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/nmi.h>
25 #include <linux/percpu.h>
26 #include <linux/clockchips.h>
27 #include <linux/completion.h>
28 #include <linux/cpufreq.h>
29 #include <linux/irq_work.h>
31 #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/processor.h>
43 #include <asm/sections.h>
44 #include <asm/tlbflush.h>
45 #include <asm/ptrace.h>
46 #include <asm/smp_plat.h>
48 #include <asm/mach/arch.h>
51 #define CREATE_TRACE_POINTS
52 #include <trace/events/ipi.h>
55 * as from 2.5, kernels no longer have an init_tasks structure
56 * so we need some other way of telling a new secondary core
57 * where to place its SVC stack
59 struct secondary_data secondary_data
;
62 * control for which core is the next to come out of the secondary
65 volatile int pen_release
= -1;
76 IPI_CPU_BACKTRACE
= 15,
79 static DECLARE_COMPLETION(cpu_running
);
81 static struct smp_operations smp_ops
;
83 void __init
smp_set_ops(const struct smp_operations
*ops
)
89 static unsigned long get_arch_pgd(pgd_t
*pgd
)
91 #ifdef CONFIG_ARM_LPAE
92 return __phys_to_pfn(virt_to_phys(pgd
));
94 return virt_to_phys(pgd
);
98 int __cpu_up(unsigned int cpu
, struct task_struct
*idle
)
102 if (!smp_ops
.smp_boot_secondary
)
106 * We need to tell the secondary core where to find
107 * its stack and the page tables.
109 secondary_data
.stack
= task_stack_page(idle
) + THREAD_START_SP
;
110 #ifdef CONFIG_ARM_MPU
111 secondary_data
.mpu_rgn_szr
= mpu_rgn_info
.rgns
[MPU_RAM_REGION
].drsr
;
115 secondary_data
.pgdir
= virt_to_phys(idmap_pgd
);
116 secondary_data
.swapper_pg_dir
= get_arch_pgd(swapper_pg_dir
);
118 sync_cache_w(&secondary_data
);
121 * Now bring the CPU into our world.
123 ret
= smp_ops
.smp_boot_secondary(cpu
, idle
);
126 * CPU was successfully started, wait for it
127 * to come online or time out.
129 wait_for_completion_timeout(&cpu_running
,
130 msecs_to_jiffies(1000));
132 if (!cpu_online(cpu
)) {
133 pr_crit("CPU%u: failed to come online\n", cpu
);
137 pr_err("CPU%u: failed to boot: %d\n", cpu
, ret
);
141 memset(&secondary_data
, 0, sizeof(secondary_data
));
145 /* platform specific SMP operations */
146 void __init
smp_init_cpus(void)
148 if (smp_ops
.smp_init_cpus
)
149 smp_ops
.smp_init_cpus();
152 int platform_can_secondary_boot(void)
154 return !!smp_ops
.smp_boot_secondary
;
157 int platform_can_cpu_hotplug(void)
159 #ifdef CONFIG_HOTPLUG_CPU
160 if (smp_ops
.cpu_kill
)
167 #ifdef CONFIG_HOTPLUG_CPU
168 static int platform_cpu_kill(unsigned int cpu
)
170 if (smp_ops
.cpu_kill
)
171 return smp_ops
.cpu_kill(cpu
);
175 static int platform_cpu_disable(unsigned int cpu
)
177 if (smp_ops
.cpu_disable
)
178 return smp_ops
.cpu_disable(cpu
);
183 int platform_can_hotplug_cpu(unsigned int cpu
)
185 /* cpu_die must be specified to support hotplug */
186 if (!smp_ops
.cpu_die
)
189 if (smp_ops
.cpu_can_disable
)
190 return smp_ops
.cpu_can_disable(cpu
);
193 * By default, allow disabling all CPUs except the first one,
194 * since this is special on a lot of platforms, e.g. because
195 * of clock tick interrupts.
201 * __cpu_disable runs on the processor to be shutdown.
203 int __cpu_disable(void)
205 unsigned int cpu
= smp_processor_id();
208 ret
= platform_cpu_disable(cpu
);
213 * Take this CPU offline. Once we clear this, we can't return,
214 * and we must not schedule until we're ready to give up the cpu.
216 set_cpu_online(cpu
, false);
219 * OK - migrate IRQs away from this CPU
224 * Flush user cache and TLB mappings, and then remove this CPU
225 * from the vm mask set of all processes.
227 * Caches are flushed to the Level of Unification Inner Shareable
228 * to write-back dirty lines to unified caches shared by all CPUs.
231 local_flush_tlb_all();
233 clear_tasks_mm_cpumask(cpu
);
238 static DECLARE_COMPLETION(cpu_died
);
241 * called on the thread which is asking for a CPU to be shutdown -
242 * waits until shutdown has completed, or it is timed out.
244 void __cpu_die(unsigned int cpu
)
246 if (!wait_for_completion_timeout(&cpu_died
, msecs_to_jiffies(5000))) {
247 pr_err("CPU%u: cpu didn't die\n", cpu
);
250 pr_notice("CPU%u: shutdown\n", cpu
);
253 * platform_cpu_kill() is generally expected to do the powering off
254 * and/or cutting of clocks to the dying CPU. Optionally, this may
255 * be done by the CPU which is dying in preference to supporting
256 * this call, but that means there is _no_ synchronisation between
257 * the requesting CPU and the dying CPU actually losing power.
259 if (!platform_cpu_kill(cpu
))
260 pr_err("CPU%u: unable to kill\n", cpu
);
264 * Called from the idle thread for the CPU which has been shutdown.
266 * Note that we disable IRQs here, but do not re-enable them
267 * before returning to the caller. This is also the behaviour
268 * of the other hotplug-cpu capable cores, so presumably coming
269 * out of idle fixes this.
271 void arch_cpu_idle_dead(void)
273 unsigned int cpu
= smp_processor_id();
280 * Flush the data out of the L1 cache for this CPU. This must be
281 * before the completion to ensure that data is safely written out
282 * before platform_cpu_kill() gets called - which may disable
283 * *this* CPU and power down its cache.
288 * Tell __cpu_die() that this CPU is now safe to dispose of. Once
289 * this returns, power and/or clocks can be removed at any point
290 * from this CPU and its cache by platform_cpu_kill().
295 * Ensure that the cache lines associated with that completion are
296 * written out. This covers the case where _this_ CPU is doing the
297 * powering down, to ensure that the completion is visible to the
298 * CPU waiting for this one.
303 * The actual CPU shutdown procedure is at least platform (if not
304 * CPU) specific. This may remove power, or it may simply spin.
306 * Platforms are generally expected *NOT* to return from this call,
307 * although there are some which do because they have no way to
308 * power down the CPU. These platforms are the _only_ reason we
309 * have a return path which uses the fragment of assembly below.
311 * The return path should not be used for platforms which can
315 smp_ops
.cpu_die(cpu
);
317 pr_warn("CPU%u: smp_ops.cpu_die() returned, trying to resuscitate\n",
321 * Do not return to the idle loop - jump back to the secondary
322 * cpu initialisation. There's some initialisation which needs
323 * to be repeated to undo the effects of taking the CPU offline.
325 __asm__("mov sp, %0\n"
327 " b secondary_start_kernel"
329 : "r" (task_stack_page(current
) + THREAD_SIZE
- 8));
331 #endif /* CONFIG_HOTPLUG_CPU */
334 * Called by both boot and secondaries to move global data into
335 * per-processor storage.
337 static void smp_store_cpu_info(unsigned int cpuid
)
339 struct cpuinfo_arm
*cpu_info
= &per_cpu(cpu_data
, cpuid
);
341 cpu_info
->loops_per_jiffy
= loops_per_jiffy
;
342 cpu_info
->cpuid
= read_cpuid_id();
344 store_cpu_topology(cpuid
);
348 * This is the secondary CPU boot entry. We're using this CPUs
349 * idle thread stack, but a set of temporary page tables.
351 asmlinkage
void secondary_start_kernel(void)
353 struct mm_struct
*mm
= &init_mm
;
357 * The identity mapping is uncached (strongly ordered), so
358 * switch away from it before attempting any exclusive accesses.
360 cpu_switch_mm(mm
->pgd
, mm
);
361 local_flush_bp_all();
362 enter_lazy_tlb(mm
, current
);
363 local_flush_tlb_all();
366 * All kernel threads share the same mm context; grab a
367 * reference and switch to it.
369 cpu
= smp_processor_id();
370 atomic_inc(&mm
->mm_count
);
371 current
->active_mm
= mm
;
372 cpumask_set_cpu(cpu
, mm_cpumask(mm
));
376 pr_debug("CPU%u: Booted secondary processor\n", cpu
);
379 trace_hardirqs_off();
382 * Give the platform a chance to do its own initialisation.
384 if (smp_ops
.smp_secondary_init
)
385 smp_ops
.smp_secondary_init(cpu
);
387 notify_cpu_starting(cpu
);
391 smp_store_cpu_info(cpu
);
394 * OK, now it's safe to let the boot CPU continue. Wait for
395 * the CPU migration code to notice that the CPU is online
396 * before we continue - which happens after __cpu_up returns.
398 set_cpu_online(cpu
, true);
399 complete(&cpu_running
);
406 * OK, it's off to the idle thread for us
408 cpu_startup_entry(CPUHP_ONLINE
);
411 void __init
smp_cpus_done(unsigned int max_cpus
)
414 unsigned long bogosum
= 0;
416 for_each_online_cpu(cpu
)
417 bogosum
+= per_cpu(cpu_data
, cpu
).loops_per_jiffy
;
419 printk(KERN_INFO
"SMP: Total of %d processors activated "
420 "(%lu.%02lu BogoMIPS).\n",
422 bogosum
/ (500000/HZ
),
423 (bogosum
/ (5000/HZ
)) % 100);
428 void __init
smp_prepare_boot_cpu(void)
430 set_my_cpu_offset(per_cpu_offset(smp_processor_id()));
433 void __init
smp_prepare_cpus(unsigned int max_cpus
)
435 unsigned int ncores
= num_possible_cpus();
439 smp_store_cpu_info(smp_processor_id());
442 * are we trying to boot more cores than exist?
444 if (max_cpus
> ncores
)
446 if (ncores
> 1 && max_cpus
) {
448 * Initialise the present map, which describes the set of CPUs
449 * actually populated at the present time. A platform should
450 * re-initialize the map in the platforms smp_prepare_cpus()
451 * if present != possible (e.g. physical hotplug).
453 init_cpu_present(cpu_possible_mask
);
456 * Initialise the SCU if there are more than one CPU
457 * and let them know where to start.
459 if (smp_ops
.smp_prepare_cpus
)
460 smp_ops
.smp_prepare_cpus(max_cpus
);
464 static void (*__smp_cross_call
)(const struct cpumask
*, unsigned int);
466 void __init
set_smp_cross_call(void (*fn
)(const struct cpumask
*, unsigned int))
468 if (!__smp_cross_call
)
469 __smp_cross_call
= fn
;
472 static const char *ipi_types
[NR_IPI
] __tracepoint_string
= {
473 #define S(x,s) [x] = s
474 S(IPI_WAKEUP
, "CPU wakeup interrupts"),
475 S(IPI_TIMER
, "Timer broadcast interrupts"),
476 S(IPI_RESCHEDULE
, "Rescheduling interrupts"),
477 S(IPI_CALL_FUNC
, "Function call interrupts"),
478 S(IPI_CALL_FUNC_SINGLE
, "Single function call interrupts"),
479 S(IPI_CPU_STOP
, "CPU stop interrupts"),
480 S(IPI_IRQ_WORK
, "IRQ work interrupts"),
481 S(IPI_COMPLETION
, "completion interrupts"),
484 static void smp_cross_call(const struct cpumask
*target
, unsigned int ipinr
)
486 trace_ipi_raise(target
, ipi_types
[ipinr
]);
487 __smp_cross_call(target
, ipinr
);
490 void show_ipi_list(struct seq_file
*p
, int prec
)
494 for (i
= 0; i
< NR_IPI
; i
++) {
495 seq_printf(p
, "%*s%u: ", prec
- 1, "IPI", i
);
497 for_each_online_cpu(cpu
)
498 seq_printf(p
, "%10u ",
499 __get_irq_stat(cpu
, ipi_irqs
[i
]));
501 seq_printf(p
, " %s\n", ipi_types
[i
]);
505 u64
smp_irq_stat_cpu(unsigned int cpu
)
510 for (i
= 0; i
< NR_IPI
; i
++)
511 sum
+= __get_irq_stat(cpu
, ipi_irqs
[i
]);
516 void arch_send_call_function_ipi_mask(const struct cpumask
*mask
)
518 smp_cross_call(mask
, IPI_CALL_FUNC
);
521 void arch_send_wakeup_ipi_mask(const struct cpumask
*mask
)
523 smp_cross_call(mask
, IPI_WAKEUP
);
526 void arch_send_call_function_single_ipi(int cpu
)
528 smp_cross_call(cpumask_of(cpu
), IPI_CALL_FUNC_SINGLE
);
531 #ifdef CONFIG_IRQ_WORK
532 void arch_irq_work_raise(void)
534 if (arch_irq_work_has_interrupt())
535 smp_cross_call(cpumask_of(smp_processor_id()), IPI_IRQ_WORK
);
539 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
540 void tick_broadcast(const struct cpumask
*mask
)
542 smp_cross_call(mask
, IPI_TIMER
);
546 static DEFINE_RAW_SPINLOCK(stop_lock
);
549 * ipi_cpu_stop - handle IPI from smp_send_stop()
551 static void ipi_cpu_stop(unsigned int cpu
)
553 if (system_state
== SYSTEM_BOOTING
||
554 system_state
== SYSTEM_RUNNING
) {
555 raw_spin_lock(&stop_lock
);
556 pr_crit("CPU%u: stopping\n", cpu
);
558 raw_spin_unlock(&stop_lock
);
561 set_cpu_online(cpu
, false);
570 static DEFINE_PER_CPU(struct completion
*, cpu_completion
);
572 int register_ipi_completion(struct completion
*completion
, int cpu
)
574 per_cpu(cpu_completion
, cpu
) = completion
;
575 return IPI_COMPLETION
;
578 static void ipi_complete(unsigned int cpu
)
580 complete(per_cpu(cpu_completion
, cpu
));
584 * Main handler for inter-processor interrupts
586 asmlinkage
void __exception_irq_entry
do_IPI(int ipinr
, struct pt_regs
*regs
)
588 handle_IPI(ipinr
, regs
);
591 void handle_IPI(int ipinr
, struct pt_regs
*regs
)
593 unsigned int cpu
= smp_processor_id();
594 struct pt_regs
*old_regs
= set_irq_regs(regs
);
596 if ((unsigned)ipinr
< NR_IPI
) {
597 trace_ipi_entry_rcuidle(ipi_types
[ipinr
]);
598 __inc_irq_stat(cpu
, ipi_irqs
[ipinr
]);
605 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
608 tick_receive_broadcast();
619 generic_smp_call_function_interrupt();
623 case IPI_CALL_FUNC_SINGLE
:
625 generic_smp_call_function_single_interrupt();
635 #ifdef CONFIG_IRQ_WORK
649 case IPI_CPU_BACKTRACE
:
651 nmi_cpu_backtrace(regs
);
656 pr_crit("CPU%u: Unknown IPI message 0x%x\n",
661 if ((unsigned)ipinr
< NR_IPI
)
662 trace_ipi_exit_rcuidle(ipi_types
[ipinr
]);
663 set_irq_regs(old_regs
);
666 void smp_send_reschedule(int cpu
)
668 smp_cross_call(cpumask_of(cpu
), IPI_RESCHEDULE
);
671 void smp_send_stop(void)
673 unsigned long timeout
;
676 cpumask_copy(&mask
, cpu_online_mask
);
677 cpumask_clear_cpu(smp_processor_id(), &mask
);
678 if (!cpumask_empty(&mask
))
679 smp_cross_call(&mask
, IPI_CPU_STOP
);
681 /* Wait up to one second for other CPUs to stop */
682 timeout
= USEC_PER_SEC
;
683 while (num_online_cpus() > 1 && timeout
--)
686 if (num_online_cpus() > 1)
687 pr_warn("SMP: failed to stop secondary CPUs\n");
693 int setup_profiling_timer(unsigned int multiplier
)
698 #ifdef CONFIG_CPU_FREQ
700 static DEFINE_PER_CPU(unsigned long, l_p_j_ref
);
701 static DEFINE_PER_CPU(unsigned long, l_p_j_ref_freq
);
702 static unsigned long global_l_p_j_ref
;
703 static unsigned long global_l_p_j_ref_freq
;
705 static int cpufreq_callback(struct notifier_block
*nb
,
706 unsigned long val
, void *data
)
708 struct cpufreq_freqs
*freq
= data
;
711 if (freq
->flags
& CPUFREQ_CONST_LOOPS
)
714 if (!per_cpu(l_p_j_ref
, cpu
)) {
715 per_cpu(l_p_j_ref
, cpu
) =
716 per_cpu(cpu_data
, cpu
).loops_per_jiffy
;
717 per_cpu(l_p_j_ref_freq
, cpu
) = freq
->old
;
718 if (!global_l_p_j_ref
) {
719 global_l_p_j_ref
= loops_per_jiffy
;
720 global_l_p_j_ref_freq
= freq
->old
;
724 if ((val
== CPUFREQ_PRECHANGE
&& freq
->old
< freq
->new) ||
725 (val
== CPUFREQ_POSTCHANGE
&& freq
->old
> freq
->new)) {
726 loops_per_jiffy
= cpufreq_scale(global_l_p_j_ref
,
727 global_l_p_j_ref_freq
,
729 per_cpu(cpu_data
, cpu
).loops_per_jiffy
=
730 cpufreq_scale(per_cpu(l_p_j_ref
, cpu
),
731 per_cpu(l_p_j_ref_freq
, cpu
),
737 static struct notifier_block cpufreq_notifier
= {
738 .notifier_call
= cpufreq_callback
,
741 static int __init
register_cpufreq_notifier(void)
743 return cpufreq_register_notifier(&cpufreq_notifier
,
744 CPUFREQ_TRANSITION_NOTIFIER
);
746 core_initcall(register_cpufreq_notifier
);
750 static void raise_nmi(cpumask_t
*mask
)
753 * Generate the backtrace directly if we are running in a calling
754 * context that is not preemptible by the backtrace IPI. Note
755 * that nmi_cpu_backtrace() automatically removes the current cpu
758 if (cpumask_test_cpu(smp_processor_id(), mask
) && irqs_disabled())
759 nmi_cpu_backtrace(NULL
);
761 smp_cross_call(mask
, IPI_CPU_BACKTRACE
);
764 void arch_trigger_all_cpu_backtrace(bool include_self
)
766 nmi_trigger_all_cpu_backtrace(include_self
, raise_nmi
);