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/mm.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>
29
30#include <linux/atomic.h>
31#include <asm/smp.h>
32#include <asm/cacheflush.h>
33#include <asm/cpu.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>
46#include <asm/virt.h>
47#include <asm/mach/arch.h>
48#include <asm/mpu.h>
49
50#define CREATE_TRACE_POINTS
51#include <trace/events/ipi.h>
52
53/*
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
57 */
58struct secondary_data secondary_data;
59
60/*
61 * control for which core is the next to come out of the secondary
62 * boot "holding pen"
63 */
64volatile int pen_release = -1;
65
66enum ipi_msg_type {
67	IPI_WAKEUP,
68	IPI_TIMER,
69	IPI_RESCHEDULE,
70	IPI_CALL_FUNC,
71	IPI_CALL_FUNC_SINGLE,
72	IPI_CPU_STOP,
73	IPI_IRQ_WORK,
74	IPI_COMPLETION,
75};
76
77static DECLARE_COMPLETION(cpu_running);
78
79static struct smp_operations smp_ops;
80
81void __init smp_set_ops(struct smp_operations *ops)
82{
83	if (ops)
84		smp_ops = *ops;
85};
86
87static unsigned long get_arch_pgd(pgd_t *pgd)
88{
89	phys_addr_t pgdir = virt_to_idmap(pgd);
90	BUG_ON(pgdir & ARCH_PGD_MASK);
91	return pgdir >> ARCH_PGD_SHIFT;
92}
93
94int __cpu_up(unsigned int cpu, struct task_struct *idle)
95{
96	int ret;
97
98	if (!smp_ops.smp_boot_secondary)
99		return -ENOSYS;
100
101	/*
102	 * We need to tell the secondary core where to find
103	 * its stack and the page tables.
104	 */
105	secondary_data.stack = task_stack_page(idle) + THREAD_START_SP;
106#ifdef CONFIG_ARM_MPU
107	secondary_data.mpu_rgn_szr = mpu_rgn_info.rgns[MPU_RAM_REGION].drsr;
108#endif
109
110#ifdef CONFIG_MMU
111	secondary_data.pgdir = get_arch_pgd(idmap_pgd);
112	secondary_data.swapper_pg_dir = get_arch_pgd(swapper_pg_dir);
113#endif
114	sync_cache_w(&secondary_data);
115
116	/*
117	 * Now bring the CPU into our world.
118	 */
119	ret = smp_ops.smp_boot_secondary(cpu, idle);
120	if (ret == 0) {
121		/*
122		 * CPU was successfully started, wait for it
123		 * to come online or time out.
124		 */
125		wait_for_completion_timeout(&cpu_running,
126						 msecs_to_jiffies(1000));
127
128		if (!cpu_online(cpu)) {
129			pr_crit("CPU%u: failed to come online\n", cpu);
130			ret = -EIO;
131		}
132	} else {
133		pr_err("CPU%u: failed to boot: %d\n", cpu, ret);
134	}
135
136
137	memset(&secondary_data, 0, sizeof(secondary_data));
138	return ret;
139}
140
141/* platform specific SMP operations */
142void __init smp_init_cpus(void)
143{
144	if (smp_ops.smp_init_cpus)
145		smp_ops.smp_init_cpus();
146}
147
148int platform_can_secondary_boot(void)
149{
150	return !!smp_ops.smp_boot_secondary;
151}
152
153int platform_can_cpu_hotplug(void)
154{
155#ifdef CONFIG_HOTPLUG_CPU
156	if (smp_ops.cpu_kill)
157		return 1;
158#endif
159
160	return 0;
161}
162
163#ifdef CONFIG_HOTPLUG_CPU
164static int platform_cpu_kill(unsigned int cpu)
165{
166	if (smp_ops.cpu_kill)
167		return smp_ops.cpu_kill(cpu);
168	return 1;
169}
170
171static int platform_cpu_disable(unsigned int cpu)
172{
173	if (smp_ops.cpu_disable)
174		return smp_ops.cpu_disable(cpu);
175
176	/*
177	 * By default, allow disabling all CPUs except the first one,
178	 * since this is special on a lot of platforms, e.g. because
179	 * of clock tick interrupts.
180	 */
181	return cpu == 0 ? -EPERM : 0;
182}
183/*
184 * __cpu_disable runs on the processor to be shutdown.
185 */
186int __cpu_disable(void)
187{
188	unsigned int cpu = smp_processor_id();
189	int ret;
190
191	ret = platform_cpu_disable(cpu);
192	if (ret)
193		return ret;
194
195	/*
196	 * Take this CPU offline.  Once we clear this, we can't return,
197	 * and we must not schedule until we're ready to give up the cpu.
198	 */
199	set_cpu_online(cpu, false);
200
201	/*
202	 * OK - migrate IRQs away from this CPU
203	 */
204	migrate_irqs();
205
206	/*
207	 * Flush user cache and TLB mappings, and then remove this CPU
208	 * from the vm mask set of all processes.
209	 *
210	 * Caches are flushed to the Level of Unification Inner Shareable
211	 * to write-back dirty lines to unified caches shared by all CPUs.
212	 */
213	flush_cache_louis();
214	local_flush_tlb_all();
215
216	clear_tasks_mm_cpumask(cpu);
217
218	return 0;
219}
220
221static DECLARE_COMPLETION(cpu_died);
222
223/*
224 * called on the thread which is asking for a CPU to be shutdown -
225 * waits until shutdown has completed, or it is timed out.
226 */
227void __cpu_die(unsigned int cpu)
228{
229	if (!wait_for_completion_timeout(&cpu_died, msecs_to_jiffies(5000))) {
230		pr_err("CPU%u: cpu didn't die\n", cpu);
231		return;
232	}
233	pr_notice("CPU%u: shutdown\n", cpu);
234
235	/*
236	 * platform_cpu_kill() is generally expected to do the powering off
237	 * and/or cutting of clocks to the dying CPU.  Optionally, this may
238	 * be done by the CPU which is dying in preference to supporting
239	 * this call, but that means there is _no_ synchronisation between
240	 * the requesting CPU and the dying CPU actually losing power.
241	 */
242	if (!platform_cpu_kill(cpu))
243		pr_err("CPU%u: unable to kill\n", cpu);
244}
245
246/*
247 * Called from the idle thread for the CPU which has been shutdown.
248 *
249 * Note that we disable IRQs here, but do not re-enable them
250 * before returning to the caller. This is also the behaviour
251 * of the other hotplug-cpu capable cores, so presumably coming
252 * out of idle fixes this.
253 */
254void __ref cpu_die(void)
255{
256	unsigned int cpu = smp_processor_id();
257
258	idle_task_exit();
259
260	local_irq_disable();
261
262	/*
263	 * Flush the data out of the L1 cache for this CPU.  This must be
264	 * before the completion to ensure that data is safely written out
265	 * before platform_cpu_kill() gets called - which may disable
266	 * *this* CPU and power down its cache.
267	 */
268	flush_cache_louis();
269
270	/*
271	 * Tell __cpu_die() that this CPU is now safe to dispose of.  Once
272	 * this returns, power and/or clocks can be removed at any point
273	 * from this CPU and its cache by platform_cpu_kill().
274	 */
275	complete(&cpu_died);
276
277	/*
278	 * Ensure that the cache lines associated with that completion are
279	 * written out.  This covers the case where _this_ CPU is doing the
280	 * powering down, to ensure that the completion is visible to the
281	 * CPU waiting for this one.
282	 */
283	flush_cache_louis();
284
285	/*
286	 * The actual CPU shutdown procedure is at least platform (if not
287	 * CPU) specific.  This may remove power, or it may simply spin.
288	 *
289	 * Platforms are generally expected *NOT* to return from this call,
290	 * although there are some which do because they have no way to
291	 * power down the CPU.  These platforms are the _only_ reason we
292	 * have a return path which uses the fragment of assembly below.
293	 *
294	 * The return path should not be used for platforms which can
295	 * power off the CPU.
296	 */
297	if (smp_ops.cpu_die)
298		smp_ops.cpu_die(cpu);
299
300	pr_warn("CPU%u: smp_ops.cpu_die() returned, trying to resuscitate\n",
301		cpu);
302
303	/*
304	 * Do not return to the idle loop - jump back to the secondary
305	 * cpu initialisation.  There's some initialisation which needs
306	 * to be repeated to undo the effects of taking the CPU offline.
307	 */
308	__asm__("mov	sp, %0\n"
309	"	mov	fp, #0\n"
310	"	b	secondary_start_kernel"
311		:
312		: "r" (task_stack_page(current) + THREAD_SIZE - 8));
313}
314#endif /* CONFIG_HOTPLUG_CPU */
315
316/*
317 * Called by both boot and secondaries to move global data into
318 * per-processor storage.
319 */
320static void smp_store_cpu_info(unsigned int cpuid)
321{
322	struct cpuinfo_arm *cpu_info = &per_cpu(cpu_data, cpuid);
323
324	cpu_info->loops_per_jiffy = loops_per_jiffy;
325	cpu_info->cpuid = read_cpuid_id();
326
327	store_cpu_topology(cpuid);
328}
329
330/*
331 * This is the secondary CPU boot entry.  We're using this CPUs
332 * idle thread stack, but a set of temporary page tables.
333 */
334asmlinkage void secondary_start_kernel(void)
335{
336	struct mm_struct *mm = &init_mm;
337	unsigned int cpu;
338
339	/*
340	 * The identity mapping is uncached (strongly ordered), so
341	 * switch away from it before attempting any exclusive accesses.
342	 */
343	cpu_switch_mm(mm->pgd, mm);
344	local_flush_bp_all();
345	enter_lazy_tlb(mm, current);
346	local_flush_tlb_all();
347
348	/*
349	 * All kernel threads share the same mm context; grab a
350	 * reference and switch to it.
351	 */
352	cpu = smp_processor_id();
353	atomic_inc(&mm->mm_count);
354	current->active_mm = mm;
355	cpumask_set_cpu(cpu, mm_cpumask(mm));
356
357	cpu_init();
358
359	pr_debug("CPU%u: Booted secondary processor\n", cpu);
360
361	preempt_disable();
362	trace_hardirqs_off();
363
364	/*
365	 * Give the platform a chance to do its own initialisation.
366	 */
367	if (smp_ops.smp_secondary_init)
368		smp_ops.smp_secondary_init(cpu);
369
370	notify_cpu_starting(cpu);
371
372	calibrate_delay();
373
374	smp_store_cpu_info(cpu);
375
376	/*
377	 * OK, now it's safe to let the boot CPU continue.  Wait for
378	 * the CPU migration code to notice that the CPU is online
379	 * before we continue - which happens after __cpu_up returns.
380	 */
381	set_cpu_online(cpu, true);
382	complete(&cpu_running);
383
384	local_irq_enable();
385	local_fiq_enable();
386
387	/*
388	 * OK, it's off to the idle thread for us
389	 */
390	cpu_startup_entry(CPUHP_ONLINE);
391}
392
393void __init smp_cpus_done(unsigned int max_cpus)
394{
395	int cpu;
396	unsigned long bogosum = 0;
397
398	for_each_online_cpu(cpu)
399		bogosum += per_cpu(cpu_data, cpu).loops_per_jiffy;
400
401	printk(KERN_INFO "SMP: Total of %d processors activated "
402	       "(%lu.%02lu BogoMIPS).\n",
403	       num_online_cpus(),
404	       bogosum / (500000/HZ),
405	       (bogosum / (5000/HZ)) % 100);
406
407	hyp_mode_check();
408}
409
410void __init smp_prepare_boot_cpu(void)
411{
412	set_my_cpu_offset(per_cpu_offset(smp_processor_id()));
413}
414
415void __init smp_prepare_cpus(unsigned int max_cpus)
416{
417	unsigned int ncores = num_possible_cpus();
418
419	init_cpu_topology();
420
421	smp_store_cpu_info(smp_processor_id());
422
423	/*
424	 * are we trying to boot more cores than exist?
425	 */
426	if (max_cpus > ncores)
427		max_cpus = ncores;
428	if (ncores > 1 && max_cpus) {
429		/*
430		 * Initialise the present map, which describes the set of CPUs
431		 * actually populated at the present time. A platform should
432		 * re-initialize the map in the platforms smp_prepare_cpus()
433		 * if present != possible (e.g. physical hotplug).
434		 */
435		init_cpu_present(cpu_possible_mask);
436
437		/*
438		 * Initialise the SCU if there are more than one CPU
439		 * and let them know where to start.
440		 */
441		if (smp_ops.smp_prepare_cpus)
442			smp_ops.smp_prepare_cpus(max_cpus);
443	}
444}
445
446static void (*__smp_cross_call)(const struct cpumask *, unsigned int);
447
448void __init set_smp_cross_call(void (*fn)(const struct cpumask *, unsigned int))
449{
450	if (!__smp_cross_call)
451		__smp_cross_call = fn;
452}
453
454static const char *ipi_types[NR_IPI] __tracepoint_string = {
455#define S(x,s)	[x] = s
456	S(IPI_WAKEUP, "CPU wakeup interrupts"),
457	S(IPI_TIMER, "Timer broadcast interrupts"),
458	S(IPI_RESCHEDULE, "Rescheduling interrupts"),
459	S(IPI_CALL_FUNC, "Function call interrupts"),
460	S(IPI_CALL_FUNC_SINGLE, "Single function call interrupts"),
461	S(IPI_CPU_STOP, "CPU stop interrupts"),
462	S(IPI_IRQ_WORK, "IRQ work interrupts"),
463	S(IPI_COMPLETION, "completion interrupts"),
464};
465
466static void smp_cross_call(const struct cpumask *target, unsigned int ipinr)
467{
468	trace_ipi_raise(target, ipi_types[ipinr]);
469	__smp_cross_call(target, ipinr);
470}
471
472void show_ipi_list(struct seq_file *p, int prec)
473{
474	unsigned int cpu, i;
475
476	for (i = 0; i < NR_IPI; i++) {
477		seq_printf(p, "%*s%u: ", prec - 1, "IPI", i);
478
479		for_each_online_cpu(cpu)
480			seq_printf(p, "%10u ",
481				   __get_irq_stat(cpu, ipi_irqs[i]));
482
483		seq_printf(p, " %s\n", ipi_types[i]);
484	}
485}
486
487u64 smp_irq_stat_cpu(unsigned int cpu)
488{
489	u64 sum = 0;
490	int i;
491
492	for (i = 0; i < NR_IPI; i++)
493		sum += __get_irq_stat(cpu, ipi_irqs[i]);
494
495	return sum;
496}
497
498void arch_send_call_function_ipi_mask(const struct cpumask *mask)
499{
500	smp_cross_call(mask, IPI_CALL_FUNC);
501}
502
503void arch_send_wakeup_ipi_mask(const struct cpumask *mask)
504{
505	smp_cross_call(mask, IPI_WAKEUP);
506}
507
508void arch_send_call_function_single_ipi(int cpu)
509{
510	smp_cross_call(cpumask_of(cpu), IPI_CALL_FUNC_SINGLE);
511}
512
513#ifdef CONFIG_IRQ_WORK
514void arch_irq_work_raise(void)
515{
516	if (arch_irq_work_has_interrupt())
517		smp_cross_call(cpumask_of(smp_processor_id()), IPI_IRQ_WORK);
518}
519#endif
520
521#ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
522void tick_broadcast(const struct cpumask *mask)
523{
524	smp_cross_call(mask, IPI_TIMER);
525}
526#endif
527
528static DEFINE_RAW_SPINLOCK(stop_lock);
529
530/*
531 * ipi_cpu_stop - handle IPI from smp_send_stop()
532 */
533static void ipi_cpu_stop(unsigned int cpu)
534{
535	if (system_state == SYSTEM_BOOTING ||
536	    system_state == SYSTEM_RUNNING) {
537		raw_spin_lock(&stop_lock);
538		pr_crit("CPU%u: stopping\n", cpu);
539		dump_stack();
540		raw_spin_unlock(&stop_lock);
541	}
542
543	set_cpu_online(cpu, false);
544
545	local_fiq_disable();
546	local_irq_disable();
547
548	while (1)
549		cpu_relax();
550}
551
552static DEFINE_PER_CPU(struct completion *, cpu_completion);
553
554int register_ipi_completion(struct completion *completion, int cpu)
555{
556	per_cpu(cpu_completion, cpu) = completion;
557	return IPI_COMPLETION;
558}
559
560static void ipi_complete(unsigned int cpu)
561{
562	complete(per_cpu(cpu_completion, cpu));
563}
564
565/*
566 * Main handler for inter-processor interrupts
567 */
568asmlinkage void __exception_irq_entry do_IPI(int ipinr, struct pt_regs *regs)
569{
570	handle_IPI(ipinr, regs);
571}
572
573void handle_IPI(int ipinr, struct pt_regs *regs)
574{
575	unsigned int cpu = smp_processor_id();
576	struct pt_regs *old_regs = set_irq_regs(regs);
577
578	if ((unsigned)ipinr < NR_IPI) {
579		trace_ipi_entry_rcuidle(ipi_types[ipinr]);
580		__inc_irq_stat(cpu, ipi_irqs[ipinr]);
581	}
582
583	switch (ipinr) {
584	case IPI_WAKEUP:
585		break;
586
587#ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
588	case IPI_TIMER:
589		irq_enter();
590		tick_receive_broadcast();
591		irq_exit();
592		break;
593#endif
594
595	case IPI_RESCHEDULE:
596		scheduler_ipi();
597		break;
598
599	case IPI_CALL_FUNC:
600		irq_enter();
601		generic_smp_call_function_interrupt();
602		irq_exit();
603		break;
604
605	case IPI_CALL_FUNC_SINGLE:
606		irq_enter();
607		generic_smp_call_function_single_interrupt();
608		irq_exit();
609		break;
610
611	case IPI_CPU_STOP:
612		irq_enter();
613		ipi_cpu_stop(cpu);
614		irq_exit();
615		break;
616
617#ifdef CONFIG_IRQ_WORK
618	case IPI_IRQ_WORK:
619		irq_enter();
620		irq_work_run();
621		irq_exit();
622		break;
623#endif
624
625	case IPI_COMPLETION:
626		irq_enter();
627		ipi_complete(cpu);
628		irq_exit();
629		break;
630
631	default:
632		pr_crit("CPU%u: Unknown IPI message 0x%x\n",
633		        cpu, ipinr);
634		break;
635	}
636
637	if ((unsigned)ipinr < NR_IPI)
638		trace_ipi_exit_rcuidle(ipi_types[ipinr]);
639	set_irq_regs(old_regs);
640}
641
642void smp_send_reschedule(int cpu)
643{
644	smp_cross_call(cpumask_of(cpu), IPI_RESCHEDULE);
645}
646
647void smp_send_stop(void)
648{
649	unsigned long timeout;
650	struct cpumask mask;
651
652	cpumask_copy(&mask, cpu_online_mask);
653	cpumask_clear_cpu(smp_processor_id(), &mask);
654	if (!cpumask_empty(&mask))
655		smp_cross_call(&mask, IPI_CPU_STOP);
656
657	/* Wait up to one second for other CPUs to stop */
658	timeout = USEC_PER_SEC;
659	while (num_online_cpus() > 1 && timeout--)
660		udelay(1);
661
662	if (num_online_cpus() > 1)
663		pr_warn("SMP: failed to stop secondary CPUs\n");
664}
665
666/*
667 * not supported here
668 */
669int setup_profiling_timer(unsigned int multiplier)
670{
671	return -EINVAL;
672}
673
674#ifdef CONFIG_CPU_FREQ
675
676static DEFINE_PER_CPU(unsigned long, l_p_j_ref);
677static DEFINE_PER_CPU(unsigned long, l_p_j_ref_freq);
678static unsigned long global_l_p_j_ref;
679static unsigned long global_l_p_j_ref_freq;
680
681static int cpufreq_callback(struct notifier_block *nb,
682					unsigned long val, void *data)
683{
684	struct cpufreq_freqs *freq = data;
685	int cpu = freq->cpu;
686
687	if (freq->flags & CPUFREQ_CONST_LOOPS)
688		return NOTIFY_OK;
689
690	if (!per_cpu(l_p_j_ref, cpu)) {
691		per_cpu(l_p_j_ref, cpu) =
692			per_cpu(cpu_data, cpu).loops_per_jiffy;
693		per_cpu(l_p_j_ref_freq, cpu) = freq->old;
694		if (!global_l_p_j_ref) {
695			global_l_p_j_ref = loops_per_jiffy;
696			global_l_p_j_ref_freq = freq->old;
697		}
698	}
699
700	if ((val == CPUFREQ_PRECHANGE  && freq->old < freq->new) ||
701	    (val == CPUFREQ_POSTCHANGE && freq->old > freq->new)) {
702		loops_per_jiffy = cpufreq_scale(global_l_p_j_ref,
703						global_l_p_j_ref_freq,
704						freq->new);
705		per_cpu(cpu_data, cpu).loops_per_jiffy =
706			cpufreq_scale(per_cpu(l_p_j_ref, cpu),
707					per_cpu(l_p_j_ref_freq, cpu),
708					freq->new);
709	}
710	return NOTIFY_OK;
711}
712
713static struct notifier_block cpufreq_notifier = {
714	.notifier_call  = cpufreq_callback,
715};
716
717static int __init register_cpufreq_notifier(void)
718{
719	return cpufreq_register_notifier(&cpufreq_notifier,
720						CPUFREQ_TRANSITION_NOTIFIER);
721}
722core_initcall(register_cpufreq_notifier);
723
724#endif
725