1 // SPDX-License-Identifier: GPL-2.0-only
3 * linux/arch/arm/kernel/smp.c
5 * Copyright (C) 2002 ARM Limited, All Rights Reserved.
7 #include <linux/module.h>
8 #include <linux/delay.h>
9 #include <linux/init.h>
10 #include <linux/spinlock.h>
11 #include <linux/sched/mm.h>
12 #include <linux/sched/hotplug.h>
13 #include <linux/sched/task_stack.h>
14 #include <linux/interrupt.h>
15 #include <linux/cache.h>
16 #include <linux/profile.h>
17 #include <linux/errno.h>
19 #include <linux/err.h>
20 #include <linux/cpu.h>
21 #include <linux/seq_file.h>
22 #include <linux/irq.h>
23 #include <linux/nmi.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>
33 #include <asm/cacheflush.h>
35 #include <asm/cputype.h>
36 #include <asm/exception.h>
37 #include <asm/idmap.h>
38 #include <asm/topology.h>
39 #include <asm/mmu_context.h>
40 #include <asm/pgtable.h>
41 #include <asm/pgalloc.h>
42 #include <asm/procinfo.h>
43 #include <asm/processor.h>
44 #include <asm/sections.h>
45 #include <asm/tlbflush.h>
46 #include <asm/ptrace.h>
47 #include <asm/smp_plat.h>
49 #include <asm/mach/arch.h>
52 #define CREATE_TRACE_POINTS
53 #include <trace/events/ipi.h>
56 * as from 2.5, kernels no longer have an init_tasks structure
57 * so we need some other way of telling a new secondary core
58 * where to place its SVC stack
60 struct secondary_data secondary_data;
71 * CPU_BACKTRACE is special and not included in NR_IPI
72 * or tracable with trace_ipi_*
76 * SGI8-15 can be reserved by secure firmware, and thus may
77 * not be usable by the kernel. Please keep the above limited
78 * to at most 8 entries.
82 static DECLARE_COMPLETION(cpu_running);
84 static struct smp_operations smp_ops __ro_after_init;
86 void __init smp_set_ops(const struct smp_operations *ops)
92 static unsigned long get_arch_pgd(pgd_t *pgd)
94 #ifdef CONFIG_ARM_LPAE
95 return __phys_to_pfn(virt_to_phys(pgd));
97 return virt_to_phys(pgd);
101 #if defined(CONFIG_BIG_LITTLE) && defined(CONFIG_HARDEN_BRANCH_PREDICTOR)
102 static int secondary_biglittle_prepare(unsigned int cpu)
104 if (!cpu_vtable[cpu])
105 cpu_vtable[cpu] = kzalloc(sizeof(*cpu_vtable[cpu]), GFP_KERNEL);
107 return cpu_vtable[cpu] ? 0 : -ENOMEM;
110 static void secondary_biglittle_init(void)
112 init_proc_vtable(lookup_processor(read_cpuid_id())->proc);
115 static int secondary_biglittle_prepare(unsigned int cpu)
120 static void secondary_biglittle_init(void)
125 int __cpu_up(unsigned int cpu, struct task_struct *idle)
129 if (!smp_ops.smp_boot_secondary)
132 ret = secondary_biglittle_prepare(cpu);
137 * We need to tell the secondary core where to find
138 * its stack and the page tables.
140 secondary_data.stack = task_stack_page(idle) + THREAD_START_SP;
141 #ifdef CONFIG_ARM_MPU
142 secondary_data.mpu_rgn_info = &mpu_rgn_info;
146 secondary_data.pgdir = virt_to_phys(idmap_pgd);
147 secondary_data.swapper_pg_dir = get_arch_pgd(swapper_pg_dir);
149 sync_cache_w(&secondary_data);
152 * Now bring the CPU into our world.
154 ret = smp_ops.smp_boot_secondary(cpu, idle);
157 * CPU was successfully started, wait for it
158 * to come online or time out.
160 wait_for_completion_timeout(&cpu_running,
161 msecs_to_jiffies(1000));
163 if (!cpu_online(cpu)) {
164 pr_crit("CPU%u: failed to come online\n", cpu);
168 pr_err("CPU%u: failed to boot: %d\n", cpu, ret);
172 memset(&secondary_data, 0, sizeof(secondary_data));
176 /* platform specific SMP operations */
177 void __init smp_init_cpus(void)
179 if (smp_ops.smp_init_cpus)
180 smp_ops.smp_init_cpus();
183 int platform_can_secondary_boot(void)
185 return !!smp_ops.smp_boot_secondary;
188 int platform_can_cpu_hotplug(void)
190 #ifdef CONFIG_HOTPLUG_CPU
191 if (smp_ops.cpu_kill)
198 #ifdef CONFIG_HOTPLUG_CPU
199 static int platform_cpu_kill(unsigned int cpu)
201 if (smp_ops.cpu_kill)
202 return smp_ops.cpu_kill(cpu);
206 static int platform_cpu_disable(unsigned int cpu)
208 if (smp_ops.cpu_disable)
209 return smp_ops.cpu_disable(cpu);
214 int platform_can_hotplug_cpu(unsigned int cpu)
216 /* cpu_die must be specified to support hotplug */
217 if (!smp_ops.cpu_die)
220 if (smp_ops.cpu_can_disable)
221 return smp_ops.cpu_can_disable(cpu);
224 * By default, allow disabling all CPUs except the first one,
225 * since this is special on a lot of platforms, e.g. because
226 * of clock tick interrupts.
232 * __cpu_disable runs on the processor to be shutdown.
234 int __cpu_disable(void)
236 unsigned int cpu = smp_processor_id();
239 ret = platform_cpu_disable(cpu);
244 * Take this CPU offline. Once we clear this, we can't return,
245 * and we must not schedule until we're ready to give up the cpu.
247 set_cpu_online(cpu, false);
250 * OK - migrate IRQs away from this CPU
252 irq_migrate_all_off_this_cpu();
255 * Flush user cache and TLB mappings, and then remove this CPU
256 * from the vm mask set of all processes.
258 * Caches are flushed to the Level of Unification Inner Shareable
259 * to write-back dirty lines to unified caches shared by all CPUs.
262 local_flush_tlb_all();
268 * called on the thread which is asking for a CPU to be shutdown -
269 * waits until shutdown has completed, or it is timed out.
271 void __cpu_die(unsigned int cpu)
273 if (!cpu_wait_death(cpu, 5)) {
274 pr_err("CPU%u: cpu didn't die\n", cpu);
277 pr_debug("CPU%u: shutdown\n", cpu);
279 clear_tasks_mm_cpumask(cpu);
281 * platform_cpu_kill() is generally expected to do the powering off
282 * and/or cutting of clocks to the dying CPU. Optionally, this may
283 * be done by the CPU which is dying in preference to supporting
284 * this call, but that means there is _no_ synchronisation between
285 * the requesting CPU and the dying CPU actually losing power.
287 if (!platform_cpu_kill(cpu))
288 pr_err("CPU%u: unable to kill\n", cpu);
292 * Called from the idle thread for the CPU which has been shutdown.
294 * Note that we disable IRQs here, but do not re-enable them
295 * before returning to the caller. This is also the behaviour
296 * of the other hotplug-cpu capable cores, so presumably coming
297 * out of idle fixes this.
299 void arch_cpu_idle_dead(void)
301 unsigned int cpu = smp_processor_id();
308 * Flush the data out of the L1 cache for this CPU. This must be
309 * before the completion to ensure that data is safely written out
310 * before platform_cpu_kill() gets called - which may disable
311 * *this* CPU and power down its cache.
316 * Tell __cpu_die() that this CPU is now safe to dispose of. Once
317 * this returns, power and/or clocks can be removed at any point
318 * from this CPU and its cache by platform_cpu_kill().
320 (void)cpu_report_death();
323 * Ensure that the cache lines associated with that completion are
324 * written out. This covers the case where _this_ CPU is doing the
325 * powering down, to ensure that the completion is visible to the
326 * CPU waiting for this one.
331 * The actual CPU shutdown procedure is at least platform (if not
332 * CPU) specific. This may remove power, or it may simply spin.
334 * Platforms are generally expected *NOT* to return from this call,
335 * although there are some which do because they have no way to
336 * power down the CPU. These platforms are the _only_ reason we
337 * have a return path which uses the fragment of assembly below.
339 * The return path should not be used for platforms which can
343 smp_ops.cpu_die(cpu);
345 pr_warn("CPU%u: smp_ops.cpu_die() returned, trying to resuscitate\n",
349 * Do not return to the idle loop - jump back to the secondary
350 * cpu initialisation. There's some initialisation which needs
351 * to be repeated to undo the effects of taking the CPU offline.
353 __asm__("mov sp, %0\n"
355 " b secondary_start_kernel"
357 : "r" (task_stack_page(current) + THREAD_SIZE - 8));
359 #endif /* CONFIG_HOTPLUG_CPU */
362 * Called by both boot and secondaries to move global data into
363 * per-processor storage.
365 static void smp_store_cpu_info(unsigned int cpuid)
367 struct cpuinfo_arm *cpu_info = &per_cpu(cpu_data, cpuid);
369 cpu_info->loops_per_jiffy = loops_per_jiffy;
370 cpu_info->cpuid = read_cpuid_id();
372 store_cpu_topology(cpuid);
373 check_cpu_icache_size(cpuid);
377 * This is the secondary CPU boot entry. We're using this CPUs
378 * idle thread stack, but a set of temporary page tables.
380 asmlinkage void secondary_start_kernel(void)
382 struct mm_struct *mm = &init_mm;
385 secondary_biglittle_init();
388 * The identity mapping is uncached (strongly ordered), so
389 * switch away from it before attempting any exclusive accesses.
391 cpu_switch_mm(mm->pgd, mm);
392 local_flush_bp_all();
393 enter_lazy_tlb(mm, current);
394 local_flush_tlb_all();
397 * All kernel threads share the same mm context; grab a
398 * reference and switch to it.
400 cpu = smp_processor_id();
402 current->active_mm = mm;
403 cpumask_set_cpu(cpu, mm_cpumask(mm));
408 setup_vectors_base();
410 pr_debug("CPU%u: Booted secondary processor\n", cpu);
413 trace_hardirqs_off();
416 * Give the platform a chance to do its own initialisation.
418 if (smp_ops.smp_secondary_init)
419 smp_ops.smp_secondary_init(cpu);
421 notify_cpu_starting(cpu);
425 smp_store_cpu_info(cpu);
428 * OK, now it's safe to let the boot CPU continue. Wait for
429 * the CPU migration code to notice that the CPU is online
430 * before we continue - which happens after __cpu_up returns.
432 set_cpu_online(cpu, true);
436 complete(&cpu_running);
443 * OK, it's off to the idle thread for us
445 cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
448 void __init smp_cpus_done(unsigned int max_cpus)
451 unsigned long bogosum = 0;
453 for_each_online_cpu(cpu)
454 bogosum += per_cpu(cpu_data, cpu).loops_per_jiffy;
456 printk(KERN_INFO "SMP: Total of %d processors activated "
457 "(%lu.%02lu BogoMIPS).\n",
459 bogosum / (500000/HZ),
460 (bogosum / (5000/HZ)) % 100);
465 void __init smp_prepare_boot_cpu(void)
467 set_my_cpu_offset(per_cpu_offset(smp_processor_id()));
470 void __init smp_prepare_cpus(unsigned int max_cpus)
472 unsigned int ncores = num_possible_cpus();
476 smp_store_cpu_info(smp_processor_id());
479 * are we trying to boot more cores than exist?
481 if (max_cpus > ncores)
483 if (ncores > 1 && max_cpus) {
485 * Initialise the present map, which describes the set of CPUs
486 * actually populated at the present time. A platform should
487 * re-initialize the map in the platforms smp_prepare_cpus()
488 * if present != possible (e.g. physical hotplug).
490 init_cpu_present(cpu_possible_mask);
493 * Initialise the SCU if there are more than one CPU
494 * and let them know where to start.
496 if (smp_ops.smp_prepare_cpus)
497 smp_ops.smp_prepare_cpus(max_cpus);
501 static void (*__smp_cross_call)(const struct cpumask *, unsigned int);
503 void __init set_smp_cross_call(void (*fn)(const struct cpumask *, unsigned int))
505 if (!__smp_cross_call)
506 __smp_cross_call = fn;
509 static const char *ipi_types[NR_IPI] __tracepoint_string = {
510 #define S(x,s) [x] = s
511 S(IPI_WAKEUP, "CPU wakeup interrupts"),
512 S(IPI_TIMER, "Timer broadcast interrupts"),
513 S(IPI_RESCHEDULE, "Rescheduling interrupts"),
514 S(IPI_CALL_FUNC, "Function call interrupts"),
515 S(IPI_CPU_STOP, "CPU stop interrupts"),
516 S(IPI_IRQ_WORK, "IRQ work interrupts"),
517 S(IPI_COMPLETION, "completion interrupts"),
520 static void smp_cross_call(const struct cpumask *target, unsigned int ipinr)
522 trace_ipi_raise_rcuidle(target, ipi_types[ipinr]);
523 __smp_cross_call(target, ipinr);
526 void show_ipi_list(struct seq_file *p, int prec)
530 for (i = 0; i < NR_IPI; i++) {
531 seq_printf(p, "%*s%u: ", prec - 1, "IPI", i);
533 for_each_online_cpu(cpu)
534 seq_printf(p, "%10u ",
535 __get_irq_stat(cpu, ipi_irqs[i]));
537 seq_printf(p, " %s\n", ipi_types[i]);
541 u64 smp_irq_stat_cpu(unsigned int cpu)
546 for (i = 0; i < NR_IPI; i++)
547 sum += __get_irq_stat(cpu, ipi_irqs[i]);
552 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
554 smp_cross_call(mask, IPI_CALL_FUNC);
557 void arch_send_wakeup_ipi_mask(const struct cpumask *mask)
559 smp_cross_call(mask, IPI_WAKEUP);
562 void arch_send_call_function_single_ipi(int cpu)
564 smp_cross_call(cpumask_of(cpu), IPI_CALL_FUNC);
567 #ifdef CONFIG_IRQ_WORK
568 void arch_irq_work_raise(void)
570 if (arch_irq_work_has_interrupt())
571 smp_cross_call(cpumask_of(smp_processor_id()), IPI_IRQ_WORK);
575 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
576 void tick_broadcast(const struct cpumask *mask)
578 smp_cross_call(mask, IPI_TIMER);
582 static DEFINE_RAW_SPINLOCK(stop_lock);
585 * ipi_cpu_stop - handle IPI from smp_send_stop()
587 static void ipi_cpu_stop(unsigned int cpu)
589 if (system_state <= SYSTEM_RUNNING) {
590 raw_spin_lock(&stop_lock);
591 pr_crit("CPU%u: stopping\n", cpu);
593 raw_spin_unlock(&stop_lock);
596 set_cpu_online(cpu, false);
607 static DEFINE_PER_CPU(struct completion *, cpu_completion);
609 int register_ipi_completion(struct completion *completion, int cpu)
611 per_cpu(cpu_completion, cpu) = completion;
612 return IPI_COMPLETION;
615 static void ipi_complete(unsigned int cpu)
617 complete(per_cpu(cpu_completion, cpu));
621 * Main handler for inter-processor interrupts
623 asmlinkage void __exception_irq_entry do_IPI(int ipinr, struct pt_regs *regs)
625 handle_IPI(ipinr, regs);
628 void handle_IPI(int ipinr, struct pt_regs *regs)
630 unsigned int cpu = smp_processor_id();
631 struct pt_regs *old_regs = set_irq_regs(regs);
633 if ((unsigned)ipinr < NR_IPI) {
634 trace_ipi_entry_rcuidle(ipi_types[ipinr]);
635 __inc_irq_stat(cpu, ipi_irqs[ipinr]);
642 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
645 tick_receive_broadcast();
656 generic_smp_call_function_interrupt();
666 #ifdef CONFIG_IRQ_WORK
680 case IPI_CPU_BACKTRACE:
683 nmi_cpu_backtrace(regs);
689 pr_crit("CPU%u: Unknown IPI message 0x%x\n",
694 if ((unsigned)ipinr < NR_IPI)
695 trace_ipi_exit_rcuidle(ipi_types[ipinr]);
696 set_irq_regs(old_regs);
699 void smp_send_reschedule(int cpu)
701 smp_cross_call(cpumask_of(cpu), IPI_RESCHEDULE);
704 void smp_send_stop(void)
706 unsigned long timeout;
709 cpumask_copy(&mask, cpu_online_mask);
710 cpumask_clear_cpu(smp_processor_id(), &mask);
711 if (!cpumask_empty(&mask))
712 smp_cross_call(&mask, IPI_CPU_STOP);
714 /* Wait up to one second for other CPUs to stop */
715 timeout = USEC_PER_SEC;
716 while (num_online_cpus() > 1 && timeout--)
719 if (num_online_cpus() > 1)
720 pr_warn("SMP: failed to stop secondary CPUs\n");
723 /* In case panic() and panic() called at the same time on CPU1 and CPU2,
724 * and CPU 1 calls panic_smp_self_stop() before crash_smp_send_stop()
725 * CPU1 can't receive the ipi irqs from CPU2, CPU1 will be always online,
726 * kdump fails. So split out the panic_smp_self_stop() and add
727 * set_cpu_online(smp_processor_id(), false).
729 void panic_smp_self_stop(void)
731 pr_debug("CPU %u will stop doing anything useful since another CPU has paniced\n",
733 set_cpu_online(smp_processor_id(), false);
741 int setup_profiling_timer(unsigned int multiplier)
746 #ifdef CONFIG_CPU_FREQ
748 static DEFINE_PER_CPU(unsigned long, l_p_j_ref);
749 static DEFINE_PER_CPU(unsigned long, l_p_j_ref_freq);
750 static unsigned long global_l_p_j_ref;
751 static unsigned long global_l_p_j_ref_freq;
753 static int cpufreq_callback(struct notifier_block *nb,
754 unsigned long val, void *data)
756 struct cpufreq_freqs *freq = data;
757 struct cpumask *cpus = freq->policy->cpus;
758 int cpu, first = cpumask_first(cpus);
761 if (freq->flags & CPUFREQ_CONST_LOOPS)
764 if (!per_cpu(l_p_j_ref, first)) {
765 for_each_cpu(cpu, cpus) {
766 per_cpu(l_p_j_ref, cpu) =
767 per_cpu(cpu_data, cpu).loops_per_jiffy;
768 per_cpu(l_p_j_ref_freq, cpu) = freq->old;
771 if (!global_l_p_j_ref) {
772 global_l_p_j_ref = loops_per_jiffy;
773 global_l_p_j_ref_freq = freq->old;
777 if ((val == CPUFREQ_PRECHANGE && freq->old < freq->new) ||
778 (val == CPUFREQ_POSTCHANGE && freq->old > freq->new)) {
779 loops_per_jiffy = cpufreq_scale(global_l_p_j_ref,
780 global_l_p_j_ref_freq,
783 lpj = cpufreq_scale(per_cpu(l_p_j_ref, first),
784 per_cpu(l_p_j_ref_freq, first), freq->new);
785 for_each_cpu(cpu, cpus)
786 per_cpu(cpu_data, cpu).loops_per_jiffy = lpj;
791 static struct notifier_block cpufreq_notifier = {
792 .notifier_call = cpufreq_callback,
795 static int __init register_cpufreq_notifier(void)
797 return cpufreq_register_notifier(&cpufreq_notifier,
798 CPUFREQ_TRANSITION_NOTIFIER);
800 core_initcall(register_cpufreq_notifier);
804 static void raise_nmi(cpumask_t *mask)
806 __smp_cross_call(mask, IPI_CPU_BACKTRACE);
809 void arch_trigger_cpumask_backtrace(const cpumask_t *mask, bool exclude_self)
811 nmi_trigger_cpumask_backtrace(mask, exclude_self, raise_nmi);