smp.c 12 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510
  1. /*
  2. * This program is free software; you can redistribute it and/or
  3. * modify it under the terms of the GNU General Public License
  4. * as published by the Free Software Foundation; either version 2
  5. * of the License, or (at your option) any later version.
  6. *
  7. * This program is distributed in the hope that it will be useful,
  8. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  9. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  10. * GNU General Public License for more details.
  11. *
  12. * You should have received a copy of the GNU General Public License
  13. * along with this program; if not, write to the Free Software
  14. * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  15. *
  16. * Copyright (C) 2000, 2001 Kanoj Sarcar
  17. * Copyright (C) 2000, 2001 Ralf Baechle
  18. * Copyright (C) 2000, 2001 Silicon Graphics, Inc.
  19. * Copyright (C) 2000, 2001, 2003 Broadcom Corporation
  20. */
  21. #include <linux/cache.h>
  22. #include <linux/delay.h>
  23. #include <linux/init.h>
  24. #include <linux/interrupt.h>
  25. #include <linux/smp.h>
  26. #include <linux/spinlock.h>
  27. #include <linux/threads.h>
  28. #include <linux/module.h>
  29. #include <linux/time.h>
  30. #include <linux/timex.h>
  31. #include <linux/sched.h>
  32. #include <linux/cpumask.h>
  33. #include <linux/cpu.h>
  34. #include <linux/err.h>
  35. #include <linux/ftrace.h>
  36. #include <linux/atomic.h>
  37. #include <asm/cpu.h>
  38. #include <asm/processor.h>
  39. #include <asm/idle.h>
  40. #include <asm/r4k-timer.h>
  41. #include <asm/mmu_context.h>
  42. #include <asm/time.h>
  43. #include <asm/setup.h>
  44. #include <asm/maar.h>
  45. cpumask_t cpu_callin_map; /* Bitmask of started secondaries */
  46. int __cpu_number_map[NR_CPUS]; /* Map physical to logical */
  47. EXPORT_SYMBOL(__cpu_number_map);
  48. int __cpu_logical_map[NR_CPUS]; /* Map logical to physical */
  49. EXPORT_SYMBOL(__cpu_logical_map);
  50. /* Number of TCs (or siblings in Intel speak) per CPU core */
  51. int smp_num_siblings = 1;
  52. EXPORT_SYMBOL(smp_num_siblings);
  53. /* representing the TCs (or siblings in Intel speak) of each logical CPU */
  54. cpumask_t cpu_sibling_map[NR_CPUS] __read_mostly;
  55. EXPORT_SYMBOL(cpu_sibling_map);
  56. /* representing the core map of multi-core chips of each logical CPU */
  57. cpumask_t cpu_core_map[NR_CPUS] __read_mostly;
  58. EXPORT_SYMBOL(cpu_core_map);
  59. static DECLARE_COMPLETION(cpu_starting);
  60. static DECLARE_COMPLETION(cpu_running);
  61. /*
  62. * A logcal cpu mask containing only one VPE per core to
  63. * reduce the number of IPIs on large MT systems.
  64. */
  65. cpumask_t cpu_foreign_map __read_mostly;
  66. EXPORT_SYMBOL(cpu_foreign_map);
  67. /* representing cpus for which sibling maps can be computed */
  68. static cpumask_t cpu_sibling_setup_map;
  69. /* representing cpus for which core maps can be computed */
  70. static cpumask_t cpu_core_setup_map;
  71. cpumask_t cpu_coherent_mask;
  72. static inline void set_cpu_sibling_map(int cpu)
  73. {
  74. int i;
  75. cpumask_set_cpu(cpu, &cpu_sibling_setup_map);
  76. if (smp_num_siblings > 1) {
  77. for_each_cpu(i, &cpu_sibling_setup_map) {
  78. if (cpu_data[cpu].package == cpu_data[i].package &&
  79. cpu_data[cpu].core == cpu_data[i].core) {
  80. cpumask_set_cpu(i, &cpu_sibling_map[cpu]);
  81. cpumask_set_cpu(cpu, &cpu_sibling_map[i]);
  82. }
  83. }
  84. } else
  85. cpumask_set_cpu(cpu, &cpu_sibling_map[cpu]);
  86. }
  87. static inline void set_cpu_core_map(int cpu)
  88. {
  89. int i;
  90. cpumask_set_cpu(cpu, &cpu_core_setup_map);
  91. for_each_cpu(i, &cpu_core_setup_map) {
  92. if (cpu_data[cpu].package == cpu_data[i].package) {
  93. cpumask_set_cpu(i, &cpu_core_map[cpu]);
  94. cpumask_set_cpu(cpu, &cpu_core_map[i]);
  95. }
  96. }
  97. }
  98. /*
  99. * Calculate a new cpu_foreign_map mask whenever a
  100. * new cpu appears or disappears.
  101. */
  102. static inline void calculate_cpu_foreign_map(void)
  103. {
  104. int i, k, core_present;
  105. cpumask_t temp_foreign_map;
  106. /* Re-calculate the mask */
  107. cpumask_clear(&temp_foreign_map);
  108. for_each_online_cpu(i) {
  109. core_present = 0;
  110. for_each_cpu(k, &temp_foreign_map)
  111. if (cpu_data[i].package == cpu_data[k].package &&
  112. cpu_data[i].core == cpu_data[k].core)
  113. core_present = 1;
  114. if (!core_present)
  115. cpumask_set_cpu(i, &temp_foreign_map);
  116. }
  117. cpumask_copy(&cpu_foreign_map, &temp_foreign_map);
  118. }
  119. struct plat_smp_ops *mp_ops;
  120. EXPORT_SYMBOL(mp_ops);
  121. void register_smp_ops(struct plat_smp_ops *ops)
  122. {
  123. if (mp_ops)
  124. printk(KERN_WARNING "Overriding previously set SMP ops\n");
  125. mp_ops = ops;
  126. }
  127. /*
  128. * First C code run on the secondary CPUs after being started up by
  129. * the master.
  130. */
  131. asmlinkage void start_secondary(void)
  132. {
  133. unsigned int cpu;
  134. cpu_probe();
  135. per_cpu_trap_init(false);
  136. mips_clockevent_init();
  137. mp_ops->init_secondary();
  138. cpu_report();
  139. maar_init();
  140. /*
  141. * XXX parity protection should be folded in here when it's converted
  142. * to an option instead of something based on .cputype
  143. */
  144. calibrate_delay();
  145. preempt_disable();
  146. cpu = smp_processor_id();
  147. cpu_data[cpu].udelay_val = loops_per_jiffy;
  148. cpumask_set_cpu(cpu, &cpu_coherent_mask);
  149. notify_cpu_starting(cpu);
  150. /* Notify boot CPU that we're starting & ready to sync counters */
  151. complete(&cpu_starting);
  152. synchronise_count_slave(cpu);
  153. /* The CPU is running and counters synchronised, now mark it online */
  154. set_cpu_online(cpu, true);
  155. set_cpu_sibling_map(cpu);
  156. set_cpu_core_map(cpu);
  157. calculate_cpu_foreign_map();
  158. /*
  159. * Notify boot CPU that we're up & online and it can safely return
  160. * from __cpu_up
  161. */
  162. complete(&cpu_running);
  163. /*
  164. * irq will be enabled in ->smp_finish(), enabling it too early
  165. * is dangerous.
  166. */
  167. WARN_ON_ONCE(!irqs_disabled());
  168. mp_ops->smp_finish();
  169. cpu_startup_entry(CPUHP_ONLINE);
  170. }
  171. static void stop_this_cpu(void *dummy)
  172. {
  173. /*
  174. * Remove this CPU. Be a bit slow here and
  175. * set the bits for every online CPU so we don't miss
  176. * any IPI whilst taking this VPE down.
  177. */
  178. cpumask_copy(&cpu_foreign_map, cpu_online_mask);
  179. /* Make it visible to every other CPU */
  180. smp_mb();
  181. set_cpu_online(smp_processor_id(), false);
  182. calculate_cpu_foreign_map();
  183. local_irq_disable();
  184. while (1);
  185. }
  186. void smp_send_stop(void)
  187. {
  188. smp_call_function(stop_this_cpu, NULL, 0);
  189. }
  190. void __init smp_cpus_done(unsigned int max_cpus)
  191. {
  192. }
  193. /* called from main before smp_init() */
  194. void __init smp_prepare_cpus(unsigned int max_cpus)
  195. {
  196. init_new_context(current, &init_mm);
  197. current_thread_info()->cpu = 0;
  198. mp_ops->prepare_cpus(max_cpus);
  199. set_cpu_sibling_map(0);
  200. set_cpu_core_map(0);
  201. calculate_cpu_foreign_map();
  202. #ifndef CONFIG_HOTPLUG_CPU
  203. init_cpu_present(cpu_possible_mask);
  204. #endif
  205. cpumask_copy(&cpu_coherent_mask, cpu_possible_mask);
  206. }
  207. /* preload SMP state for boot cpu */
  208. void smp_prepare_boot_cpu(void)
  209. {
  210. set_cpu_possible(0, true);
  211. set_cpu_online(0, true);
  212. }
  213. int __cpu_up(unsigned int cpu, struct task_struct *tidle)
  214. {
  215. mp_ops->boot_secondary(cpu, tidle);
  216. /* Wait for CPU to start and be ready to sync counters */
  217. if (!wait_for_completion_timeout(&cpu_starting,
  218. msecs_to_jiffies(1000))) {
  219. pr_crit("CPU%u: failed to start\n", cpu);
  220. return -EIO;
  221. }
  222. synchronise_count_master(cpu);
  223. /* Wait for CPU to finish startup & mark itself online before return */
  224. wait_for_completion(&cpu_running);
  225. return 0;
  226. }
  227. /* Not really SMP stuff ... */
  228. int setup_profiling_timer(unsigned int multiplier)
  229. {
  230. return 0;
  231. }
  232. static void flush_tlb_all_ipi(void *info)
  233. {
  234. local_flush_tlb_all();
  235. }
  236. void flush_tlb_all(void)
  237. {
  238. on_each_cpu(flush_tlb_all_ipi, NULL, 1);
  239. }
  240. static void flush_tlb_mm_ipi(void *mm)
  241. {
  242. local_flush_tlb_mm((struct mm_struct *)mm);
  243. }
  244. /*
  245. * Special Variant of smp_call_function for use by TLB functions:
  246. *
  247. * o No return value
  248. * o collapses to normal function call on UP kernels
  249. * o collapses to normal function call on systems with a single shared
  250. * primary cache.
  251. */
  252. static inline void smp_on_other_tlbs(void (*func) (void *info), void *info)
  253. {
  254. smp_call_function(func, info, 1);
  255. }
  256. static inline void smp_on_each_tlb(void (*func) (void *info), void *info)
  257. {
  258. preempt_disable();
  259. smp_on_other_tlbs(func, info);
  260. func(info);
  261. preempt_enable();
  262. }
  263. /*
  264. * The following tlb flush calls are invoked when old translations are
  265. * being torn down, or pte attributes are changing. For single threaded
  266. * address spaces, a new context is obtained on the current cpu, and tlb
  267. * context on other cpus are invalidated to force a new context allocation
  268. * at switch_mm time, should the mm ever be used on other cpus. For
  269. * multithreaded address spaces, intercpu interrupts have to be sent.
  270. * Another case where intercpu interrupts are required is when the target
  271. * mm might be active on another cpu (eg debuggers doing the flushes on
  272. * behalf of debugees, kswapd stealing pages from another process etc).
  273. * Kanoj 07/00.
  274. */
  275. void flush_tlb_mm(struct mm_struct *mm)
  276. {
  277. preempt_disable();
  278. if ((atomic_read(&mm->mm_users) != 1) || (current->mm != mm)) {
  279. smp_on_other_tlbs(flush_tlb_mm_ipi, mm);
  280. } else {
  281. unsigned int cpu;
  282. for_each_online_cpu(cpu) {
  283. if (cpu != smp_processor_id() && cpu_context(cpu, mm))
  284. cpu_context(cpu, mm) = 0;
  285. }
  286. }
  287. local_flush_tlb_mm(mm);
  288. preempt_enable();
  289. }
  290. struct flush_tlb_data {
  291. struct vm_area_struct *vma;
  292. unsigned long addr1;
  293. unsigned long addr2;
  294. };
  295. static void flush_tlb_range_ipi(void *info)
  296. {
  297. struct flush_tlb_data *fd = info;
  298. local_flush_tlb_range(fd->vma, fd->addr1, fd->addr2);
  299. }
  300. void flush_tlb_range(struct vm_area_struct *vma, unsigned long start, unsigned long end)
  301. {
  302. struct mm_struct *mm = vma->vm_mm;
  303. preempt_disable();
  304. if ((atomic_read(&mm->mm_users) != 1) || (current->mm != mm)) {
  305. struct flush_tlb_data fd = {
  306. .vma = vma,
  307. .addr1 = start,
  308. .addr2 = end,
  309. };
  310. smp_on_other_tlbs(flush_tlb_range_ipi, &fd);
  311. } else {
  312. unsigned int cpu;
  313. for_each_online_cpu(cpu) {
  314. if (cpu != smp_processor_id() && cpu_context(cpu, mm))
  315. cpu_context(cpu, mm) = 0;
  316. }
  317. }
  318. local_flush_tlb_range(vma, start, end);
  319. preempt_enable();
  320. }
  321. static void flush_tlb_kernel_range_ipi(void *info)
  322. {
  323. struct flush_tlb_data *fd = info;
  324. local_flush_tlb_kernel_range(fd->addr1, fd->addr2);
  325. }
  326. void flush_tlb_kernel_range(unsigned long start, unsigned long end)
  327. {
  328. struct flush_tlb_data fd = {
  329. .addr1 = start,
  330. .addr2 = end,
  331. };
  332. on_each_cpu(flush_tlb_kernel_range_ipi, &fd, 1);
  333. }
  334. static void flush_tlb_page_ipi(void *info)
  335. {
  336. struct flush_tlb_data *fd = info;
  337. local_flush_tlb_page(fd->vma, fd->addr1);
  338. }
  339. void flush_tlb_page(struct vm_area_struct *vma, unsigned long page)
  340. {
  341. preempt_disable();
  342. if ((atomic_read(&vma->vm_mm->mm_users) != 1) || (current->mm != vma->vm_mm)) {
  343. struct flush_tlb_data fd = {
  344. .vma = vma,
  345. .addr1 = page,
  346. };
  347. smp_on_other_tlbs(flush_tlb_page_ipi, &fd);
  348. } else {
  349. unsigned int cpu;
  350. for_each_online_cpu(cpu) {
  351. if (cpu != smp_processor_id() && cpu_context(cpu, vma->vm_mm))
  352. cpu_context(cpu, vma->vm_mm) = 0;
  353. }
  354. }
  355. local_flush_tlb_page(vma, page);
  356. preempt_enable();
  357. }
  358. static void flush_tlb_one_ipi(void *info)
  359. {
  360. unsigned long vaddr = (unsigned long) info;
  361. local_flush_tlb_one(vaddr);
  362. }
  363. void flush_tlb_one(unsigned long vaddr)
  364. {
  365. smp_on_each_tlb(flush_tlb_one_ipi, (void *) vaddr);
  366. }
  367. EXPORT_SYMBOL(flush_tlb_page);
  368. EXPORT_SYMBOL(flush_tlb_one);
  369. #if defined(CONFIG_KEXEC)
  370. void (*dump_ipi_function_ptr)(void *) = NULL;
  371. void dump_send_ipi(void (*dump_ipi_callback)(void *))
  372. {
  373. int i;
  374. int cpu = smp_processor_id();
  375. dump_ipi_function_ptr = dump_ipi_callback;
  376. smp_mb();
  377. for_each_online_cpu(i)
  378. if (i != cpu)
  379. mp_ops->send_ipi_single(i, SMP_DUMP);
  380. }
  381. EXPORT_SYMBOL(dump_send_ipi);
  382. #endif
  383. #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
  384. static DEFINE_PER_CPU(atomic_t, tick_broadcast_count);
  385. static DEFINE_PER_CPU(struct call_single_data, tick_broadcast_csd);
  386. void tick_broadcast(const struct cpumask *mask)
  387. {
  388. atomic_t *count;
  389. struct call_single_data *csd;
  390. int cpu;
  391. for_each_cpu(cpu, mask) {
  392. count = &per_cpu(tick_broadcast_count, cpu);
  393. csd = &per_cpu(tick_broadcast_csd, cpu);
  394. if (atomic_inc_return(count) == 1)
  395. smp_call_function_single_async(cpu, csd);
  396. }
  397. }
  398. static void tick_broadcast_callee(void *info)
  399. {
  400. int cpu = smp_processor_id();
  401. tick_receive_broadcast();
  402. atomic_set(&per_cpu(tick_broadcast_count, cpu), 0);
  403. }
  404. static int __init tick_broadcast_init(void)
  405. {
  406. struct call_single_data *csd;
  407. int cpu;
  408. for (cpu = 0; cpu < NR_CPUS; cpu++) {
  409. csd = &per_cpu(tick_broadcast_csd, cpu);
  410. csd->func = tick_broadcast_callee;
  411. }
  412. return 0;
  413. }
  414. early_initcall(tick_broadcast_init);
  415. #endif /* CONFIG_GENERIC_CLOCKEVENTS_BROADCAST */