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/mm.h>
15 #include <linux/sched/hotplug.h>
16 #include <linux/sched/task_stack.h>
17 #include <linux/interrupt.h>
18 #include <linux/cache.h>
19 #include <linux/profile.h>
20 #include <linux/errno.h>
22 #include <linux/err.h>
23 #include <linux/cpu.h>
24 #include <linux/seq_file.h>
25 #include <linux/irq.h>
26 #include <linux/nmi.h>
27 #include <linux/percpu.h>
28 #include <linux/clockchips.h>
29 #include <linux/completion.h>
30 #include <linux/cpufreq.h>
31 #include <linux/irq_work.h>
33 #include <linux/atomic.h>
35 #include <asm/cacheflush.h>
37 #include <asm/cputype.h>
38 #include <asm/exception.h>
39 #include <asm/idmap.h>
40 #include <asm/topology.h>
41 #include <asm/mmu_context.h>
42 #include <asm/pgtable.h>
43 #include <asm/pgalloc.h>
44 #include <asm/processor.h>
45 #include <asm/sections.h>
46 #include <asm/tlbflush.h>
47 #include <asm/ptrace.h>
48 #include <asm/smp_plat.h>
50 #include <asm/mach/arch.h>
53 #define CREATE_TRACE_POINTS
54 #include <trace/events/ipi.h>
57 * as from 2.5, kernels no longer have an init_tasks structure
58 * so we need some other way of telling a new secondary core
59 * where to place its SVC stack
61 struct secondary_data secondary_data
;
64 * control for which core is the next to come out of the secondary
67 volatile int pen_release
= -1;
79 * SGI8-15 can be reserved by secure firmware, and thus may
80 * not be usable by the kernel. Please keep the above limited
81 * to at most 8 entries.
85 static DECLARE_COMPLETION(cpu_running
);
87 static struct smp_operations smp_ops __ro_after_init
;
89 void __init
smp_set_ops(const struct smp_operations
*ops
)
95 static unsigned long get_arch_pgd(pgd_t
*pgd
)
97 #ifdef CONFIG_ARM_LPAE
98 return __phys_to_pfn(virt_to_phys(pgd
));
100 return virt_to_phys(pgd
);
104 int __cpu_up(unsigned int cpu
, struct task_struct
*idle
)
108 if (!smp_ops
.smp_boot_secondary
)
112 * We need to tell the secondary core where to find
113 * its stack and the page tables.
115 secondary_data
.stack
= task_stack_page(idle
) + THREAD_START_SP
;
116 #ifdef CONFIG_ARM_MPU
117 secondary_data
.mpu_rgn_info
= &mpu_rgn_info
;
121 secondary_data
.pgdir
= virt_to_phys(idmap_pgd
);
122 secondary_data
.swapper_pg_dir
= get_arch_pgd(swapper_pg_dir
);
124 sync_cache_w(&secondary_data
);
127 * Now bring the CPU into our world.
129 ret
= smp_ops
.smp_boot_secondary(cpu
, idle
);
132 * CPU was successfully started, wait for it
133 * to come online or time out.
135 wait_for_completion_timeout(&cpu_running
,
136 msecs_to_jiffies(1000));
138 if (!cpu_online(cpu
)) {
139 pr_crit("CPU%u: failed to come online\n", cpu
);
143 pr_err("CPU%u: failed to boot: %d\n", cpu
, ret
);
147 memset(&secondary_data
, 0, sizeof(secondary_data
));
151 /* platform specific SMP operations */
152 void __init
smp_init_cpus(void)
154 if (smp_ops
.smp_init_cpus
)
155 smp_ops
.smp_init_cpus();
158 int platform_can_secondary_boot(void)
160 return !!smp_ops
.smp_boot_secondary
;
163 int platform_can_cpu_hotplug(void)
165 #ifdef CONFIG_HOTPLUG_CPU
166 if (smp_ops
.cpu_kill
)
173 #ifdef CONFIG_HOTPLUG_CPU
174 static int platform_cpu_kill(unsigned int cpu
)
176 if (smp_ops
.cpu_kill
)
177 return smp_ops
.cpu_kill(cpu
);
181 static int platform_cpu_disable(unsigned int cpu
)
183 if (smp_ops
.cpu_disable
)
184 return smp_ops
.cpu_disable(cpu
);
189 int platform_can_hotplug_cpu(unsigned int cpu
)
191 /* cpu_die must be specified to support hotplug */
192 if (!smp_ops
.cpu_die
)
195 if (smp_ops
.cpu_can_disable
)
196 return smp_ops
.cpu_can_disable(cpu
);
199 * By default, allow disabling all CPUs except the first one,
200 * since this is special on a lot of platforms, e.g. because
201 * of clock tick interrupts.
207 * __cpu_disable runs on the processor to be shutdown.
209 int __cpu_disable(void)
211 unsigned int cpu
= smp_processor_id();
214 ret
= platform_cpu_disable(cpu
);
219 * Take this CPU offline. Once we clear this, we can't return,
220 * and we must not schedule until we're ready to give up the cpu.
222 set_cpu_online(cpu
, false);
225 * OK - migrate IRQs away from this CPU
230 * Flush user cache and TLB mappings, and then remove this CPU
231 * from the vm mask set of all processes.
233 * Caches are flushed to the Level of Unification Inner Shareable
234 * to write-back dirty lines to unified caches shared by all CPUs.
237 local_flush_tlb_all();
239 clear_tasks_mm_cpumask(cpu
);
244 static DECLARE_COMPLETION(cpu_died
);
247 * called on the thread which is asking for a CPU to be shutdown -
248 * waits until shutdown has completed, or it is timed out.
250 void __cpu_die(unsigned int cpu
)
252 if (!wait_for_completion_timeout(&cpu_died
, msecs_to_jiffies(5000))) {
253 pr_err("CPU%u: cpu didn't die\n", cpu
);
256 pr_debug("CPU%u: shutdown\n", cpu
);
259 * platform_cpu_kill() is generally expected to do the powering off
260 * and/or cutting of clocks to the dying CPU. Optionally, this may
261 * be done by the CPU which is dying in preference to supporting
262 * this call, but that means there is _no_ synchronisation between
263 * the requesting CPU and the dying CPU actually losing power.
265 if (!platform_cpu_kill(cpu
))
266 pr_err("CPU%u: unable to kill\n", cpu
);
270 * Called from the idle thread for the CPU which has been shutdown.
272 * Note that we disable IRQs here, but do not re-enable them
273 * before returning to the caller. This is also the behaviour
274 * of the other hotplug-cpu capable cores, so presumably coming
275 * out of idle fixes this.
277 void arch_cpu_idle_dead(void)
279 unsigned int cpu
= smp_processor_id();
286 * Flush the data out of the L1 cache for this CPU. This must be
287 * before the completion to ensure that data is safely written out
288 * before platform_cpu_kill() gets called - which may disable
289 * *this* CPU and power down its cache.
294 * Tell __cpu_die() that this CPU is now safe to dispose of. Once
295 * this returns, power and/or clocks can be removed at any point
296 * from this CPU and its cache by platform_cpu_kill().
301 * Ensure that the cache lines associated with that completion are
302 * written out. This covers the case where _this_ CPU is doing the
303 * powering down, to ensure that the completion is visible to the
304 * CPU waiting for this one.
309 * The actual CPU shutdown procedure is at least platform (if not
310 * CPU) specific. This may remove power, or it may simply spin.
312 * Platforms are generally expected *NOT* to return from this call,
313 * although there are some which do because they have no way to
314 * power down the CPU. These platforms are the _only_ reason we
315 * have a return path which uses the fragment of assembly below.
317 * The return path should not be used for platforms which can
321 smp_ops
.cpu_die(cpu
);
323 pr_warn("CPU%u: smp_ops.cpu_die() returned, trying to resuscitate\n",
327 * Do not return to the idle loop - jump back to the secondary
328 * cpu initialisation. There's some initialisation which needs
329 * to be repeated to undo the effects of taking the CPU offline.
331 __asm__("mov sp, %0\n"
333 " b secondary_start_kernel"
335 : "r" (task_stack_page(current
) + THREAD_SIZE
- 8));
337 #endif /* CONFIG_HOTPLUG_CPU */
340 * Called by both boot and secondaries to move global data into
341 * per-processor storage.
343 static void smp_store_cpu_info(unsigned int cpuid
)
345 struct cpuinfo_arm
*cpu_info
= &per_cpu(cpu_data
, cpuid
);
347 cpu_info
->loops_per_jiffy
= loops_per_jiffy
;
348 cpu_info
->cpuid
= read_cpuid_id();
350 store_cpu_topology(cpuid
);
354 * This is the secondary CPU boot entry. We're using this CPUs
355 * idle thread stack, but a set of temporary page tables.
357 asmlinkage
void secondary_start_kernel(void)
359 struct mm_struct
*mm
= &init_mm
;
363 * The identity mapping is uncached (strongly ordered), so
364 * switch away from it before attempting any exclusive accesses.
366 cpu_switch_mm(mm
->pgd
, mm
);
367 local_flush_bp_all();
368 enter_lazy_tlb(mm
, current
);
369 local_flush_tlb_all();
372 * All kernel threads share the same mm context; grab a
373 * reference and switch to it.
375 cpu
= smp_processor_id();
377 current
->active_mm
= mm
;
378 cpumask_set_cpu(cpu
, mm_cpumask(mm
));
383 setup_vectors_base();
385 pr_debug("CPU%u: Booted secondary processor\n", cpu
);
388 trace_hardirqs_off();
391 * Give the platform a chance to do its own initialisation.
393 if (smp_ops
.smp_secondary_init
)
394 smp_ops
.smp_secondary_init(cpu
);
396 notify_cpu_starting(cpu
);
400 smp_store_cpu_info(cpu
);
403 * OK, now it's safe to let the boot CPU continue. Wait for
404 * the CPU migration code to notice that the CPU is online
405 * before we continue - which happens after __cpu_up returns.
407 set_cpu_online(cpu
, true);
408 complete(&cpu_running
);
415 * OK, it's off to the idle thread for us
417 cpu_startup_entry(CPUHP_AP_ONLINE_IDLE
);
420 void __init
smp_cpus_done(unsigned int max_cpus
)
423 unsigned long bogosum
= 0;
425 for_each_online_cpu(cpu
)
426 bogosum
+= per_cpu(cpu_data
, cpu
).loops_per_jiffy
;
428 printk(KERN_INFO
"SMP: Total of %d processors activated "
429 "(%lu.%02lu BogoMIPS).\n",
431 bogosum
/ (500000/HZ
),
432 (bogosum
/ (5000/HZ
)) % 100);
437 void __init
smp_prepare_boot_cpu(void)
439 set_my_cpu_offset(per_cpu_offset(smp_processor_id()));
442 void __init
smp_prepare_cpus(unsigned int max_cpus
)
444 unsigned int ncores
= num_possible_cpus();
448 smp_store_cpu_info(smp_processor_id());
451 * are we trying to boot more cores than exist?
453 if (max_cpus
> ncores
)
455 if (ncores
> 1 && max_cpus
) {
457 * Initialise the present map, which describes the set of CPUs
458 * actually populated at the present time. A platform should
459 * re-initialize the map in the platforms smp_prepare_cpus()
460 * if present != possible (e.g. physical hotplug).
462 init_cpu_present(cpu_possible_mask
);
465 * Initialise the SCU if there are more than one CPU
466 * and let them know where to start.
468 if (smp_ops
.smp_prepare_cpus
)
469 smp_ops
.smp_prepare_cpus(max_cpus
);
473 static void (*__smp_cross_call
)(const struct cpumask
*, unsigned int);
475 void __init
set_smp_cross_call(void (*fn
)(const struct cpumask
*, unsigned int))
477 if (!__smp_cross_call
)
478 __smp_cross_call
= fn
;
481 static const char *ipi_types
[NR_IPI
] __tracepoint_string
= {
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_CPU_STOP
, "CPU stop interrupts"),
488 S(IPI_IRQ_WORK
, "IRQ work interrupts"),
489 S(IPI_COMPLETION
, "completion interrupts"),
492 static void smp_cross_call(const struct cpumask
*target
, unsigned int ipinr
)
494 trace_ipi_raise_rcuidle(target
, ipi_types
[ipinr
]);
495 __smp_cross_call(target
, ipinr
);
498 void show_ipi_list(struct seq_file
*p
, int prec
)
502 for (i
= 0; i
< NR_IPI
; i
++) {
503 seq_printf(p
, "%*s%u: ", prec
- 1, "IPI", i
);
505 for_each_online_cpu(cpu
)
506 seq_printf(p
, "%10u ",
507 __get_irq_stat(cpu
, ipi_irqs
[i
]));
509 seq_printf(p
, " %s\n", ipi_types
[i
]);
513 u64
smp_irq_stat_cpu(unsigned int cpu
)
518 for (i
= 0; i
< NR_IPI
; i
++)
519 sum
+= __get_irq_stat(cpu
, ipi_irqs
[i
]);
524 void arch_send_call_function_ipi_mask(const struct cpumask
*mask
)
526 smp_cross_call(mask
, IPI_CALL_FUNC
);
529 void arch_send_wakeup_ipi_mask(const struct cpumask
*mask
)
531 smp_cross_call(mask
, IPI_WAKEUP
);
534 void arch_send_call_function_single_ipi(int cpu
)
536 smp_cross_call(cpumask_of(cpu
), IPI_CALL_FUNC
);
539 #ifdef CONFIG_IRQ_WORK
540 void arch_irq_work_raise(void)
542 if (arch_irq_work_has_interrupt())
543 smp_cross_call(cpumask_of(smp_processor_id()), IPI_IRQ_WORK
);
547 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
548 void tick_broadcast(const struct cpumask
*mask
)
550 smp_cross_call(mask
, IPI_TIMER
);
554 static DEFINE_RAW_SPINLOCK(stop_lock
);
557 * ipi_cpu_stop - handle IPI from smp_send_stop()
559 static void ipi_cpu_stop(unsigned int cpu
)
561 if (system_state
<= SYSTEM_RUNNING
) {
562 raw_spin_lock(&stop_lock
);
563 pr_crit("CPU%u: stopping\n", cpu
);
565 raw_spin_unlock(&stop_lock
);
568 set_cpu_online(cpu
, false);
577 static DEFINE_PER_CPU(struct completion
*, cpu_completion
);
579 int register_ipi_completion(struct completion
*completion
, int cpu
)
581 per_cpu(cpu_completion
, cpu
) = completion
;
582 return IPI_COMPLETION
;
585 static void ipi_complete(unsigned int cpu
)
587 complete(per_cpu(cpu_completion
, cpu
));
591 * Main handler for inter-processor interrupts
593 asmlinkage
void __exception_irq_entry
do_IPI(int ipinr
, struct pt_regs
*regs
)
595 handle_IPI(ipinr
, regs
);
598 void handle_IPI(int ipinr
, struct pt_regs
*regs
)
600 unsigned int cpu
= smp_processor_id();
601 struct pt_regs
*old_regs
= set_irq_regs(regs
);
603 if ((unsigned)ipinr
< NR_IPI
) {
604 trace_ipi_entry_rcuidle(ipi_types
[ipinr
]);
605 __inc_irq_stat(cpu
, ipi_irqs
[ipinr
]);
612 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
615 tick_receive_broadcast();
626 generic_smp_call_function_interrupt();
636 #ifdef CONFIG_IRQ_WORK
650 case IPI_CPU_BACKTRACE
:
653 nmi_cpu_backtrace(regs
);
659 pr_crit("CPU%u: Unknown IPI message 0x%x\n",
664 if ((unsigned)ipinr
< NR_IPI
)
665 trace_ipi_exit_rcuidle(ipi_types
[ipinr
]);
666 set_irq_regs(old_regs
);
669 void smp_send_reschedule(int cpu
)
671 smp_cross_call(cpumask_of(cpu
), IPI_RESCHEDULE
);
674 void smp_send_stop(void)
676 unsigned long timeout
;
679 cpumask_copy(&mask
, cpu_online_mask
);
680 cpumask_clear_cpu(smp_processor_id(), &mask
);
681 if (!cpumask_empty(&mask
))
682 smp_cross_call(&mask
, IPI_CPU_STOP
);
684 /* Wait up to one second for other CPUs to stop */
685 timeout
= USEC_PER_SEC
;
686 while (num_online_cpus() > 1 && timeout
--)
689 if (num_online_cpus() > 1)
690 pr_warn("SMP: failed to stop secondary CPUs\n");
696 int setup_profiling_timer(unsigned int multiplier
)
701 #ifdef CONFIG_CPU_FREQ
703 static DEFINE_PER_CPU(unsigned long, l_p_j_ref
);
704 static DEFINE_PER_CPU(unsigned long, l_p_j_ref_freq
);
705 static unsigned long global_l_p_j_ref
;
706 static unsigned long global_l_p_j_ref_freq
;
708 static int cpufreq_callback(struct notifier_block
*nb
,
709 unsigned long val
, void *data
)
711 struct cpufreq_freqs
*freq
= data
;
714 if (freq
->flags
& CPUFREQ_CONST_LOOPS
)
717 if (!per_cpu(l_p_j_ref
, cpu
)) {
718 per_cpu(l_p_j_ref
, cpu
) =
719 per_cpu(cpu_data
, cpu
).loops_per_jiffy
;
720 per_cpu(l_p_j_ref_freq
, cpu
) = freq
->old
;
721 if (!global_l_p_j_ref
) {
722 global_l_p_j_ref
= loops_per_jiffy
;
723 global_l_p_j_ref_freq
= freq
->old
;
727 if ((val
== CPUFREQ_PRECHANGE
&& freq
->old
< freq
->new) ||
728 (val
== CPUFREQ_POSTCHANGE
&& freq
->old
> freq
->new)) {
729 loops_per_jiffy
= cpufreq_scale(global_l_p_j_ref
,
730 global_l_p_j_ref_freq
,
732 per_cpu(cpu_data
, cpu
).loops_per_jiffy
=
733 cpufreq_scale(per_cpu(l_p_j_ref
, cpu
),
734 per_cpu(l_p_j_ref_freq
, cpu
),
740 static struct notifier_block cpufreq_notifier
= {
741 .notifier_call
= cpufreq_callback
,
744 static int __init
register_cpufreq_notifier(void)
746 return cpufreq_register_notifier(&cpufreq_notifier
,
747 CPUFREQ_TRANSITION_NOTIFIER
);
749 core_initcall(register_cpufreq_notifier
);
753 static void raise_nmi(cpumask_t
*mask
)
755 smp_cross_call(mask
, IPI_CPU_BACKTRACE
);
758 void arch_trigger_cpumask_backtrace(const cpumask_t
*mask
, bool exclude_self
)
760 nmi_trigger_cpumask_backtrace(mask
, exclude_self
, raise_nmi
);