data_get_node(irq_data);
  306. #ifdef CONFIG_X86_IO_APIC
  307. if (virq + i < nr_legacy_irqs() && legacy_irq_data[virq + i])
  308. data = legacy_irq_data[virq + i];
  309. else
  310. #endif
  311. data = alloc_apic_chip_data(node);
  312. if (!data) {
  313. err = -ENOMEM;
  314. goto error;
  315. }
  316. irq_data->chip = &lapic_controller;
  317. irq_data->chip_data = data;
  318. irq_data->hwirq = virq + i;
  319. err = assign_irq_vector_policy(virq + i, node, data, info);
  320. if (err)
  321. goto error;
  322. }
  323. return 0;
  324. error:
  325. x86_vector_free_irqs(domain, virq, i + 1);
  326. return err;
  327. }
  328. static const struct irq_domain_ops x86_vector_domain_ops = {
  329. .alloc = x86_vector_alloc_irqs,
  330. .free = x86_vector_free_irqs,
  331. };
  332. int __init arch_probe_nr_irqs(void)
  333. {
  334. int nr;
  335. if (nr_irqs > (NR_VECTORS * nr_cpu_ids))
  336. nr_irqs = NR_VECTORS * nr_cpu_ids;
  337. nr = (gsi_top + nr_legacy_irqs()) + 8 * nr_cpu_ids;
  338. #if defined(CONFIG_PCI_MSI) || defined(CONFIG_HT_IRQ)
  339. /*
  340. * for MSI and HT dyn irq
  341. */
  342. if (gsi_top <= NR_IRQS_LEGACY)
  343. nr += 8 * nr_cpu_ids;
  344. else
  345. nr += gsi_top * 16;
  346. #endif
  347. if (nr < nr_irqs)
  348. nr_irqs = nr;
  349. /*
  350. * We don't know if PIC is present at this point so we need to do
  351. * probe() to get the right number of legacy IRQs.
  352. */
  353. return legacy_pic->probe();
  354. }
  355. #ifdef CONFIG_X86_IO_APIC
  356. static void init_legacy_irqs(void)
  357. {
  358. int i, node = cpu_to_node(0);
  359. struct apic_chip_data *data;
  360. /*
  361. * For legacy IRQ's, start with assigning irq0 to irq15 to
  362. * ISA_IRQ_VECTOR(i) for all cpu's.
  363. */
  364. for (i = 0; i < nr_legacy_irqs(); i++) {
  365. data = legacy_irq_data[i] = alloc_apic_chip_data(node);
  366. BUG_ON(!data);
  367. data->cfg.vector = ISA_IRQ_VECTOR(i);
  368. cpumask_setall(data->domain);
  369. irq_set_chip_data(i, data);
  370. }
  371. }
  372. #else
  373. static void init_legacy_irqs(void) { }
  374. #endif
  375. int __init arch_early_irq_init(void)
  376. {
  377. init_legacy_irqs();
  378. x86_vector_domain = irq_domain_add_tree(NULL, &x86_vector_domain_ops,
  379. NULL);
  380. BUG_ON(x86_vector_domain == NULL);
  381. irq_set_default_host(x86_vector_domain);
  382. arch_init_msi_domain(x86_vector_domain);
  383. arch_init_htirq_domain(x86_vector_domain);
  384. BUG_ON(!alloc_cpumask_var(&vector_cpumask, GFP_KERNEL));
  385. BUG_ON(!alloc_cpumask_var(&vector_searchmask, GFP_KERNEL));
  386. BUG_ON(!alloc_cpumask_var(&searched_cpumask, GFP_KERNEL));
  387. return arch_early_ioapic_init();
  388. }
  389. /* Initialize vector_irq on a new cpu */
  390. static void __setup_vector_irq(int cpu)
  391. {
  392. struct apic_chip_data *data;
  393. struct irq_desc *desc;
  394. int irq, vector;
  395. /* Mark the inuse vectors */
  396. for_each_irq_desc(irq, desc) {
  397. struct irq_data *idata = irq_desc_get_irq_data(desc);
  398. data = apic_chip_data(idata);
  399. if (!data || !cpumask_test_cpu(cpu, data->domain))
  400. continue;
  401. vector = data->cfg.vector;
  402. per_cpu(vector_irq, cpu)[vector] = desc;
  403. }
  404. /* Mark the free vectors */
  405. for (vector = 0; vector < NR_VECTORS; ++vector) {
  406. desc = per_cpu(vector_irq, cpu)[vector];
  407. if (IS_ERR_OR_NULL(desc))
  408. continue;
  409. data = apic_chip_data(irq_desc_get_irq_data(desc));
  410. if (!cpumask_test_cpu(cpu, data->domain))
  411. per_cpu(vector_irq, cpu)[vector] = VECTOR_UNUSED;
  412. }
  413. }
  414. /*
  415. * Setup the vector to irq mappings. Must be called with vector_lock held.
  416. */
  417. void setup_vector_irq(int cpu)
  418. {
  419. int irq;
  420. lockdep_assert_held(&vector_lock);
  421. /*
  422. * On most of the platforms, legacy PIC delivers the interrupts on the
  423. * boot cpu. But there are certain platforms where PIC interrupts are
  424. * delivered to multiple cpu's. If the legacy IRQ is handled by the
  425. * legacy PIC, for the new cpu that is coming online, setup the static
  426. * legacy vector to irq mapping:
  427. */
  428. for (irq = 0; irq < nr_legacy_irqs(); irq++)
  429. per_cpu(vector_irq, cpu)[ISA_IRQ_VECTOR(irq)] = irq_to_desc(irq);
  430. __setup_vector_irq(cpu);
  431. }
  432. static int apic_retrigger_irq(struct irq_data *irq_data)
  433. {
  434. struct apic_chip_data *data = apic_chip_data(irq_data);
  435. unsigned long flags;
  436. int cpu;
  437. raw_spin_lock_irqsave(&vector_lock, flags);
  438. cpu = cpumask_first_and(data->domain, cpu_online_mask);
  439. apic->send_IPI_mask(cpumask_of(cpu), data->cfg.vector);
  440. raw_spin_unlock_irqrestore(&vector_lock, flags);
  441. return 1;
  442. }
  443. void apic_ack_edge(struct irq_data *data)
  444. {
  445. irq_complete_move(irqd_cfg(data));
  446. irq_move_irq(data);
  447. ack_APIC_irq();
  448. }
  449. static int apic_set_affinity(struct irq_data *irq_data,
  450. const struct cpumask *dest, bool force)
  451. {
  452. struct apic_chip_data *data = irq_data->chip_data;
  453. int err, irq = irq_data->irq;
  454. if (!config_enabled(CONFIG_SMP))
  455. return -EPERM;
  456. if (!cpumask_intersects(dest, cpu_online_mask))
  457. return -EINVAL;
  458. err = assign_irq_vector(irq, data, dest);
  459. return err ? err : IRQ_SET_MASK_OK;
  460. }
  461. static struct irq_chip lapic_controller = {
  462. .irq_ack = apic_ack_edge,
  463. .irq_set_affinity = apic_set_affinity,
  464. .irq_retrigger = apic_retrigger_irq,
  465. };
  466. #ifdef CONFIG_SMP
  467. static void __send_cleanup_vector(struct apic_chip_data *data)
  468. {
  469. raw_spin_lock(&vector_lock);
  470. cpumask_and(data->old_domain, data->old_domain, cpu_online_mask);
  471. data->move_in_progress = 0;
  472. if (!cpumask_empty(data->old_domain))
  473. apic->send_IPI_mask(data->old_domain, IRQ_MOVE_CLEANUP_VECTOR);
  474. raw_spin_unlock(&vector_lock);
  475. }
  476. void send_cleanup_vector(struct irq_cfg *cfg)
  477. {
  478. struct apic_chip_data *data;
  479. data = container_of(cfg, struct apic_chip_data, cfg);
  480. if (data->move_in_progress)
  481. __send_cleanup_vector(data);
  482. }
  483. asmlinkage __visible void smp_irq_move_cleanup_interrupt(void)
  484. {
  485. unsigned vector, me;
  486. entering_ack_irq();
  487. /* Prevent vectors vanishing under us */
  488. raw_spin_lock(&vector_lock);
  489. me = smp_processor_id();
  490. for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
  491. struct apic_chip_data *data;
  492. struct irq_desc *desc;
  493. unsigned int irr;
  494. retry:
  495. desc = __this_cpu_read(vector_irq[vector]);
  496. if (IS_ERR_OR_NULL(desc))
  497. continue;
  498. if (!raw_spin_trylock(&desc->lock)) {
  499. raw_spin_unlock(&vector_lock);
  500. cpu_relax();
  501. raw_spin_lock(&vector_lock);
  502. goto retry;
  503. }
  504. data = apic_chip_data(irq_desc_get_irq_data(desc));
  505. if (!data)
  506. goto unlock;
  507. /*
  508. * Nothing to cleanup if irq migration is in progress
  509. * or this cpu is not set in the cleanup mask.
  510. */
  511. if (data->move_in_progress ||
  512. !cpumask_test_cpu(me, data->old_domain))
  513. goto unlock;
  514. /*
  515. * We have two cases to handle here:
  516. * 1) vector is unchanged but the target mask got reduced
  517. * 2) vector and the target mask has changed
  518. *
  519. * #1 is obvious, but in #2 we have two vectors with the same
  520. * irq descriptor: the old and the new vector. So we need to
  521. * make sure that we only cleanup the old vector. The new
  522. * vector has the current @vector number in the config and
  523. * this cpu is part of the target mask. We better leave that
  524. * one alone.
  525. */
  526. if (vector == data->cfg.vector &&
  527. cpumask_test_cpu(me, data->domain))
  528. goto unlock;
  529. irr = apic_read(APIC_IRR + (vector / 32 * 0x10));
  530. /*
  531. * Check if the vector that needs to be cleanedup is
  532. * registered at the cpu's IRR. If so, then this is not
  533. * the best time to clean it up. Lets clean it up in the
  534. * next attempt by sending another IRQ_MOVE_CLEANUP_VECTOR
  535. * to myself.
  536. */
  537. if (irr & (1 << (vector % 32))) {
  538. apic->send_IPI_self(IRQ_MOVE_CLEANUP_VECTOR);
  539. goto unlock;
  540. }
  541. __this_cpu_write(vector_irq[vector], VECTOR_UNUSED);
  542. cpumask_clear_cpu(me, data->old_domain);
  543. unlock:
  544. raw_spin_unlock(&desc->lock);
  545. }
  546. raw_spin_unlock(&vector_lock);
  547. exiting_irq();
  548. }
  549. static void __irq_complete_move(struct irq_cfg *cfg, unsigned vector)
  550. {
  551. unsigned me;
  552. struct apic_chip_data *data;
  553. data = container_of(cfg, struct apic_chip_data, cfg);
  554. if (likely(!data->move_in_progress))
  555. return;
  556. me = smp_processor_id();
  557. if (vector == data->cfg.vector && cpumask_test_cpu(me, data->domain))
  558. __send_cleanup_vector(data);
  559. }
  560. void irq_complete_move(struct irq_cfg *cfg)
  561. {
  562. __irq_complete_move(cfg, ~get_irq_regs()->orig_ax);
  563. }
  564. /*
  565. * Called from fixup_irqs() with @desc->lock held and interrupts disabled.
  566. */
  567. void irq_force_complete_move(struct irq_desc *desc)
  568. {
  569. struct irq_data *irqdata = irq_desc_get_irq_data(desc);
  570. struct apic_chip_data *data = apic_chip_data(irqdata);
  571. struct irq_cfg *cfg = data ? &data->cfg : NULL;
  572. unsigned int cpu;
  573. if (!cfg)
  574. return;
  575. /*
  576. * This is tricky. If the cleanup of @data->old_domain has not been
  577. * done yet, then the following setaffinity call will fail with
  578. * -EBUSY. This can leave the interrupt in a stale state.
  579. *
  580. * All CPUs are stuck in stop machine with interrupts disabled so
  581. * calling __irq_complete_move() would be completely pointless.
  582. */
  583. raw_spin_lock(&vector_lock);
  584. /*
  585. * Clean out all offline cpus (including the outgoing one) from the
  586. * old_domain mask.
  587. */
  588. cpumask_and(data->old_domain, data->old_domain, cpu_online_mask);
  589. /*
  590. * If move_in_progress is cleared and the old_domain mask is empty,
  591. * then there is nothing to cleanup. fixup_irqs() will take care of
  592. * the stale vectors on the outgoing cpu.
  593. */
  594. if (!data->move_in_progress && cpumask_empty(data->old_domain)) {
  595. raw_spin_unlock(&vector_lock);
  596. return;
  597. }
  598. /*
  599. * 1) The interrupt is in move_in_progress state. That means that we
  600. * have not seen an interrupt since the io_apic was reprogrammed to
  601. * the new vector.
  602. *
  603. * 2) The interrupt has fired on the new vector, but the cleanup IPIs
  604. * have not been processed yet.
  605. */
  606. if (data->move_in_progress) {
  607. /*
  608. * In theory there is a race:
  609. *
  610. * set_ioapic(new_vector) <-- Interrupt is raised before update
  611. * is effective, i.e. it's raised on
  612. * the old vector.
  613. *
  614. * So if the target cpu cannot handle that interrupt before
  615. * the old vector is cleaned up, we get a spurious interrupt
  616. * and in the worst case the ioapic irq line becomes stale.
  617. *
  618. * But in case of cpu hotplug this should be a non issue
  619. * because if the affinity update happens right before all
  620. * cpus rendevouz in stop machine, there is no way that the
  621. * interrupt can be blocked on the target cpu because all cpus
  622. * loops first with interrupts enabled in stop machine, so the
  623. * old vector is not yet cleaned up when the interrupt fires.
  624. *
  625. * So the only way to run into this issue is if the delivery
  626. * of the interrupt on the apic/system bus would be delayed
  627. * beyond the point where the target cpu disables interrupts
  628. * in stop machine. I doubt that it can happen, but at least
  629. * there is a theroretical chance. Virtualization might be
  630. * able to expose this, but AFAICT the IOAPIC emulation is not
  631. * as stupid as the real hardware.
  632. *
  633. * Anyway, there is nothing we can do about that at this point
  634. * w/o refactoring the whole fixup_irq() business completely.
  635. * We print at least the irq number and the old vector number,
  636. * so we have the necessary information when a problem in that
  637. * area arises.
  638. */
  639. pr_warn("IRQ fixup: irq %d move in progress, old vector %d\n",
  640. irqdata->irq, cfg->old_vector);
  641. }
  642. /*
  643. * If old_domain is not empty, then other cpus still have the irq
  644. * descriptor set in their vector array. Clean it up.
  645. */
  646. for_each_cpu(cpu, data->old_domain)
  647. per_cpu(vector_irq, cpu)[cfg->old_vector] = VECTOR_UNUSED;
  648. /* Cleanup the left overs of the (half finished) move */
  649. cpumask_clear(data->old_domain);
  650. data->move_in_progress = 0;
  651. raw_spin_unlock(&vector_lock);
  652. }
  653. #endif
  654. static void __init print_APIC_field(int base)
  655. {
  656. int i;
  657. printk(KERN_DEBUG);
  658. for (i = 0; i < 8; i++)
  659. pr_cont("%08x", apic_read(base + i*0x10));
  660. pr_cont("\n");
  661. }
  662. static void __init print_local_APIC(void *dummy)
  663. {
  664. unsigned int i, v, ver, maxlvt;
  665. u64 icr;
  666. pr_debug("printing local APIC contents on CPU#%d/%d:\n",
  667. smp_processor_id(), hard_smp_processor_id());
  668. v = apic_read(APIC_ID);
  669. pr_info("... APIC ID: %08x (%01x)\n", v, read_apic_id());
  670. v = apic_read(APIC_LVR);
  671. pr_info("... APIC VERSION: %08x\n", v);
  672. ver = GET_APIC_VERSION(v);
  673. maxlvt = lapic_get_maxlvt();
  674. v = apic_read(APIC_TASKPRI);
  675. pr_debug("... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
  676. /* !82489DX */
  677. if (APIC_INTEGRATED(ver)) {
  678. if (!APIC_XAPIC(ver)) {
  679. v = apic_read(APIC_ARBPRI);
  680. pr_debug("... APIC ARBPRI: %08x (%02x)\n",
  681. v, v & APIC_ARBPRI_MASK);
  682. }
  683. v = apic_read(APIC_PROCPRI);
  684. pr_debug("... APIC PROCPRI: %08x\n", v);
  685. }
  686. /*
  687. * Remote read supported only in the 82489DX and local APIC for
  688. * Pentium processors.
  689. */
  690. if (!APIC_INTEGRATED(ver) || maxlvt == 3) {
  691. v = apic_read(APIC_RRR);
  692. pr_debug("... APIC RRR: %08x\n", v);
  693. }
  694. v = apic_read(APIC_LDR);
  695. pr_debug("... APIC LDR: %08x\n", v);
  696. if (!x2apic_enabled()) {
  697. v = apic_read(APIC_DFR);
  698. pr_debug("... APIC DFR: %08x\n", v);
  699. }
  700. v = apic_read(APIC_SPIV);
  701. pr_debug("... APIC SPIV: %08x\n", v);
  702. pr_debug("... APIC ISR field:\n");
  703. print_APIC_field(APIC_ISR);
  704. pr_debug("... APIC TMR field:\n");
  705. print_APIC_field(APIC_TMR);
  706. pr_debug("... APIC IRR field:\n");
  707. print_APIC_field(APIC_IRR);
  708. /* !82489DX */
  709. if (APIC_INTEGRATED(ver)) {
  710. /* Due to the Pentium erratum 3AP. */
  711. if (maxlvt > 3)
  712. apic_write(APIC_ESR, 0);
  713. v = apic_read(APIC_ESR);
  714. pr_debug("... APIC ESR: %08x\n", v);
  715. }
  716. icr = apic_icr_read();
  717. pr_debug("... APIC ICR: %08x\n", (u32)icr);
  718. pr_debug("... APIC ICR2: %08x\n", (u32)(icr >> 32));
  719. v = apic_read(APIC_LVTT);
  720. pr_debug("... APIC LVTT: %08x\n", v);
  721. if (maxlvt > 3) {
  722. /* PC is LVT#4. */
  723. v = apic_read(APIC_LVTPC);
  724. pr_debug("... APIC LVTPC: %08x\n", v);
  725. }
  726. v = apic_read(APIC_LVT0);
  727. pr_debug("... APIC LVT0: %08x\n", v);
  728. v = apic_read(APIC_LVT1);
  729. pr_debug("... APIC LVT1: %08x\n", v);
  730. if (maxlvt > 2) {
  731. /* ERR is LVT#3. */
  732. v = apic_read(APIC_LVTERR);
  733. pr_debug("... APIC LVTERR: %08x\n", v);
  734. }
  735. v = apic_read(APIC_TMICT);
  736. pr_debug("... APIC TMICT: %08x\n", v);
  737. v = apic_read(APIC_TMCCT);
  738. pr_debug("... APIC TMCCT: %08x\n", v);
  739. v = apic_read(APIC_TDCR);
  740. pr_debug("... APIC TDCR: %08x\n", v);
  741. if (boot_cpu_has(X86_FEATURE_EXTAPIC)) {
  742. v = apic_read(APIC_EFEAT);
  743. maxlvt = (v >> 16) & 0xff;
  744. pr_debug("... APIC EFEAT: %08x\n", v);
  745. v = apic_read(APIC_ECTRL);
  746. pr_debug("... APIC ECTRL: %08x\n", v);
  747. for (i = 0; i < maxlvt; i++) {
  748. v = apic_read(APIC_EILVTn(i));
  749. pr_debug("... APIC EILVT%d: %08x\n", i, v);
  750. }
  751. }
  752. pr_cont("\n");
  753. }
  754. static void __init print_local_APICs(int maxcpu)
  755. {
  756. int cpu;
  757. if (!maxcpu)
  758. return;
  759. preempt_disable();
  760. for_each_online_cpu(cpu) {
  761. if (cpu >= maxcpu)
  762. break;
  763. smp_call_function_single(cpu, print_local_APIC, NULL, 1);
  764. }
  765. preempt_enable();
  766. }
  767. static void __init print_PIC(void)
  768. {
  769. unsigned int v;
  770. unsigned long flags;
  771. if (!nr_legacy_irqs())
  772. return;
  773. pr_debug("\nprinting PIC contents\n");
  774. raw_spin_lock_irqsave(&i8259A_lock, flags);
  775. v = inb(0xa1) << 8 | inb(0x21);
  776. pr_debug("... PIC IMR: %04x\n", v);
  777. v = inb(0xa0) << 8 | inb(0x20);
  778. pr_debug("... PIC IRR: %04x\n", v);
  779. outb(0x0b, 0xa0);
  780. outb(0x0b, 0x20);
  781. v = inb(0xa0) << 8 | inb(0x20);
  782. outb(0x0a, 0xa0);
  783. outb(0x0a, 0x20);
  784. raw_spin_unlock_irqrestore(&i8259A_lock, flags);
  785. pr_debug("... PIC ISR: %04x\n", v);
  786. v = inb(0x4d1) << 8 | inb(0x4d0);
  787. pr_debug("... PIC ELCR: %04x\n", v);
  788. }
  789. static int show_lapic __initdata = 1;
  790. static __init int setup_show_lapic(char *arg)
  791. {
  792. int num = -1;
  793. if (strcmp(arg, "all") == 0) {
  794. show_lapic = CONFIG_NR_CPUS;
  795. } else {
  796. get_option(&arg, &num);
  797. if (num >= 0)
  798. show_lapic = num;
  799. }
  800. return 1;
  801. }
  802. __setup("show_lapic=", setup_show_lapic);
  803. static int __init print_ICs(void)
  804. {
  805. if (apic_verbosity == APIC_QUIET)
  806. return 0;
  807. print_PIC();
  808. /* don't print out if apic is not there */
  809. if (!cpu_has_apic && !apic_from_smp_config())
  810. return 0;
  811. print_local_APICs(show_lapic);
  812. print_IO_APICs();
  813. return 0;
  814. }
  815. late_initcall(print_ICs);