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>
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;
74 static DECLARE_COMPLETION(cpu_running);
76 static struct smp_operations smp_ops;
78 void __init smp_set_ops(struct smp_operations *ops)
84 static unsigned long get_arch_pgd(pgd_t *pgd)
86 phys_addr_t pgdir = virt_to_idmap(pgd);
87 BUG_ON(pgdir & ARCH_PGD_MASK);
88 return pgdir >> ARCH_PGD_SHIFT;
91 int __cpu_up(unsigned int cpu, struct task_struct *idle)
96 * We need to tell the secondary core where to find
97 * its stack and the page tables.
99 secondary_data.stack = task_stack_page(idle) + THREAD_START_SP;
100 #ifdef CONFIG_ARM_MPU
101 secondary_data.mpu_rgn_szr = mpu_rgn_info.rgns[MPU_RAM_REGION].drsr;
105 secondary_data.pgdir = get_arch_pgd(idmap_pgd);
106 secondary_data.swapper_pg_dir = get_arch_pgd(swapper_pg_dir);
108 sync_cache_w(&secondary_data);
111 * Now bring the CPU into our world.
113 ret = boot_secondary(cpu, idle);
116 * CPU was successfully started, wait for it
117 * to come online or time out.
119 wait_for_completion_timeout(&cpu_running,
120 msecs_to_jiffies(1000));
122 if (!cpu_online(cpu)) {
123 pr_crit("CPU%u: failed to come online\n", cpu);
127 pr_err("CPU%u: failed to boot: %d\n", cpu, ret);
131 memset(&secondary_data, 0, sizeof(secondary_data));
135 /* platform specific SMP operations */
136 void __init smp_init_cpus(void)
138 if (smp_ops.smp_init_cpus)
139 smp_ops.smp_init_cpus();
142 int boot_secondary(unsigned int cpu, struct task_struct *idle)
144 if (smp_ops.smp_boot_secondary)
145 return smp_ops.smp_boot_secondary(cpu, idle);
149 int platform_can_cpu_hotplug(void)
151 #ifdef CONFIG_HOTPLUG_CPU
152 if (smp_ops.cpu_kill)
159 #ifdef CONFIG_HOTPLUG_CPU
160 static int platform_cpu_kill(unsigned int cpu)
162 if (smp_ops.cpu_kill)
163 return smp_ops.cpu_kill(cpu);
167 static int platform_cpu_disable(unsigned int cpu)
169 if (smp_ops.cpu_disable)
170 return smp_ops.cpu_disable(cpu);
173 * By default, allow disabling all CPUs except the first one,
174 * since this is special on a lot of platforms, e.g. because
175 * of clock tick interrupts.
177 return cpu == 0 ? -EPERM : 0;
180 * __cpu_disable runs on the processor to be shutdown.
182 int __cpu_disable(void)
184 unsigned int cpu = smp_processor_id();
187 ret = platform_cpu_disable(cpu);
192 * Take this CPU offline. Once we clear this, we can't return,
193 * and we must not schedule until we're ready to give up the cpu.
195 set_cpu_online(cpu, false);
198 * OK - migrate IRQs away from this CPU
203 * Flush user cache and TLB mappings, and then remove this CPU
204 * from the vm mask set of all processes.
206 * Caches are flushed to the Level of Unification Inner Shareable
207 * to write-back dirty lines to unified caches shared by all CPUs.
210 local_flush_tlb_all();
212 clear_tasks_mm_cpumask(cpu);
217 static DECLARE_COMPLETION(cpu_died);
220 * called on the thread which is asking for a CPU to be shutdown -
221 * waits until shutdown has completed, or it is timed out.
223 void __cpu_die(unsigned int cpu)
225 if (!wait_for_completion_timeout(&cpu_died, msecs_to_jiffies(5000))) {
226 pr_err("CPU%u: cpu didn't die\n", cpu);
229 printk(KERN_NOTICE "CPU%u: shutdown\n", cpu);
232 * platform_cpu_kill() is generally expected to do the powering off
233 * and/or cutting of clocks to the dying CPU. Optionally, this may
234 * be done by the CPU which is dying in preference to supporting
235 * this call, but that means there is _no_ synchronisation between
236 * the requesting CPU and the dying CPU actually losing power.
238 if (!platform_cpu_kill(cpu))
239 printk("CPU%u: unable to kill\n", cpu);
243 * Called from the idle thread for the CPU which has been shutdown.
245 * Note that we disable IRQs here, but do not re-enable them
246 * before returning to the caller. This is also the behaviour
247 * of the other hotplug-cpu capable cores, so presumably coming
248 * out of idle fixes this.
250 void __ref cpu_die(void)
252 unsigned int cpu = smp_processor_id();
259 * Flush the data out of the L1 cache for this CPU. This must be
260 * before the completion to ensure that data is safely written out
261 * before platform_cpu_kill() gets called - which may disable
262 * *this* CPU and power down its cache.
267 * Tell __cpu_die() that this CPU is now safe to dispose of. Once
268 * this returns, power and/or clocks can be removed at any point
269 * from this CPU and its cache by platform_cpu_kill().
274 * Ensure that the cache lines associated with that completion are
275 * written out. This covers the case where _this_ CPU is doing the
276 * powering down, to ensure that the completion is visible to the
277 * CPU waiting for this one.
282 * The actual CPU shutdown procedure is at least platform (if not
283 * CPU) specific. This may remove power, or it may simply spin.
285 * Platforms are generally expected *NOT* to return from this call,
286 * although there are some which do because they have no way to
287 * power down the CPU. These platforms are the _only_ reason we
288 * have a return path which uses the fragment of assembly below.
290 * The return path should not be used for platforms which can
294 smp_ops.cpu_die(cpu);
297 * Do not return to the idle loop - jump back to the secondary
298 * cpu initialisation. There's some initialisation which needs
299 * to be repeated to undo the effects of taking the CPU offline.
301 __asm__("mov sp, %0\n"
303 " b secondary_start_kernel"
305 : "r" (task_stack_page(current) + THREAD_SIZE - 8));
307 #endif /* CONFIG_HOTPLUG_CPU */
310 * Called by both boot and secondaries to move global data into
311 * per-processor storage.
313 static void smp_store_cpu_info(unsigned int cpuid)
315 struct cpuinfo_arm *cpu_info = &per_cpu(cpu_data, cpuid);
317 cpu_info->loops_per_jiffy = loops_per_jiffy;
318 cpu_info->cpuid = read_cpuid_id();
320 store_cpu_topology(cpuid);
324 * This is the secondary CPU boot entry. We're using this CPUs
325 * idle thread stack, but a set of temporary page tables.
327 asmlinkage void secondary_start_kernel(void)
329 struct mm_struct *mm = &init_mm;
333 * The identity mapping is uncached (strongly ordered), so
334 * switch away from it before attempting any exclusive accesses.
336 cpu_switch_mm(mm->pgd, mm);
337 local_flush_bp_all();
338 enter_lazy_tlb(mm, current);
339 local_flush_tlb_all();
342 * All kernel threads share the same mm context; grab a
343 * reference and switch to it.
345 cpu = smp_processor_id();
346 atomic_inc(&mm->mm_count);
347 current->active_mm = mm;
348 cpumask_set_cpu(cpu, mm_cpumask(mm));
352 printk("CPU%u: Booted secondary processor\n", cpu);
355 trace_hardirqs_off();
358 * Give the platform a chance to do its own initialisation.
360 if (smp_ops.smp_secondary_init)
361 smp_ops.smp_secondary_init(cpu);
363 notify_cpu_starting(cpu);
367 smp_store_cpu_info(cpu);
370 * OK, now it's safe to let the boot CPU continue. Wait for
371 * the CPU migration code to notice that the CPU is online
372 * before we continue - which happens after __cpu_up returns.
374 set_cpu_online(cpu, true);
375 complete(&cpu_running);
381 * OK, it's off to the idle thread for us
383 cpu_startup_entry(CPUHP_ONLINE);
386 void __init smp_cpus_done(unsigned int max_cpus)
388 printk(KERN_INFO "SMP: Total of %d processors activated.\n",
394 void __init smp_prepare_boot_cpu(void)
396 set_my_cpu_offset(per_cpu_offset(smp_processor_id()));
399 void __init smp_prepare_cpus(unsigned int max_cpus)
401 unsigned int ncores = num_possible_cpus();
405 smp_store_cpu_info(smp_processor_id());
408 * are we trying to boot more cores than exist?
410 if (max_cpus > ncores)
412 if (ncores > 1 && max_cpus) {
414 * Initialise the present map, which describes the set of CPUs
415 * actually populated at the present time. A platform should
416 * re-initialize the map in the platforms smp_prepare_cpus()
417 * if present != possible (e.g. physical hotplug).
419 init_cpu_present(cpu_possible_mask);
422 * Initialise the SCU if there are more than one CPU
423 * and let them know where to start.
425 if (smp_ops.smp_prepare_cpus)
426 smp_ops.smp_prepare_cpus(max_cpus);
430 static void (*smp_cross_call)(const struct cpumask *, unsigned int);
432 void __init set_smp_cross_call(void (*fn)(const struct cpumask *, unsigned int))
438 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
440 smp_cross_call(mask, IPI_CALL_FUNC);
443 void arch_send_wakeup_ipi_mask(const struct cpumask *mask)
445 smp_cross_call(mask, IPI_WAKEUP);
448 void arch_send_call_function_single_ipi(int cpu)
450 smp_cross_call(cpumask_of(cpu), IPI_CALL_FUNC_SINGLE);
453 #ifdef CONFIG_IRQ_WORK
454 void arch_irq_work_raise(void)
457 smp_cross_call(cpumask_of(smp_processor_id()), IPI_IRQ_WORK);
461 static const char *ipi_types[NR_IPI] = {
462 #define S(x,s) [x] = s
463 S(IPI_WAKEUP, "CPU wakeup interrupts"),
464 S(IPI_TIMER, "Timer broadcast interrupts"),
465 S(IPI_RESCHEDULE, "Rescheduling interrupts"),
466 S(IPI_CALL_FUNC, "Function call interrupts"),
467 S(IPI_CALL_FUNC_SINGLE, "Single function call interrupts"),
468 S(IPI_CPU_STOP, "CPU stop interrupts"),
469 S(IPI_IRQ_WORK, "IRQ work interrupts"),
470 S(IPI_COMPLETION, "completion interrupts"),
473 void show_ipi_list(struct seq_file *p, int prec)
477 for (i = 0; i < NR_IPI; i++) {
478 seq_printf(p, "%*s%u: ", prec - 1, "IPI", i);
480 for_each_online_cpu(cpu)
481 seq_printf(p, "%10u ",
482 __get_irq_stat(cpu, ipi_irqs[i]));
484 seq_printf(p, " %s\n", ipi_types[i]);
488 u64 smp_irq_stat_cpu(unsigned int cpu)
493 for (i = 0; i < NR_IPI; i++)
494 sum += __get_irq_stat(cpu, ipi_irqs[i]);
499 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
500 void tick_broadcast(const struct cpumask *mask)
502 smp_cross_call(mask, IPI_TIMER);
506 static DEFINE_RAW_SPINLOCK(stop_lock);
509 * ipi_cpu_stop - handle IPI from smp_send_stop()
511 static void ipi_cpu_stop(unsigned int cpu)
513 if (system_state == SYSTEM_BOOTING ||
514 system_state == SYSTEM_RUNNING) {
515 raw_spin_lock(&stop_lock);
516 printk(KERN_CRIT "CPU%u: stopping\n", cpu);
518 raw_spin_unlock(&stop_lock);
521 set_cpu_online(cpu, false);
530 static DEFINE_PER_CPU(struct completion *, cpu_completion);
532 int register_ipi_completion(struct completion *completion, int cpu)
534 per_cpu(cpu_completion, cpu) = completion;
535 return IPI_COMPLETION;
538 static void ipi_complete(unsigned int cpu)
540 complete(per_cpu(cpu_completion, cpu));
544 * Main handler for inter-processor interrupts
546 asmlinkage void __exception_irq_entry do_IPI(int ipinr, struct pt_regs *regs)
548 handle_IPI(ipinr, regs);
551 void handle_IPI(int ipinr, struct pt_regs *regs)
553 unsigned int cpu = smp_processor_id();
554 struct pt_regs *old_regs = set_irq_regs(regs);
557 __inc_irq_stat(cpu, ipi_irqs[ipinr]);
563 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
566 tick_receive_broadcast();
577 generic_smp_call_function_interrupt();
581 case IPI_CALL_FUNC_SINGLE:
583 generic_smp_call_function_single_interrupt();
593 #ifdef CONFIG_IRQ_WORK
608 printk(KERN_CRIT "CPU%u: Unknown IPI message 0x%x\n",
612 set_irq_regs(old_regs);
615 void smp_send_reschedule(int cpu)
617 smp_cross_call(cpumask_of(cpu), IPI_RESCHEDULE);
620 void smp_send_stop(void)
622 unsigned long timeout;
625 cpumask_copy(&mask, cpu_online_mask);
626 cpumask_clear_cpu(smp_processor_id(), &mask);
627 if (!cpumask_empty(&mask))
628 smp_cross_call(&mask, IPI_CPU_STOP);
630 /* Wait up to one second for other CPUs to stop */
631 timeout = USEC_PER_SEC;
632 while (num_online_cpus() > 1 && timeout--)
635 if (num_online_cpus() > 1)
636 pr_warning("SMP: failed to stop secondary CPUs\n");
642 int setup_profiling_timer(unsigned int multiplier)
647 #ifdef CONFIG_CPU_FREQ
649 static DEFINE_PER_CPU(unsigned long, l_p_j_ref);
650 static DEFINE_PER_CPU(unsigned long, l_p_j_ref_freq);
651 static unsigned long global_l_p_j_ref;
652 static unsigned long global_l_p_j_ref_freq;
654 static int cpufreq_callback(struct notifier_block *nb,
655 unsigned long val, void *data)
657 struct cpufreq_freqs *freq = data;
660 if (freq->flags & CPUFREQ_CONST_LOOPS)
663 if (!per_cpu(l_p_j_ref, cpu)) {
664 per_cpu(l_p_j_ref, cpu) =
665 per_cpu(cpu_data, cpu).loops_per_jiffy;
666 per_cpu(l_p_j_ref_freq, cpu) = freq->old;
667 if (!global_l_p_j_ref) {
668 global_l_p_j_ref = loops_per_jiffy;
669 global_l_p_j_ref_freq = freq->old;
673 if ((val == CPUFREQ_PRECHANGE && freq->old < freq->new) ||
674 (val == CPUFREQ_POSTCHANGE && freq->old > freq->new) ||
675 (val == CPUFREQ_RESUMECHANGE || val == CPUFREQ_SUSPENDCHANGE)) {
676 loops_per_jiffy = cpufreq_scale(global_l_p_j_ref,
677 global_l_p_j_ref_freq,
679 per_cpu(cpu_data, cpu).loops_per_jiffy =
680 cpufreq_scale(per_cpu(l_p_j_ref, cpu),
681 per_cpu(l_p_j_ref_freq, cpu),
687 static struct notifier_block cpufreq_notifier = {
688 .notifier_call = cpufreq_callback,
691 static int __init register_cpufreq_notifier(void)
693 return cpufreq_register_notifier(&cpufreq_notifier,
694 CPUFREQ_TRANSITION_NOTIFIER);
696 core_initcall(register_cpufreq_notifier);