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>
50 #define CREATE_TRACE_POINTS
51 #include <trace/events/ipi.h>
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
58 struct secondary_data secondary_data
;
61 * control for which core is the next to come out of the secondary
64 volatile int pen_release
= -1;
77 static DECLARE_COMPLETION(cpu_running
);
79 static struct smp_operations smp_ops
;
81 void __init
smp_set_ops(struct smp_operations
*ops
)
87 static unsigned long get_arch_pgd(pgd_t
*pgd
)
89 #ifdef CONFIG_ARM_LPAE
90 return __phys_to_pfn(virt_to_phys(pgd
));
92 return virt_to_phys(pgd
);
96 int __cpu_up(unsigned int cpu
, struct task_struct
*idle
)
100 if (!smp_ops
.smp_boot_secondary
)
104 * We need to tell the secondary core where to find
105 * its stack and the page tables.
107 secondary_data
.stack
= task_stack_page(idle
) + THREAD_START_SP
;
108 #ifdef CONFIG_ARM_MPU
109 secondary_data
.mpu_rgn_szr
= mpu_rgn_info
.rgns
[MPU_RAM_REGION
].drsr
;
113 secondary_data
.pgdir
= virt_to_phys(idmap_pgd
);
114 secondary_data
.swapper_pg_dir
= get_arch_pgd(swapper_pg_dir
);
116 sync_cache_w(&secondary_data
);
119 * Now bring the CPU into our world.
121 ret
= smp_ops
.smp_boot_secondary(cpu
, idle
);
124 * CPU was successfully started, wait for it
125 * to come online or time out.
127 wait_for_completion_timeout(&cpu_running
,
128 msecs_to_jiffies(1000));
130 if (!cpu_online(cpu
)) {
131 pr_crit("CPU%u: failed to come online\n", cpu
);
135 pr_err("CPU%u: failed to boot: %d\n", cpu
, ret
);
139 memset(&secondary_data
, 0, sizeof(secondary_data
));
143 /* platform specific SMP operations */
144 void __init
smp_init_cpus(void)
146 if (smp_ops
.smp_init_cpus
)
147 smp_ops
.smp_init_cpus();
150 int platform_can_secondary_boot(void)
152 return !!smp_ops
.smp_boot_secondary
;
155 int platform_can_cpu_hotplug(void)
157 #ifdef CONFIG_HOTPLUG_CPU
158 if (smp_ops
.cpu_kill
)
165 #ifdef CONFIG_HOTPLUG_CPU
166 static int platform_cpu_kill(unsigned int cpu
)
168 if (smp_ops
.cpu_kill
)
169 return smp_ops
.cpu_kill(cpu
);
173 static int platform_cpu_disable(unsigned int cpu
)
175 if (smp_ops
.cpu_disable
)
176 return smp_ops
.cpu_disable(cpu
);
179 * By default, allow disabling all CPUs except the first one,
180 * since this is special on a lot of platforms, e.g. because
181 * of clock tick interrupts.
183 return cpu
== 0 ? -EPERM
: 0;
186 * __cpu_disable runs on the processor to be shutdown.
188 int __cpu_disable(void)
190 unsigned int cpu
= smp_processor_id();
193 ret
= platform_cpu_disable(cpu
);
198 * Take this CPU offline. Once we clear this, we can't return,
199 * and we must not schedule until we're ready to give up the cpu.
201 set_cpu_online(cpu
, false);
204 * OK - migrate IRQs away from this CPU
209 * Flush user cache and TLB mappings, and then remove this CPU
210 * from the vm mask set of all processes.
212 * Caches are flushed to the Level of Unification Inner Shareable
213 * to write-back dirty lines to unified caches shared by all CPUs.
216 local_flush_tlb_all();
218 clear_tasks_mm_cpumask(cpu
);
223 static DECLARE_COMPLETION(cpu_died
);
226 * called on the thread which is asking for a CPU to be shutdown -
227 * waits until shutdown has completed, or it is timed out.
229 void __cpu_die(unsigned int cpu
)
231 if (!wait_for_completion_timeout(&cpu_died
, msecs_to_jiffies(5000))) {
232 pr_err("CPU%u: cpu didn't die\n", cpu
);
235 pr_notice("CPU%u: shutdown\n", cpu
);
238 * platform_cpu_kill() is generally expected to do the powering off
239 * and/or cutting of clocks to the dying CPU. Optionally, this may
240 * be done by the CPU which is dying in preference to supporting
241 * this call, but that means there is _no_ synchronisation between
242 * the requesting CPU and the dying CPU actually losing power.
244 if (!platform_cpu_kill(cpu
))
245 pr_err("CPU%u: unable to kill\n", cpu
);
249 * Called from the idle thread for the CPU which has been shutdown.
251 * Note that we disable IRQs here, but do not re-enable them
252 * before returning to the caller. This is also the behaviour
253 * of the other hotplug-cpu capable cores, so presumably coming
254 * out of idle fixes this.
256 void __ref
cpu_die(void)
258 unsigned int cpu
= smp_processor_id();
265 * Flush the data out of the L1 cache for this CPU. This must be
266 * before the completion to ensure that data is safely written out
267 * before platform_cpu_kill() gets called - which may disable
268 * *this* CPU and power down its cache.
273 * Tell __cpu_die() that this CPU is now safe to dispose of. Once
274 * this returns, power and/or clocks can be removed at any point
275 * from this CPU and its cache by platform_cpu_kill().
280 * Ensure that the cache lines associated with that completion are
281 * written out. This covers the case where _this_ CPU is doing the
282 * powering down, to ensure that the completion is visible to the
283 * CPU waiting for this one.
288 * The actual CPU shutdown procedure is at least platform (if not
289 * CPU) specific. This may remove power, or it may simply spin.
291 * Platforms are generally expected *NOT* to return from this call,
292 * although there are some which do because they have no way to
293 * power down the CPU. These platforms are the _only_ reason we
294 * have a return path which uses the fragment of assembly below.
296 * The return path should not be used for platforms which can
300 smp_ops
.cpu_die(cpu
);
302 pr_warn("CPU%u: smp_ops.cpu_die() returned, trying to resuscitate\n",
306 * Do not return to the idle loop - jump back to the secondary
307 * cpu initialisation. There's some initialisation which needs
308 * to be repeated to undo the effects of taking the CPU offline.
310 __asm__("mov sp, %0\n"
312 " b secondary_start_kernel"
314 : "r" (task_stack_page(current
) + THREAD_SIZE
- 8));
316 #endif /* CONFIG_HOTPLUG_CPU */
319 * Called by both boot and secondaries to move global data into
320 * per-processor storage.
322 static void smp_store_cpu_info(unsigned int cpuid
)
324 struct cpuinfo_arm
*cpu_info
= &per_cpu(cpu_data
, cpuid
);
326 cpu_info
->loops_per_jiffy
= loops_per_jiffy
;
327 cpu_info
->cpuid
= read_cpuid_id();
329 store_cpu_topology(cpuid
);
333 * This is the secondary CPU boot entry. We're using this CPUs
334 * idle thread stack, but a set of temporary page tables.
336 asmlinkage
void secondary_start_kernel(void)
338 struct mm_struct
*mm
= &init_mm
;
342 * The identity mapping is uncached (strongly ordered), so
343 * switch away from it before attempting any exclusive accesses.
345 cpu_switch_mm(mm
->pgd
, mm
);
346 local_flush_bp_all();
347 enter_lazy_tlb(mm
, current
);
348 local_flush_tlb_all();
351 * All kernel threads share the same mm context; grab a
352 * reference and switch to it.
354 cpu
= smp_processor_id();
355 atomic_inc(&mm
->mm_count
);
356 current
->active_mm
= mm
;
357 cpumask_set_cpu(cpu
, mm_cpumask(mm
));
361 pr_debug("CPU%u: Booted secondary processor\n", cpu
);
364 trace_hardirqs_off();
367 * Give the platform a chance to do its own initialisation.
369 if (smp_ops
.smp_secondary_init
)
370 smp_ops
.smp_secondary_init(cpu
);
372 notify_cpu_starting(cpu
);
376 smp_store_cpu_info(cpu
);
379 * OK, now it's safe to let the boot CPU continue. Wait for
380 * the CPU migration code to notice that the CPU is online
381 * before we continue - which happens after __cpu_up returns.
383 set_cpu_online(cpu
, true);
384 complete(&cpu_running
);
390 * OK, it's off to the idle thread for us
392 cpu_startup_entry(CPUHP_ONLINE
);
395 void __init
smp_cpus_done(unsigned int max_cpus
)
398 unsigned long bogosum
= 0;
400 for_each_online_cpu(cpu
)
401 bogosum
+= per_cpu(cpu_data
, cpu
).loops_per_jiffy
;
403 printk(KERN_INFO
"SMP: Total of %d processors activated "
404 "(%lu.%02lu BogoMIPS).\n",
406 bogosum
/ (500000/HZ
),
407 (bogosum
/ (5000/HZ
)) % 100);
412 void __init
smp_prepare_boot_cpu(void)
414 set_my_cpu_offset(per_cpu_offset(smp_processor_id()));
417 void __init
smp_prepare_cpus(unsigned int max_cpus
)
419 unsigned int ncores
= num_possible_cpus();
423 smp_store_cpu_info(smp_processor_id());
426 * are we trying to boot more cores than exist?
428 if (max_cpus
> ncores
)
430 if (ncores
> 1 && max_cpus
) {
432 * Initialise the present map, which describes the set of CPUs
433 * actually populated at the present time. A platform should
434 * re-initialize the map in the platforms smp_prepare_cpus()
435 * if present != possible (e.g. physical hotplug).
437 init_cpu_present(cpu_possible_mask
);
440 * Initialise the SCU if there are more than one CPU
441 * and let them know where to start.
443 if (smp_ops
.smp_prepare_cpus
)
444 smp_ops
.smp_prepare_cpus(max_cpus
);
448 static void (*__smp_cross_call
)(const struct cpumask
*, unsigned int);
450 void __init
set_smp_cross_call(void (*fn
)(const struct cpumask
*, unsigned int))
452 if (!__smp_cross_call
)
453 __smp_cross_call
= fn
;
456 static const char *ipi_types
[NR_IPI
] __tracepoint_string
= {
457 #define S(x,s) [x] = s
458 S(IPI_WAKEUP
, "CPU wakeup interrupts"),
459 S(IPI_TIMER
, "Timer broadcast interrupts"),
460 S(IPI_RESCHEDULE
, "Rescheduling interrupts"),
461 S(IPI_CALL_FUNC
, "Function call interrupts"),
462 S(IPI_CALL_FUNC_SINGLE
, "Single function call interrupts"),
463 S(IPI_CPU_STOP
, "CPU stop interrupts"),
464 S(IPI_IRQ_WORK
, "IRQ work interrupts"),
465 S(IPI_COMPLETION
, "completion interrupts"),
468 static void smp_cross_call(const struct cpumask
*target
, unsigned int ipinr
)
470 trace_ipi_raise(target
, ipi_types
[ipinr
]);
471 __smp_cross_call(target
, ipinr
);
474 void show_ipi_list(struct seq_file
*p
, int prec
)
478 for (i
= 0; i
< NR_IPI
; i
++) {
479 seq_printf(p
, "%*s%u: ", prec
- 1, "IPI", i
);
481 for_each_online_cpu(cpu
)
482 seq_printf(p
, "%10u ",
483 __get_irq_stat(cpu
, ipi_irqs
[i
]));
485 seq_printf(p
, " %s\n", ipi_types
[i
]);
489 u64
smp_irq_stat_cpu(unsigned int cpu
)
494 for (i
= 0; i
< NR_IPI
; i
++)
495 sum
+= __get_irq_stat(cpu
, ipi_irqs
[i
]);
500 void arch_send_call_function_ipi_mask(const struct cpumask
*mask
)
502 smp_cross_call(mask
, IPI_CALL_FUNC
);
505 void arch_send_wakeup_ipi_mask(const struct cpumask
*mask
)
507 smp_cross_call(mask
, IPI_WAKEUP
);
510 void arch_send_call_function_single_ipi(int cpu
)
512 smp_cross_call(cpumask_of(cpu
), IPI_CALL_FUNC_SINGLE
);
515 #ifdef CONFIG_IRQ_WORK
516 void arch_irq_work_raise(void)
518 if (arch_irq_work_has_interrupt())
519 smp_cross_call(cpumask_of(smp_processor_id()), IPI_IRQ_WORK
);
523 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
524 void tick_broadcast(const struct cpumask
*mask
)
526 smp_cross_call(mask
, IPI_TIMER
);
530 static DEFINE_RAW_SPINLOCK(stop_lock
);
533 * ipi_cpu_stop - handle IPI from smp_send_stop()
535 static void ipi_cpu_stop(unsigned int cpu
)
537 if (system_state
== SYSTEM_BOOTING
||
538 system_state
== SYSTEM_RUNNING
) {
539 raw_spin_lock(&stop_lock
);
540 pr_crit("CPU%u: stopping\n", cpu
);
542 raw_spin_unlock(&stop_lock
);
545 set_cpu_online(cpu
, false);
554 static DEFINE_PER_CPU(struct completion
*, cpu_completion
);
556 int register_ipi_completion(struct completion
*completion
, int cpu
)
558 per_cpu(cpu_completion
, cpu
) = completion
;
559 return IPI_COMPLETION
;
562 static void ipi_complete(unsigned int cpu
)
564 complete(per_cpu(cpu_completion
, cpu
));
568 * Main handler for inter-processor interrupts
570 asmlinkage
void __exception_irq_entry
do_IPI(int ipinr
, struct pt_regs
*regs
)
572 handle_IPI(ipinr
, regs
);
575 void handle_IPI(int ipinr
, struct pt_regs
*regs
)
577 unsigned int cpu
= smp_processor_id();
578 struct pt_regs
*old_regs
= set_irq_regs(regs
);
580 if ((unsigned)ipinr
< NR_IPI
) {
581 trace_ipi_entry_rcuidle(ipi_types
[ipinr
]);
582 __inc_irq_stat(cpu
, ipi_irqs
[ipinr
]);
589 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
592 tick_receive_broadcast();
603 generic_smp_call_function_interrupt();
607 case IPI_CALL_FUNC_SINGLE
:
609 generic_smp_call_function_single_interrupt();
619 #ifdef CONFIG_IRQ_WORK
634 pr_crit("CPU%u: Unknown IPI message 0x%x\n",
639 if ((unsigned)ipinr
< NR_IPI
)
640 trace_ipi_exit_rcuidle(ipi_types
[ipinr
]);
641 set_irq_regs(old_regs
);
644 void smp_send_reschedule(int cpu
)
646 smp_cross_call(cpumask_of(cpu
), IPI_RESCHEDULE
);
649 void smp_send_stop(void)
651 unsigned long timeout
;
654 cpumask_copy(&mask
, cpu_online_mask
);
655 cpumask_clear_cpu(smp_processor_id(), &mask
);
656 if (!cpumask_empty(&mask
))
657 smp_cross_call(&mask
, IPI_CPU_STOP
);
659 /* Wait up to one second for other CPUs to stop */
660 timeout
= USEC_PER_SEC
;
661 while (num_online_cpus() > 1 && timeout
--)
664 if (num_online_cpus() > 1)
665 pr_warn("SMP: failed to stop secondary CPUs\n");
671 int setup_profiling_timer(unsigned int multiplier
)
676 #ifdef CONFIG_CPU_FREQ
678 static DEFINE_PER_CPU(unsigned long, l_p_j_ref
);
679 static DEFINE_PER_CPU(unsigned long, l_p_j_ref_freq
);
680 static unsigned long global_l_p_j_ref
;
681 static unsigned long global_l_p_j_ref_freq
;
683 static int cpufreq_callback(struct notifier_block
*nb
,
684 unsigned long val
, void *data
)
686 struct cpufreq_freqs
*freq
= data
;
689 if (freq
->flags
& CPUFREQ_CONST_LOOPS
)
692 if (!per_cpu(l_p_j_ref
, cpu
)) {
693 per_cpu(l_p_j_ref
, cpu
) =
694 per_cpu(cpu_data
, cpu
).loops_per_jiffy
;
695 per_cpu(l_p_j_ref_freq
, cpu
) = freq
->old
;
696 if (!global_l_p_j_ref
) {
697 global_l_p_j_ref
= loops_per_jiffy
;
698 global_l_p_j_ref_freq
= freq
->old
;
702 if ((val
== CPUFREQ_PRECHANGE
&& freq
->old
< freq
->new) ||
703 (val
== CPUFREQ_POSTCHANGE
&& freq
->old
> freq
->new)) {
704 loops_per_jiffy
= cpufreq_scale(global_l_p_j_ref
,
705 global_l_p_j_ref_freq
,
707 per_cpu(cpu_data
, cpu
).loops_per_jiffy
=
708 cpufreq_scale(per_cpu(l_p_j_ref
, cpu
),
709 per_cpu(l_p_j_ref_freq
, cpu
),
715 static struct notifier_block cpufreq_notifier
= {
716 .notifier_call
= cpufreq_callback
,
719 static int __init
register_cpufreq_notifier(void)
721 return cpufreq_register_notifier(&cpufreq_notifier
,
722 CPUFREQ_TRANSITION_NOTIFIER
);
724 core_initcall(register_cpufreq_notifier
);