smp.c 15 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695
  1. /*
  2. * linux/arch/arm/kernel/smp.c
  3. *
  4. * Copyright (C) 2002 ARM Limited, All Rights Reserved.
  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 version 2 as
  8. * published by the Free Software Foundation.
  9. */
  10. #include <linux/module.h>
  11. #include <linux/delay.h>
  12. #include <linux/init.h>
  13. #include <linux/spinlock.h>
  14. #include <linux/sched.h>
  15. #include <linux/interrupt.h>
  16. #include <linux/cache.h>
  17. #include <linux/profile.h>
  18. #include <linux/errno.h>
  19. #include <linux/ftrace.h>
  20. #include <linux/mm.h>
  21. #include <linux/err.h>
  22. #include <linux/cpu.h>
  23. #include <linux/smp.h>
  24. #include <linux/seq_file.h>
  25. #include <linux/irq.h>
  26. #include <linux/percpu.h>
  27. #include <linux/clockchips.h>
  28. #include <linux/completion.h>
  29. #include <linux/cpufreq.h>
  30. #include <asm/atomic.h>
  31. #include <asm/cacheflush.h>
  32. #include <asm/cpu.h>
  33. #include <asm/cputype.h>
  34. #include <asm/mmu_context.h>
  35. #include <asm/pgtable.h>
  36. #include <asm/pgalloc.h>
  37. #include <asm/processor.h>
  38. #include <asm/sections.h>
  39. #include <asm/tlbflush.h>
  40. #include <asm/ptrace.h>
  41. #include <asm/localtimer.h>
  42. /*
  43. * as from 2.5, kernels no longer have an init_tasks structure
  44. * so we need some other way of telling a new secondary core
  45. * where to place its SVC stack
  46. */
  47. struct secondary_data secondary_data;
  48. enum ipi_msg_type {
  49. IPI_TIMER = 2,
  50. IPI_RESCHEDULE,
  51. IPI_CALL_FUNC,
  52. IPI_CALL_FUNC_SINGLE,
  53. IPI_CPU_STOP,
  54. IPI_CPU_BACKTRACE,
  55. };
  56. int __cpuinit __cpu_up(unsigned int cpu)
  57. {
  58. struct cpuinfo_arm *ci = &per_cpu(cpu_data, cpu);
  59. struct task_struct *idle = ci->idle;
  60. pgd_t *pgd;
  61. int ret;
  62. /*
  63. * Spawn a new process manually, if not already done.
  64. * Grab a pointer to its task struct so we can mess with it
  65. */
  66. if (!idle) {
  67. idle = fork_idle(cpu);
  68. if (IS_ERR(idle)) {
  69. printk(KERN_ERR "CPU%u: fork() failed\n", cpu);
  70. return PTR_ERR(idle);
  71. }
  72. ci->idle = idle;
  73. } else {
  74. /*
  75. * Since this idle thread is being re-used, call
  76. * init_idle() to reinitialize the thread structure.
  77. */
  78. init_idle(idle, cpu);
  79. }
  80. /*
  81. * Allocate initial page tables to allow the new CPU to
  82. * enable the MMU safely. This essentially means a set
  83. * of our "standard" page tables, with the addition of
  84. * a 1:1 mapping for the physical address of the kernel.
  85. */
  86. pgd = pgd_alloc(&init_mm);
  87. if (!pgd)
  88. return -ENOMEM;
  89. if (PHYS_OFFSET != PAGE_OFFSET) {
  90. #ifndef CONFIG_HOTPLUG_CPU
  91. identity_mapping_add(pgd, __pa(__init_begin), __pa(__init_end));
  92. #endif
  93. identity_mapping_add(pgd, __pa(_stext), __pa(_etext));
  94. identity_mapping_add(pgd, __pa(_sdata), __pa(_edata));
  95. }
  96. /*
  97. * We need to tell the secondary core where to find
  98. * its stack and the page tables.
  99. */
  100. secondary_data.stack = task_stack_page(idle) + THREAD_START_SP;
  101. secondary_data.pgdir = virt_to_phys(pgd);
  102. secondary_data.swapper_pg_dir = virt_to_phys(swapper_pg_dir);
  103. __cpuc_flush_dcache_area(&secondary_data, sizeof(secondary_data));
  104. outer_clean_range(__pa(&secondary_data), __pa(&secondary_data + 1));
  105. /*
  106. * Now bring the CPU into our world.
  107. */
  108. ret = boot_secondary(cpu, idle);
  109. if (ret == 0) {
  110. unsigned long timeout;
  111. /*
  112. * CPU was successfully started, wait for it
  113. * to come online or time out.
  114. */
  115. timeout = jiffies + (10 * HZ);
  116. while (time_before(jiffies, timeout)) {
  117. if (cpu_online(cpu))
  118. break;
  119. udelay(10);
  120. barrier();
  121. }
  122. if (!cpu_online(cpu)) {
  123. pr_crit("CPU%u: failed to come online\n", cpu);
  124. ret = -EIO;
  125. }
  126. } else {
  127. pr_err("CPU%u: failed to boot: %d\n", cpu, ret);
  128. }
  129. secondary_data.stack = NULL;
  130. secondary_data.pgdir = 0;
  131. if (PHYS_OFFSET != PAGE_OFFSET) {
  132. #ifndef CONFIG_HOTPLUG_CPU
  133. identity_mapping_del(pgd, __pa(__init_begin), __pa(__init_end));
  134. #endif
  135. identity_mapping_del(pgd, __pa(_stext), __pa(_etext));
  136. identity_mapping_del(pgd, __pa(_sdata), __pa(_edata));
  137. }
  138. pgd_free(&init_mm, pgd);
  139. return ret;
  140. }
  141. #ifdef CONFIG_HOTPLUG_CPU
  142. static void percpu_timer_stop(void);
  143. /*
  144. * __cpu_disable runs on the processor to be shutdown.
  145. */
  146. int __cpu_disable(void)
  147. {
  148. unsigned int cpu = smp_processor_id();
  149. struct task_struct *p;
  150. int ret;
  151. ret = platform_cpu_disable(cpu);
  152. if (ret)
  153. return ret;
  154. /*
  155. * Take this CPU offline. Once we clear this, we can't return,
  156. * and we must not schedule until we're ready to give up the cpu.
  157. */
  158. set_cpu_online(cpu, false);
  159. /*
  160. * OK - migrate IRQs away from this CPU
  161. */
  162. migrate_irqs();
  163. /*
  164. * Stop the local timer for this CPU.
  165. */
  166. percpu_timer_stop();
  167. /*
  168. * Flush user cache and TLB mappings, and then remove this CPU
  169. * from the vm mask set of all processes.
  170. */
  171. flush_cache_all();
  172. local_flush_tlb_all();
  173. read_lock(&tasklist_lock);
  174. for_each_process(p) {
  175. if (p->mm)
  176. cpumask_clear_cpu(cpu, mm_cpumask(p->mm));
  177. }
  178. read_unlock(&tasklist_lock);
  179. return 0;
  180. }
  181. static DECLARE_COMPLETION(cpu_died);
  182. /*
  183. * called on the thread which is asking for a CPU to be shutdown -
  184. * waits until shutdown has completed, or it is timed out.
  185. */
  186. void __cpu_die(unsigned int cpu)
  187. {
  188. if (!wait_for_completion_timeout(&cpu_died, msecs_to_jiffies(5000))) {
  189. pr_err("CPU%u: cpu didn't die\n", cpu);
  190. return;
  191. }
  192. printk(KERN_NOTICE "CPU%u: shutdown\n", cpu);
  193. if (!platform_cpu_kill(cpu))
  194. printk("CPU%u: unable to kill\n", cpu);
  195. }
  196. /*
  197. * Called from the idle thread for the CPU which has been shutdown.
  198. *
  199. * Note that we disable IRQs here, but do not re-enable them
  200. * before returning to the caller. This is also the behaviour
  201. * of the other hotplug-cpu capable cores, so presumably coming
  202. * out of idle fixes this.
  203. */
  204. void __ref cpu_die(void)
  205. {
  206. unsigned int cpu = smp_processor_id();
  207. idle_task_exit();
  208. local_irq_disable();
  209. mb();
  210. /* Tell __cpu_die() that this CPU is now safe to dispose of */
  211. complete(&cpu_died);
  212. /*
  213. * actual CPU shutdown procedure is at least platform (if not
  214. * CPU) specific.
  215. */
  216. platform_cpu_die(cpu);
  217. /*
  218. * Do not return to the idle loop - jump back to the secondary
  219. * cpu initialisation. There's some initialisation which needs
  220. * to be repeated to undo the effects of taking the CPU offline.
  221. */
  222. __asm__("mov sp, %0\n"
  223. " mov fp, #0\n"
  224. " b secondary_start_kernel"
  225. :
  226. : "r" (task_stack_page(current) + THREAD_SIZE - 8));
  227. }
  228. #endif /* CONFIG_HOTPLUG_CPU */
  229. /*
  230. * Called by both boot and secondaries to move global data into
  231. * per-processor storage.
  232. */
  233. static void __cpuinit smp_store_cpu_info(unsigned int cpuid)
  234. {
  235. struct cpuinfo_arm *cpu_info = &per_cpu(cpu_data, cpuid);
  236. cpu_info->loops_per_jiffy = loops_per_jiffy;
  237. }
  238. /*
  239. * This is the secondary CPU boot entry. We're using this CPUs
  240. * idle thread stack, but a set of temporary page tables.
  241. */
  242. asmlinkage void __cpuinit secondary_start_kernel(void)
  243. {
  244. struct mm_struct *mm = &init_mm;
  245. unsigned int cpu = smp_processor_id();
  246. static bool booted;
  247. printk("CPU%u: Booted secondary processor\n", cpu);
  248. /*
  249. * All kernel threads share the same mm context; grab a
  250. * reference and switch to it.
  251. */
  252. atomic_inc(&mm->mm_count);
  253. current->active_mm = mm;
  254. cpumask_set_cpu(cpu, mm_cpumask(mm));
  255. cpu_switch_mm(mm->pgd, mm);
  256. enter_lazy_tlb(mm, current);
  257. local_flush_tlb_all();
  258. cpu_init();
  259. preempt_disable();
  260. trace_hardirqs_off();
  261. /*
  262. * Give the platform a chance to do its own initialisation.
  263. */
  264. platform_secondary_init(cpu);
  265. notify_cpu_starting(cpu);
  266. if (!booted)
  267. calibrate_delay();
  268. booted = true;
  269. smp_store_cpu_info(cpu);
  270. /*
  271. * OK, now it's safe to let the boot CPU continue. Wait for
  272. * the CPU migration code to notice that the CPU is online
  273. * before we continue.
  274. */
  275. set_cpu_online(cpu, true);
  276. /*
  277. * Setup the percpu timer for this CPU.
  278. */
  279. percpu_timer_setup();
  280. while (!cpu_active(cpu))
  281. cpu_relax();
  282. /*
  283. * cpu_active bit is set, so it's safe to enable interrupts
  284. * now.
  285. */
  286. local_irq_enable();
  287. local_fiq_enable();
  288. /*
  289. * OK, it's off to the idle thread for us
  290. */
  291. cpu_idle();
  292. }
  293. void __init smp_cpus_done(unsigned int max_cpus)
  294. {
  295. int cpu;
  296. unsigned long bogosum = 0;
  297. for_each_online_cpu(cpu)
  298. bogosum += per_cpu(cpu_data, cpu).loops_per_jiffy;
  299. printk(KERN_INFO "SMP: Total of %d processors activated "
  300. "(%lu.%02lu BogoMIPS).\n",
  301. num_online_cpus(),
  302. bogosum / (500000/HZ),
  303. (bogosum / (5000/HZ)) % 100);
  304. }
  305. void __init smp_prepare_boot_cpu(void)
  306. {
  307. unsigned int cpu = smp_processor_id();
  308. per_cpu(cpu_data, cpu).idle = current;
  309. }
  310. void __init smp_prepare_cpus(unsigned int max_cpus)
  311. {
  312. unsigned int ncores = num_possible_cpus();
  313. smp_store_cpu_info(smp_processor_id());
  314. /*
  315. * are we trying to boot more cores than exist?
  316. */
  317. if (max_cpus > ncores)
  318. max_cpus = ncores;
  319. if (max_cpus > 1) {
  320. /*
  321. * Enable the local timer or broadcast device for the
  322. * boot CPU, but only if we have more than one CPU.
  323. */
  324. percpu_timer_setup();
  325. /*
  326. * Initialise the SCU if there are more than one CPU
  327. * and let them know where to start.
  328. */
  329. platform_smp_prepare_cpus(max_cpus);
  330. }
  331. }
  332. static void (*smp_cross_call)(const struct cpumask *, unsigned int);
  333. void __init set_smp_cross_call(void (*fn)(const struct cpumask *, unsigned int))
  334. {
  335. smp_cross_call = fn;
  336. }
  337. void arch_send_call_function_ipi_mask(const struct cpumask *mask)
  338. {
  339. smp_cross_call(mask, IPI_CALL_FUNC);
  340. }
  341. void arch_send_call_function_single_ipi(int cpu)
  342. {
  343. smp_cross_call(cpumask_of(cpu), IPI_CALL_FUNC_SINGLE);
  344. }
  345. static const char *ipi_types[NR_IPI] = {
  346. #define S(x,s) [x - IPI_TIMER] = s
  347. S(IPI_TIMER, "Timer broadcast interrupts"),
  348. S(IPI_RESCHEDULE, "Rescheduling interrupts"),
  349. S(IPI_CALL_FUNC, "Function call interrupts"),
  350. S(IPI_CALL_FUNC_SINGLE, "Single function call interrupts"),
  351. S(IPI_CPU_STOP, "CPU stop interrupts"),
  352. S(IPI_CPU_BACKTRACE, "CPU backtrace"),
  353. };
  354. void show_ipi_list(struct seq_file *p, int prec)
  355. {
  356. unsigned int cpu, i;
  357. for (i = 0; i < NR_IPI; i++) {
  358. seq_printf(p, "%*s%u: ", prec - 1, "IPI", i);
  359. for_each_present_cpu(cpu)
  360. seq_printf(p, "%10u ",
  361. __get_irq_stat(cpu, ipi_irqs[i]));
  362. seq_printf(p, " %s\n", ipi_types[i]);
  363. }
  364. }
  365. u64 smp_irq_stat_cpu(unsigned int cpu)
  366. {
  367. u64 sum = 0;
  368. int i;
  369. for (i = 0; i < NR_IPI; i++)
  370. sum += __get_irq_stat(cpu, ipi_irqs[i]);
  371. #ifdef CONFIG_LOCAL_TIMERS
  372. sum += __get_irq_stat(cpu, local_timer_irqs);
  373. #endif
  374. return sum;
  375. }
  376. /*
  377. * Timer (local or broadcast) support
  378. */
  379. static DEFINE_PER_CPU(struct clock_event_device, percpu_clockevent);
  380. static void ipi_timer(void)
  381. {
  382. struct clock_event_device *evt = &__get_cpu_var(percpu_clockevent);
  383. irq_enter();
  384. evt->event_handler(evt);
  385. irq_exit();
  386. }
  387. #ifdef CONFIG_LOCAL_TIMERS
  388. asmlinkage void __exception_irq_entry do_local_timer(struct pt_regs *regs)
  389. {
  390. struct pt_regs *old_regs = set_irq_regs(regs);
  391. int cpu = smp_processor_id();
  392. if (local_timer_ack()) {
  393. __inc_irq_stat(cpu, local_timer_irqs);
  394. ipi_timer();
  395. }
  396. set_irq_regs(old_regs);
  397. }
  398. void show_local_irqs(struct seq_file *p, int prec)
  399. {
  400. unsigned int cpu;
  401. seq_printf(p, "%*s: ", prec, "LOC");
  402. for_each_present_cpu(cpu)
  403. seq_printf(p, "%10u ", __get_irq_stat(cpu, local_timer_irqs));
  404. seq_printf(p, " Local timer interrupts\n");
  405. }
  406. #endif
  407. #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
  408. static void smp_timer_broadcast(const struct cpumask *mask)
  409. {
  410. smp_cross_call(mask, IPI_TIMER);
  411. }
  412. #else
  413. #define smp_timer_broadcast NULL
  414. #endif
  415. static void broadcast_timer_set_mode(enum clock_event_mode mode,
  416. struct clock_event_device *evt)
  417. {
  418. }
  419. static void __cpuinit broadcast_timer_setup(struct clock_event_device *evt)
  420. {
  421. evt->name = "dummy_timer";
  422. evt->features = CLOCK_EVT_FEAT_ONESHOT |
  423. CLOCK_EVT_FEAT_PERIODIC |
  424. CLOCK_EVT_FEAT_DUMMY;
  425. evt->rating = 400;
  426. evt->mult = 1;
  427. evt->set_mode = broadcast_timer_set_mode;
  428. clockevents_register_device(evt);
  429. }
  430. void __cpuinit percpu_timer_setup(void)
  431. {
  432. unsigned int cpu = smp_processor_id();
  433. struct clock_event_device *evt = &per_cpu(percpu_clockevent, cpu);
  434. evt->cpumask = cpumask_of(cpu);
  435. evt->broadcast = smp_timer_broadcast;
  436. if (local_timer_setup(evt))
  437. broadcast_timer_setup(evt);
  438. }
  439. #ifdef CONFIG_HOTPLUG_CPU
  440. /*
  441. * The generic clock events code purposely does not stop the local timer
  442. * on CPU_DEAD/CPU_DEAD_FROZEN hotplug events, so we have to do it
  443. * manually here.
  444. */
  445. static void percpu_timer_stop(void)
  446. {
  447. unsigned int cpu = smp_processor_id();
  448. struct clock_event_device *evt = &per_cpu(percpu_clockevent, cpu);
  449. evt->set_mode(CLOCK_EVT_MODE_UNUSED, evt);
  450. }
  451. #endif
  452. static DEFINE_SPINLOCK(stop_lock);
  453. /*
  454. * ipi_cpu_stop - handle IPI from smp_send_stop()
  455. */
  456. static void ipi_cpu_stop(unsigned int cpu)
  457. {
  458. if (system_state == SYSTEM_BOOTING ||
  459. system_state == SYSTEM_RUNNING) {
  460. spin_lock(&stop_lock);
  461. printk(KERN_CRIT "CPU%u: stopping\n", cpu);
  462. dump_stack();
  463. spin_unlock(&stop_lock);
  464. }
  465. set_cpu_online(cpu, false);
  466. local_fiq_disable();
  467. local_irq_disable();
  468. while (1)
  469. cpu_relax();
  470. }
  471. static cpumask_t backtrace_mask;
  472. static DEFINE_RAW_SPINLOCK(backtrace_lock);
  473. /* "in progress" flag of arch_trigger_all_cpu_backtrace */
  474. static unsigned long backtrace_flag;
  475. void smp_send_all_cpu_backtrace(void)
  476. {
  477. unsigned int this_cpu = smp_processor_id();
  478. int i;
  479. if (test_and_set_bit(0, &backtrace_flag))
  480. /*
  481. * If there is already a trigger_all_cpu_backtrace() in progress
  482. * (backtrace_flag == 1), don't output double cpu dump infos.
  483. */
  484. return;
  485. cpumask_copy(&backtrace_mask, cpu_online_mask);
  486. cpu_clear(this_cpu, backtrace_mask);
  487. pr_info("Backtrace for cpu %d (current):\n", this_cpu);
  488. dump_stack();
  489. pr_info("\nsending IPI to all other CPUs:\n");
  490. smp_cross_call(&backtrace_mask, IPI_CPU_BACKTRACE);
  491. /* Wait for up to 10 seconds for all other CPUs to do the backtrace */
  492. for (i = 0; i < 10 * 1000; i++) {
  493. if (cpumask_empty(&backtrace_mask))
  494. break;
  495. mdelay(1);
  496. }
  497. clear_bit(0, &backtrace_flag);
  498. smp_mb__after_clear_bit();
  499. }
  500. /*
  501. * ipi_cpu_backtrace - handle IPI from smp_send_all_cpu_backtrace()
  502. */
  503. static void ipi_cpu_backtrace(unsigned int cpu, struct pt_regs *regs)
  504. {
  505. if (cpu_isset(cpu, backtrace_mask)) {
  506. raw_spin_lock(&backtrace_lock);
  507. pr_warning("IPI backtrace for cpu %d\n", cpu);
  508. show_regs(regs);
  509. raw_spin_unlock(&backtrace_lock);
  510. cpu_clear(cpu, backtrace_mask);
  511. }
  512. }
  513. /*
  514. * Main handler for inter-processor interrupts
  515. */
  516. asmlinkage void __exception_irq_entry do_IPI(int ipinr, struct pt_regs *regs)
  517. {
  518. unsigned int cpu = smp_processor_id();
  519. struct pt_regs *old_regs = set_irq_regs(regs);
  520. if (ipinr >= IPI_TIMER && ipinr < IPI_TIMER + NR_IPI)
  521. __inc_irq_stat(cpu, ipi_irqs[ipinr - IPI_TIMER]);
  522. switch (ipinr) {
  523. case IPI_TIMER:
  524. ipi_timer();
  525. break;
  526. case IPI_RESCHEDULE:
  527. scheduler_ipi();
  528. break;
  529. case IPI_CALL_FUNC:
  530. generic_smp_call_function_interrupt();
  531. break;
  532. case IPI_CALL_FUNC_SINGLE:
  533. generic_smp_call_function_single_interrupt();
  534. break;
  535. case IPI_CPU_STOP:
  536. ipi_cpu_stop(cpu);
  537. break;
  538. case IPI_CPU_BACKTRACE:
  539. ipi_cpu_backtrace(cpu, regs);
  540. break;
  541. default:
  542. printk(KERN_CRIT "CPU%u: Unknown IPI message 0x%x\n",
  543. cpu, ipinr);
  544. break;
  545. }
  546. set_irq_regs(old_regs);
  547. }
  548. void smp_send_reschedule(int cpu)
  549. {
  550. smp_cross_call(cpumask_of(cpu), IPI_RESCHEDULE);
  551. }
  552. void smp_send_stop(void)
  553. {
  554. unsigned long timeout;
  555. if (num_online_cpus() > 1) {
  556. cpumask_t mask = cpu_online_map;
  557. cpu_clear(smp_processor_id(), mask);
  558. smp_cross_call(&mask, IPI_CPU_STOP);
  559. }
  560. /* Wait up to one second for other CPUs to stop */
  561. timeout = USEC_PER_SEC;
  562. while (num_online_cpus() > 1 && timeout--)
  563. udelay(1);
  564. if (num_online_cpus() > 1)
  565. pr_warning("SMP: failed to stop secondary CPUs\n");
  566. }
  567. /*
  568. * not supported here
  569. */
  570. int setup_profiling_timer(unsigned int multiplier)
  571. {
  572. return -EINVAL;
  573. }