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>
28 #include <linux/irq_work.h>
30 #include <linux/atomic.h>
32 #include <asm/cacheflush.h>
34 #include <asm/cputype.h>
35 #include <asm/exception.h>
36 #include <asm/idmap.h>
37 #include <asm/topology.h>
38 #include <asm/mmu_context.h>
39 #include <asm/pgtable.h>
40 #include <asm/pgalloc.h>
41 #include <asm/processor.h>
42 #include <asm/sections.h>
43 #include <asm/tlbflush.h>
44 #include <asm/ptrace.h>
45 #include <asm/smp_plat.h>
47 #include <asm/mach/arch.h>
50 #define CREATE_TRACE_POINTS
51 #include <trace/events/ipi.h>
54 * as from 2.5, kernels no longer have an init_tasks structure
55 * so we need some other way of telling a new secondary core
56 * where to place its SVC stack
58 struct secondary_data secondary_data;
61 * control for which core is the next to come out of the secondary
64 volatile int pen_release = -1;
77 static DECLARE_COMPLETION(cpu_running);
79 static struct smp_operations smp_ops;
81 void __init smp_set_ops(struct smp_operations *ops)
87 static unsigned long get_arch_pgd(pgd_t *pgd)
89 #ifdef CONFIG_ARM_LPAE
90 return __phys_to_pfn(virt_to_phys(pgd));
92 return virt_to_phys(pgd);
96 int __cpu_up(unsigned int cpu, struct task_struct *idle)
100 if (!smp_ops.smp_boot_secondary)
104 * We need to tell the secondary core where to find
105 * its stack and the page tables.
107 secondary_data.stack = task_stack_page(idle) + THREAD_START_SP;
108 #ifdef CONFIG_ARM_MPU
109 secondary_data.mpu_rgn_szr = mpu_rgn_info.rgns[MPU_RAM_REGION].drsr;
113 secondary_data.pgdir = virt_to_phys(idmap_pgd);
114 secondary_data.swapper_pg_dir = get_arch_pgd(swapper_pg_dir);
116 sync_cache_w(&secondary_data);
119 * Now bring the CPU into our world.
121 ret = smp_ops.smp_boot_secondary(cpu, idle);
124 * CPU was successfully started, wait for it
125 * to come online or time out.
127 wait_for_completion_timeout(&cpu_running,
128 msecs_to_jiffies(1000));
130 if (!cpu_online(cpu)) {
131 pr_crit("CPU%u: failed to come online\n", cpu);
135 pr_err("CPU%u: failed to boot: %d\n", cpu, ret);
139 memset(&secondary_data, 0, sizeof(secondary_data));
143 /* platform specific SMP operations */
144 void __init smp_init_cpus(void)
146 if (smp_ops.smp_init_cpus)
147 smp_ops.smp_init_cpus();
150 int platform_can_secondary_boot(void)
152 return !!smp_ops.smp_boot_secondary;
155 int platform_can_cpu_hotplug(void)
157 #ifdef CONFIG_HOTPLUG_CPU
158 if (smp_ops.cpu_kill)
165 #ifdef CONFIG_HOTPLUG_CPU
166 static int platform_cpu_kill(unsigned int cpu)
168 if (smp_ops.cpu_kill)
169 return smp_ops.cpu_kill(cpu);
173 static int platform_cpu_disable(unsigned int cpu)
175 if (smp_ops.cpu_disable)
176 return smp_ops.cpu_disable(cpu);
181 int platform_can_hotplug_cpu(unsigned int cpu)
183 /* cpu_die must be specified to support hotplug */
184 if (!smp_ops.cpu_die)
187 if (smp_ops.cpu_can_disable)
188 return smp_ops.cpu_can_disable(cpu);
191 * By default, allow disabling all CPUs except the first one,
192 * since this is special on a lot of platforms, e.g. because
193 * of clock tick interrupts.
199 * __cpu_disable runs on the processor to be shutdown.
201 int __cpu_disable(void)
203 unsigned int cpu = smp_processor_id();
206 ret = platform_cpu_disable(cpu);
211 * Take this CPU offline. Once we clear this, we can't return,
212 * and we must not schedule until we're ready to give up the cpu.
214 set_cpu_online(cpu, false);
217 * OK - migrate IRQs away from this CPU
222 * Flush user cache and TLB mappings, and then remove this CPU
223 * from the vm mask set of all processes.
225 * Caches are flushed to the Level of Unification Inner Shareable
226 * to write-back dirty lines to unified caches shared by all CPUs.
229 local_flush_tlb_all();
231 clear_tasks_mm_cpumask(cpu);
236 static DECLARE_COMPLETION(cpu_died);
239 * called on the thread which is asking for a CPU to be shutdown -
240 * waits until shutdown has completed, or it is timed out.
242 void __cpu_die(unsigned int cpu)
244 if (!wait_for_completion_timeout(&cpu_died, msecs_to_jiffies(5000))) {
245 pr_err("CPU%u: cpu didn't die\n", cpu);
248 pr_notice("CPU%u: shutdown\n", cpu);
251 * platform_cpu_kill() is generally expected to do the powering off
252 * and/or cutting of clocks to the dying CPU. Optionally, this may
253 * be done by the CPU which is dying in preference to supporting
254 * this call, but that means there is _no_ synchronisation between
255 * the requesting CPU and the dying CPU actually losing power.
257 if (!platform_cpu_kill(cpu))
258 pr_err("CPU%u: unable to kill\n", cpu);
262 * Called from the idle thread for the CPU which has been shutdown.
264 * Note that we disable IRQs here, but do not re-enable them
265 * before returning to the caller. This is also the behaviour
266 * of the other hotplug-cpu capable cores, so presumably coming
267 * out of idle fixes this.
269 void arch_cpu_idle_dead(void)
271 unsigned int cpu = smp_processor_id();
278 * Flush the data out of the L1 cache for this CPU. This must be
279 * before the completion to ensure that data is safely written out
280 * before platform_cpu_kill() gets called - which may disable
281 * *this* CPU and power down its cache.
286 * Tell __cpu_die() that this CPU is now safe to dispose of. Once
287 * this returns, power and/or clocks can be removed at any point
288 * from this CPU and its cache by platform_cpu_kill().
293 * Ensure that the cache lines associated with that completion are
294 * written out. This covers the case where _this_ CPU is doing the
295 * powering down, to ensure that the completion is visible to the
296 * CPU waiting for this one.
301 * The actual CPU shutdown procedure is at least platform (if not
302 * CPU) specific. This may remove power, or it may simply spin.
304 * Platforms are generally expected *NOT* to return from this call,
305 * although there are some which do because they have no way to
306 * power down the CPU. These platforms are the _only_ reason we
307 * have a return path which uses the fragment of assembly below.
309 * The return path should not be used for platforms which can
313 smp_ops.cpu_die(cpu);
315 pr_warn("CPU%u: smp_ops.cpu_die() returned, trying to resuscitate\n",
319 * Do not return to the idle loop - jump back to the secondary
320 * cpu initialisation. There's some initialisation which needs
321 * to be repeated to undo the effects of taking the CPU offline.
323 __asm__("mov sp, %0\n"
325 " b secondary_start_kernel"
327 : "r" (task_stack_page(current) + THREAD_SIZE - 8));
329 #endif /* CONFIG_HOTPLUG_CPU */
332 * Called by both boot and secondaries to move global data into
333 * per-processor storage.
335 static void smp_store_cpu_info(unsigned int cpuid)
337 struct cpuinfo_arm *cpu_info = &per_cpu(cpu_data, cpuid);
339 cpu_info->loops_per_jiffy = loops_per_jiffy;
340 cpu_info->cpuid = read_cpuid_id();
342 store_cpu_topology(cpuid);
346 * This is the secondary CPU boot entry. We're using this CPUs
347 * idle thread stack, but a set of temporary page tables.
349 asmlinkage void secondary_start_kernel(void)
351 struct mm_struct *mm = &init_mm;
355 * The identity mapping is uncached (strongly ordered), so
356 * switch away from it before attempting any exclusive accesses.
358 cpu_switch_mm(mm->pgd, mm);
359 local_flush_bp_all();
360 enter_lazy_tlb(mm, current);
361 local_flush_tlb_all();
364 * All kernel threads share the same mm context; grab a
365 * reference and switch to it.
367 cpu = smp_processor_id();
368 atomic_inc(&mm->mm_count);
369 current->active_mm = mm;
370 cpumask_set_cpu(cpu, mm_cpumask(mm));
374 pr_debug("CPU%u: Booted secondary processor\n", cpu);
377 trace_hardirqs_off();
380 * Give the platform a chance to do its own initialisation.
382 if (smp_ops.smp_secondary_init)
383 smp_ops.smp_secondary_init(cpu);
385 notify_cpu_starting(cpu);
389 smp_store_cpu_info(cpu);
392 * OK, now it's safe to let the boot CPU continue. Wait for
393 * the CPU migration code to notice that the CPU is online
394 * before we continue - which happens after __cpu_up returns.
396 set_cpu_online(cpu, true);
397 complete(&cpu_running);
403 * OK, it's off to the idle thread for us
405 cpu_startup_entry(CPUHP_ONLINE);
408 void __init smp_cpus_done(unsigned int max_cpus)
411 unsigned long bogosum = 0;
413 for_each_online_cpu(cpu)
414 bogosum += per_cpu(cpu_data, cpu).loops_per_jiffy;
416 printk(KERN_INFO "SMP: Total of %d processors activated "
417 "(%lu.%02lu BogoMIPS).\n",
419 bogosum / (500000/HZ),
420 (bogosum / (5000/HZ)) % 100);
425 void __init smp_prepare_boot_cpu(void)
427 set_my_cpu_offset(per_cpu_offset(smp_processor_id()));
430 void __init smp_prepare_cpus(unsigned int max_cpus)
432 unsigned int ncores = num_possible_cpus();
436 smp_store_cpu_info(smp_processor_id());
439 * are we trying to boot more cores than exist?
441 if (max_cpus > ncores)
443 if (ncores > 1 && max_cpus) {
445 * Initialise the present map, which describes the set of CPUs
446 * actually populated at the present time. A platform should
447 * re-initialize the map in the platforms smp_prepare_cpus()
448 * if present != possible (e.g. physical hotplug).
450 init_cpu_present(cpu_possible_mask);
453 * Initialise the SCU if there are more than one CPU
454 * and let them know where to start.
456 if (smp_ops.smp_prepare_cpus)
457 smp_ops.smp_prepare_cpus(max_cpus);
461 static void (*__smp_cross_call)(const struct cpumask *, unsigned int);
463 void __init set_smp_cross_call(void (*fn)(const struct cpumask *, unsigned int))
465 if (!__smp_cross_call)
466 __smp_cross_call = fn;
469 static const char *ipi_types[NR_IPI] __tracepoint_string = {
470 #define S(x,s) [x] = s
471 S(IPI_WAKEUP, "CPU wakeup interrupts"),
472 S(IPI_TIMER, "Timer broadcast interrupts"),
473 S(IPI_RESCHEDULE, "Rescheduling interrupts"),
474 S(IPI_CALL_FUNC, "Function call interrupts"),
475 S(IPI_CALL_FUNC_SINGLE, "Single function call interrupts"),
476 S(IPI_CPU_STOP, "CPU stop interrupts"),
477 S(IPI_IRQ_WORK, "IRQ work interrupts"),
478 S(IPI_COMPLETION, "completion interrupts"),
481 static void smp_cross_call(const struct cpumask *target, unsigned int ipinr)
483 trace_ipi_raise(target, ipi_types[ipinr]);
484 __smp_cross_call(target, ipinr);
487 void show_ipi_list(struct seq_file *p, int prec)
491 for (i = 0; i < NR_IPI; i++) {
492 seq_printf(p, "%*s%u: ", prec - 1, "IPI", i);
494 for_each_online_cpu(cpu)
495 seq_printf(p, "%10u ",
496 __get_irq_stat(cpu, ipi_irqs[i]));
498 seq_printf(p, " %s\n", ipi_types[i]);
502 u64 smp_irq_stat_cpu(unsigned int cpu)
507 for (i = 0; i < NR_IPI; i++)
508 sum += __get_irq_stat(cpu, ipi_irqs[i]);
513 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
515 smp_cross_call(mask, IPI_CALL_FUNC);
518 void arch_send_wakeup_ipi_mask(const struct cpumask *mask)
520 smp_cross_call(mask, IPI_WAKEUP);
523 void arch_send_call_function_single_ipi(int cpu)
525 smp_cross_call(cpumask_of(cpu), IPI_CALL_FUNC_SINGLE);
528 #ifdef CONFIG_IRQ_WORK
529 void arch_irq_work_raise(void)
531 if (arch_irq_work_has_interrupt())
532 smp_cross_call(cpumask_of(smp_processor_id()), IPI_IRQ_WORK);
536 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
537 void tick_broadcast(const struct cpumask *mask)
539 smp_cross_call(mask, IPI_TIMER);
543 static DEFINE_RAW_SPINLOCK(stop_lock);
546 * ipi_cpu_stop - handle IPI from smp_send_stop()
548 static void ipi_cpu_stop(unsigned int cpu)
550 if (system_state == SYSTEM_BOOTING ||
551 system_state == SYSTEM_RUNNING) {
552 raw_spin_lock(&stop_lock);
553 pr_crit("CPU%u: stopping\n", cpu);
555 raw_spin_unlock(&stop_lock);
558 set_cpu_online(cpu, false);
567 static DEFINE_PER_CPU(struct completion *, cpu_completion);
569 int register_ipi_completion(struct completion *completion, int cpu)
571 per_cpu(cpu_completion, cpu) = completion;
572 return IPI_COMPLETION;
575 static void ipi_complete(unsigned int cpu)
577 complete(per_cpu(cpu_completion, cpu));
581 * Main handler for inter-processor interrupts
583 asmlinkage void __exception_irq_entry do_IPI(int ipinr, struct pt_regs *regs)
585 handle_IPI(ipinr, regs);
588 void handle_IPI(int ipinr, struct pt_regs *regs)
590 unsigned int cpu = smp_processor_id();
591 struct pt_regs *old_regs = set_irq_regs(regs);
593 if ((unsigned)ipinr < NR_IPI) {
594 trace_ipi_entry(ipi_types[ipinr]);
595 __inc_irq_stat(cpu, ipi_irqs[ipinr]);
602 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
605 tick_receive_broadcast();
616 generic_smp_call_function_interrupt();
620 case IPI_CALL_FUNC_SINGLE:
622 generic_smp_call_function_single_interrupt();
632 #ifdef CONFIG_IRQ_WORK
647 pr_crit("CPU%u: Unknown IPI message 0x%x\n",
652 if ((unsigned)ipinr < NR_IPI)
653 trace_ipi_exit(ipi_types[ipinr]);
654 set_irq_regs(old_regs);
657 void smp_send_reschedule(int cpu)
659 smp_cross_call(cpumask_of(cpu), IPI_RESCHEDULE);
662 void smp_send_stop(void)
664 unsigned long timeout;
667 cpumask_copy(&mask, cpu_online_mask);
668 cpumask_clear_cpu(smp_processor_id(), &mask);
669 if (!cpumask_empty(&mask))
670 smp_cross_call(&mask, IPI_CPU_STOP);
672 /* Wait up to one second for other CPUs to stop */
673 timeout = USEC_PER_SEC;
674 while (num_online_cpus() > 1 && timeout--)
677 if (num_online_cpus() > 1)
678 pr_warn("SMP: failed to stop secondary CPUs\n");
684 int setup_profiling_timer(unsigned int multiplier)
689 #ifdef CONFIG_CPU_FREQ
691 static DEFINE_PER_CPU(unsigned long, l_p_j_ref);
692 static DEFINE_PER_CPU(unsigned long, l_p_j_ref_freq);
693 static unsigned long global_l_p_j_ref;
694 static unsigned long global_l_p_j_ref_freq;
696 static int cpufreq_callback(struct notifier_block *nb,
697 unsigned long val, void *data)
699 struct cpufreq_freqs *freq = data;
702 if (freq->flags & CPUFREQ_CONST_LOOPS)
705 if (!per_cpu(l_p_j_ref, cpu)) {
706 per_cpu(l_p_j_ref, cpu) =
707 per_cpu(cpu_data, cpu).loops_per_jiffy;
708 per_cpu(l_p_j_ref_freq, cpu) = freq->old;
709 if (!global_l_p_j_ref) {
710 global_l_p_j_ref = loops_per_jiffy;
711 global_l_p_j_ref_freq = freq->old;
715 if ((val == CPUFREQ_PRECHANGE && freq->old < freq->new) ||
716 (val == CPUFREQ_POSTCHANGE && freq->old > freq->new)) {
717 loops_per_jiffy = cpufreq_scale(global_l_p_j_ref,
718 global_l_p_j_ref_freq,
720 per_cpu(cpu_data, cpu).loops_per_jiffy =
721 cpufreq_scale(per_cpu(l_p_j_ref, cpu),
722 per_cpu(l_p_j_ref_freq, cpu),
728 static struct notifier_block cpufreq_notifier = {
729 .notifier_call = cpufreq_callback,
732 static int __init register_cpufreq_notifier(void)
734 return cpufreq_register_notifier(&cpufreq_notifier,
735 CPUFREQ_TRANSITION_NOTIFIER);
737 core_initcall(register_cpufreq_notifier);