2 * linux/arch/arm/kernel/smp.c
4 * Copyright (C) 2002 ARM Limited, All Rights Reserved.
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.
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>
20 #include <linux/err.h>
21 #include <linux/cpu.h>
22 #include <linux/seq_file.h>
23 #include <linux/irq.h>
24 #include <linux/percpu.h>
25 #include <linux/clockchips.h>
26 #include <linux/completion.h>
27 #include <linux/cpufreq.h>
29 #include <linux/atomic.h>
31 #include <asm/cacheflush.h>
33 #include <asm/cputype.h>
34 #include <asm/exception.h>
35 #include <asm/idmap.h>
36 #include <asm/topology.h>
37 #include <asm/mmu_context.h>
38 #include <asm/pgtable.h>
39 #include <asm/pgalloc.h>
40 #include <asm/processor.h>
41 #include <asm/sections.h>
42 #include <asm/tlbflush.h>
43 #include <asm/ptrace.h>
44 #include <asm/localtimer.h>
45 #include <asm/smp_plat.h>
47 #include <asm/mach/arch.h>
51 * as from 2.5, kernels no longer have an init_tasks structure
52 * so we need some other way of telling a new secondary core
53 * where to place its SVC stack
55 struct secondary_data secondary_data;
58 * control for which core is the next to come out of the secondary
61 volatile int pen_release = -1;
72 static DECLARE_COMPLETION(cpu_running);
74 static struct smp_operations smp_ops;
76 void __init smp_set_ops(struct smp_operations *ops)
82 static unsigned long get_arch_pgd(pgd_t *pgd)
84 phys_addr_t pgdir = virt_to_phys(pgd);
85 BUG_ON(pgdir & ARCH_PGD_MASK);
86 return pgdir >> ARCH_PGD_SHIFT;
89 int __cpu_up(unsigned int cpu, struct task_struct *idle)
94 * We need to tell the secondary core where to find
95 * its stack and the page tables.
97 secondary_data.stack = task_stack_page(idle) + THREAD_START_SP;
99 secondary_data.mpu_rgn_szr = mpu_rgn_info.rgns[MPU_RAM_REGION].drsr;
103 secondary_data.pgdir = get_arch_pgd(idmap_pgd);
104 secondary_data.swapper_pg_dir = get_arch_pgd(swapper_pg_dir);
106 __cpuc_flush_dcache_area(&secondary_data, sizeof(secondary_data));
107 outer_clean_range(__pa(&secondary_data), __pa(&secondary_data + 1));
110 * Now bring the CPU into our world.
112 ret = boot_secondary(cpu, idle);
115 * CPU was successfully started, wait for it
116 * to come online or time out.
118 wait_for_completion_timeout(&cpu_running,
119 msecs_to_jiffies(1000));
121 if (!cpu_online(cpu)) {
122 pr_crit("CPU%u: failed to come online\n", cpu);
126 pr_err("CPU%u: failed to boot: %d\n", cpu, ret);
130 memset(&secondary_data, 0, sizeof(secondary_data));
134 /* platform specific SMP operations */
135 void __init smp_init_cpus(void)
137 if (smp_ops.smp_init_cpus)
138 smp_ops.smp_init_cpus();
141 int boot_secondary(unsigned int cpu, struct task_struct *idle)
143 if (smp_ops.smp_boot_secondary)
144 return smp_ops.smp_boot_secondary(cpu, idle);
148 int platform_can_cpu_hotplug(void)
150 #ifdef CONFIG_HOTPLUG_CPU
151 if (smp_ops.cpu_kill)
158 #ifdef CONFIG_HOTPLUG_CPU
159 static void percpu_timer_stop(void);
161 static int platform_cpu_kill(unsigned int cpu)
163 if (smp_ops.cpu_kill)
164 return smp_ops.cpu_kill(cpu);
168 static int platform_cpu_disable(unsigned int cpu)
170 if (smp_ops.cpu_disable)
171 return smp_ops.cpu_disable(cpu);
174 * By default, allow disabling all CPUs except the first one,
175 * since this is special on a lot of platforms, e.g. because
176 * of clock tick interrupts.
178 return cpu == 0 ? -EPERM : 0;
181 * __cpu_disable runs on the processor to be shutdown.
183 int __cpu_disable(void)
185 unsigned int cpu = smp_processor_id();
188 ret = platform_cpu_disable(cpu);
193 * Take this CPU offline. Once we clear this, we can't return,
194 * and we must not schedule until we're ready to give up the cpu.
196 set_cpu_online(cpu, false);
199 * OK - migrate IRQs away from this CPU
204 * Stop the local timer for this CPU.
209 * Flush user cache and TLB mappings, and then remove this CPU
210 * from the vm mask set of all processes.
212 * Caches are flushed to the Level of Unification Inner Shareable
213 * to write-back dirty lines to unified caches shared by all CPUs.
216 local_flush_tlb_all();
218 clear_tasks_mm_cpumask(cpu);
223 static DECLARE_COMPLETION(cpu_died);
226 * called on the thread which is asking for a CPU to be shutdown -
227 * waits until shutdown has completed, or it is timed out.
229 void __cpu_die(unsigned int cpu)
231 if (!wait_for_completion_timeout(&cpu_died, msecs_to_jiffies(5000))) {
232 pr_err("CPU%u: cpu didn't die\n", cpu);
235 printk(KERN_NOTICE "CPU%u: shutdown\n", cpu);
238 * platform_cpu_kill() is generally expected to do the powering off
239 * and/or cutting of clocks to the dying CPU. Optionally, this may
240 * be done by the CPU which is dying in preference to supporting
241 * this call, but that means there is _no_ synchronisation between
242 * the requesting CPU and the dying CPU actually losing power.
244 if (!platform_cpu_kill(cpu))
245 printk("CPU%u: unable to kill\n", cpu);
249 * Called from the idle thread for the CPU which has been shutdown.
251 * Note that we disable IRQs here, but do not re-enable them
252 * before returning to the caller. This is also the behaviour
253 * of the other hotplug-cpu capable cores, so presumably coming
254 * out of idle fixes this.
256 void __ref cpu_die(void)
258 unsigned int cpu = smp_processor_id();
265 * Flush the data out of the L1 cache for this CPU. This must be
266 * before the completion to ensure that data is safely written out
267 * before platform_cpu_kill() gets called - which may disable
268 * *this* CPU and power down its cache.
273 * Tell __cpu_die() that this CPU is now safe to dispose of. Once
274 * this returns, power and/or clocks can be removed at any point
275 * from this CPU and its cache by platform_cpu_kill().
280 * Ensure that the cache lines associated with that completion are
281 * written out. This covers the case where _this_ CPU is doing the
282 * powering down, to ensure that the completion is visible to the
283 * CPU waiting for this one.
288 * The actual CPU shutdown procedure is at least platform (if not
289 * CPU) specific. This may remove power, or it may simply spin.
291 * Platforms are generally expected *NOT* to return from this call,
292 * although there are some which do because they have no way to
293 * power down the CPU. These platforms are the _only_ reason we
294 * have a return path which uses the fragment of assembly below.
296 * The return path should not be used for platforms which can
300 smp_ops.cpu_die(cpu);
303 * Do not return to the idle loop - jump back to the secondary
304 * cpu initialisation. There's some initialisation which needs
305 * to be repeated to undo the effects of taking the CPU offline.
307 __asm__("mov sp, %0\n"
309 " b secondary_start_kernel"
311 : "r" (task_stack_page(current) + THREAD_SIZE - 8));
313 #endif /* CONFIG_HOTPLUG_CPU */
316 * Called by both boot and secondaries to move global data into
317 * per-processor storage.
319 static void smp_store_cpu_info(unsigned int cpuid)
321 struct cpuinfo_arm *cpu_info = &per_cpu(cpu_data, cpuid);
323 cpu_info->loops_per_jiffy = loops_per_jiffy;
324 cpu_info->cpuid = read_cpuid_id();
326 store_cpu_topology(cpuid);
329 static void percpu_timer_setup(void);
332 * This is the secondary CPU boot entry. We're using this CPUs
333 * idle thread stack, but a set of temporary page tables.
335 asmlinkage void secondary_start_kernel(void)
337 struct mm_struct *mm = &init_mm;
341 * The identity mapping is uncached (strongly ordered), so
342 * switch away from it before attempting any exclusive accesses.
344 cpu_switch_mm(mm->pgd, mm);
345 local_flush_bp_all();
346 enter_lazy_tlb(mm, current);
347 local_flush_tlb_all();
350 * All kernel threads share the same mm context; grab a
351 * reference and switch to it.
353 cpu = smp_processor_id();
354 atomic_inc(&mm->mm_count);
355 current->active_mm = mm;
356 cpumask_set_cpu(cpu, mm_cpumask(mm));
360 printk("CPU%u: Booted secondary processor\n", cpu);
363 trace_hardirqs_off();
366 * Give the platform a chance to do its own initialisation.
368 if (smp_ops.smp_secondary_init)
369 smp_ops.smp_secondary_init(cpu);
371 notify_cpu_starting(cpu);
375 smp_store_cpu_info(cpu);
378 * OK, now it's safe to let the boot CPU continue. Wait for
379 * the CPU migration code to notice that the CPU is online
380 * before we continue - which happens after __cpu_up returns.
382 set_cpu_online(cpu, true);
383 complete(&cpu_running);
386 * Setup the percpu timer for this CPU.
388 percpu_timer_setup();
394 * OK, it's off to the idle thread for us
396 cpu_startup_entry(CPUHP_ONLINE);
399 void __init smp_cpus_done(unsigned int max_cpus)
402 unsigned long bogosum = 0;
404 for_each_online_cpu(cpu)
405 bogosum += per_cpu(cpu_data, cpu).loops_per_jiffy;
407 printk(KERN_INFO "SMP: Total of %d processors activated "
408 "(%lu.%02lu BogoMIPS).\n",
410 bogosum / (500000/HZ),
411 (bogosum / (5000/HZ)) % 100);
416 void __init smp_prepare_boot_cpu(void)
418 set_my_cpu_offset(per_cpu_offset(smp_processor_id()));
421 void __init smp_prepare_cpus(unsigned int max_cpus)
423 unsigned int ncores = num_possible_cpus();
427 smp_store_cpu_info(smp_processor_id());
430 * are we trying to boot more cores than exist?
432 if (max_cpus > ncores)
434 if (ncores > 1 && max_cpus) {
436 * Enable the local timer or broadcast device for the
437 * boot CPU, but only if we have more than one CPU.
439 percpu_timer_setup();
442 * Initialise the present map, which describes the set of CPUs
443 * actually populated at the present time. A platform should
444 * re-initialize the map in the platforms smp_prepare_cpus()
445 * if present != possible (e.g. physical hotplug).
447 init_cpu_present(cpu_possible_mask);
450 * Initialise the SCU if there are more than one CPU
451 * and let them know where to start.
453 if (smp_ops.smp_prepare_cpus)
454 smp_ops.smp_prepare_cpus(max_cpus);
458 static void (*smp_cross_call)(const struct cpumask *, unsigned int);
460 void __init set_smp_cross_call(void (*fn)(const struct cpumask *, unsigned int))
466 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
468 smp_cross_call(mask, IPI_CALL_FUNC);
471 void arch_send_wakeup_ipi_mask(const struct cpumask *mask)
473 smp_cross_call(mask, IPI_WAKEUP);
476 void arch_send_call_function_single_ipi(int cpu)
478 smp_cross_call(cpumask_of(cpu), IPI_CALL_FUNC_SINGLE);
481 static const char *ipi_types[NR_IPI] = {
482 #define S(x,s) [x] = s
483 S(IPI_WAKEUP, "CPU wakeup interrupts"),
484 S(IPI_TIMER, "Timer broadcast interrupts"),
485 S(IPI_RESCHEDULE, "Rescheduling interrupts"),
486 S(IPI_CALL_FUNC, "Function call interrupts"),
487 S(IPI_CALL_FUNC_SINGLE, "Single function call interrupts"),
488 S(IPI_CPU_STOP, "CPU stop interrupts"),
491 void show_ipi_list(struct seq_file *p, int prec)
495 for (i = 0; i < NR_IPI; i++) {
496 seq_printf(p, "%*s%u: ", prec - 1, "IPI", i);
498 for_each_online_cpu(cpu)
499 seq_printf(p, "%10u ",
500 __get_irq_stat(cpu, ipi_irqs[i]));
502 seq_printf(p, " %s\n", ipi_types[i]);
506 u64 smp_irq_stat_cpu(unsigned int cpu)
511 for (i = 0; i < NR_IPI; i++)
512 sum += __get_irq_stat(cpu, ipi_irqs[i]);
518 * Timer (local or broadcast) support
520 static DEFINE_PER_CPU(struct clock_event_device, percpu_clockevent);
522 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
523 void tick_broadcast(const struct cpumask *mask)
525 smp_cross_call(mask, IPI_TIMER);
529 static void broadcast_timer_set_mode(enum clock_event_mode mode,
530 struct clock_event_device *evt)
534 static void broadcast_timer_setup(struct clock_event_device *evt)
536 evt->name = "dummy_timer";
537 evt->features = CLOCK_EVT_FEAT_ONESHOT |
538 CLOCK_EVT_FEAT_PERIODIC |
539 CLOCK_EVT_FEAT_DUMMY;
542 evt->set_mode = broadcast_timer_set_mode;
544 clockevents_register_device(evt);
547 static struct local_timer_ops *lt_ops;
549 #ifdef CONFIG_LOCAL_TIMERS
550 int local_timer_register(struct local_timer_ops *ops)
552 if (!is_smp() || !setup_max_cpus)
563 static void percpu_timer_setup(void)
565 unsigned int cpu = smp_processor_id();
566 struct clock_event_device *evt = &per_cpu(percpu_clockevent, cpu);
568 evt->cpumask = cpumask_of(cpu);
570 if (!lt_ops || lt_ops->setup(evt))
571 broadcast_timer_setup(evt);
574 #ifdef CONFIG_HOTPLUG_CPU
576 * The generic clock events code purposely does not stop the local timer
577 * on CPU_DEAD/CPU_DEAD_FROZEN hotplug events, so we have to do it
580 static void percpu_timer_stop(void)
582 unsigned int cpu = smp_processor_id();
583 struct clock_event_device *evt = &per_cpu(percpu_clockevent, cpu);
590 static DEFINE_RAW_SPINLOCK(stop_lock);
593 * ipi_cpu_stop - handle IPI from smp_send_stop()
595 static void ipi_cpu_stop(unsigned int cpu)
597 if (system_state == SYSTEM_BOOTING ||
598 system_state == SYSTEM_RUNNING) {
599 raw_spin_lock(&stop_lock);
600 printk(KERN_CRIT "CPU%u: stopping\n", cpu);
602 raw_spin_unlock(&stop_lock);
605 set_cpu_online(cpu, false);
615 * Main handler for inter-processor interrupts
617 asmlinkage void __exception_irq_entry do_IPI(int ipinr, struct pt_regs *regs)
619 handle_IPI(ipinr, regs);
622 void handle_IPI(int ipinr, struct pt_regs *regs)
624 unsigned int cpu = smp_processor_id();
625 struct pt_regs *old_regs = set_irq_regs(regs);
628 __inc_irq_stat(cpu, ipi_irqs[ipinr]);
634 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
637 tick_receive_broadcast();
648 generic_smp_call_function_interrupt();
652 case IPI_CALL_FUNC_SINGLE:
654 generic_smp_call_function_single_interrupt();
665 printk(KERN_CRIT "CPU%u: Unknown IPI message 0x%x\n",
669 set_irq_regs(old_regs);
672 void smp_send_reschedule(int cpu)
674 smp_cross_call(cpumask_of(cpu), IPI_RESCHEDULE);
677 void smp_send_stop(void)
679 unsigned long timeout;
682 cpumask_copy(&mask, cpu_online_mask);
683 cpumask_clear_cpu(smp_processor_id(), &mask);
684 if (!cpumask_empty(&mask))
685 smp_cross_call(&mask, IPI_CPU_STOP);
687 /* Wait up to one second for other CPUs to stop */
688 timeout = USEC_PER_SEC;
689 while (num_online_cpus() > 1 && timeout--)
692 if (num_online_cpus() > 1)
693 pr_warning("SMP: failed to stop secondary CPUs\n");
699 int setup_profiling_timer(unsigned int multiplier)
704 #ifdef CONFIG_CPU_FREQ
706 static DEFINE_PER_CPU(unsigned long, l_p_j_ref);
707 static DEFINE_PER_CPU(unsigned long, l_p_j_ref_freq);
708 static unsigned long global_l_p_j_ref;
709 static unsigned long global_l_p_j_ref_freq;
711 static int cpufreq_callback(struct notifier_block *nb,
712 unsigned long val, void *data)
714 struct cpufreq_freqs *freq = data;
717 if (freq->flags & CPUFREQ_CONST_LOOPS)
720 if (!per_cpu(l_p_j_ref, cpu)) {
721 per_cpu(l_p_j_ref, cpu) =
722 per_cpu(cpu_data, cpu).loops_per_jiffy;
723 per_cpu(l_p_j_ref_freq, cpu) = freq->old;
724 if (!global_l_p_j_ref) {
725 global_l_p_j_ref = loops_per_jiffy;
726 global_l_p_j_ref_freq = freq->old;
730 if ((val == CPUFREQ_PRECHANGE && freq->old < freq->new) ||
731 (val == CPUFREQ_POSTCHANGE && freq->old > freq->new) ||
732 (val == CPUFREQ_RESUMECHANGE || val == CPUFREQ_SUSPENDCHANGE)) {
733 loops_per_jiffy = cpufreq_scale(global_l_p_j_ref,
734 global_l_p_j_ref_freq,
736 per_cpu(cpu_data, cpu).loops_per_jiffy =
737 cpufreq_scale(per_cpu(l_p_j_ref, cpu),
738 per_cpu(l_p_j_ref_freq, cpu),
744 static struct notifier_block cpufreq_notifier = {
745 .notifier_call = cpufreq_callback,
748 static int __init register_cpufreq_notifier(void)
750 return cpufreq_register_notifier(&cpufreq_notifier,
751 CPUFREQ_TRANSITION_NOTIFIER);
753 core_initcall(register_cpufreq_notifier);