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>
29 #include <linux/kernel_stat.h>
31 #include <linux/atomic.h>
34 #include <asm/cacheflush.h>
36 #include <asm/cputype.h>
37 #include <asm/exception.h>
38 #include <asm/idmap.h>
39 #include <asm/topology.h>
40 #include <asm/mmu_context.h>
41 #include <asm/procinfo.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
;
71 * CPU_BACKTRACE is special and not included in NR_IPI
72 * or tracable with trace_ipi_*
74 IPI_CPU_BACKTRACE
= NR_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.
83 static int ipi_irq_base __read_mostly
;
84 static int nr_ipi __read_mostly
= NR_IPI
;
85 static struct irq_desc
*ipi_desc
[MAX_IPI
] __read_mostly
;
87 static void ipi_setup(int cpu
);
88 static void ipi_teardown(int cpu
);
90 static DECLARE_COMPLETION(cpu_running
);
92 static struct smp_operations smp_ops __ro_after_init
;
94 void __init
smp_set_ops(const struct smp_operations
*ops
)
100 static unsigned long get_arch_pgd(pgd_t
*pgd
)
102 #ifdef CONFIG_ARM_LPAE
103 return __phys_to_pfn(virt_to_phys(pgd
));
105 return virt_to_phys(pgd
);
109 #if defined(CONFIG_BIG_LITTLE) && defined(CONFIG_HARDEN_BRANCH_PREDICTOR)
110 static int secondary_biglittle_prepare(unsigned int cpu
)
112 if (!cpu_vtable
[cpu
])
113 cpu_vtable
[cpu
] = kzalloc(sizeof(*cpu_vtable
[cpu
]), GFP_KERNEL
);
115 return cpu_vtable
[cpu
] ? 0 : -ENOMEM
;
118 static void secondary_biglittle_init(void)
120 init_proc_vtable(lookup_processor(read_cpuid_id())->proc
);
123 static int secondary_biglittle_prepare(unsigned int cpu
)
128 static void secondary_biglittle_init(void)
133 int __cpu_up(unsigned int cpu
, struct task_struct
*idle
)
137 if (!smp_ops
.smp_boot_secondary
)
140 ret
= secondary_biglittle_prepare(cpu
);
145 * We need to tell the secondary core where to find
146 * its stack and the page tables.
148 secondary_data
.stack
= task_stack_page(idle
) + THREAD_START_SP
;
149 #ifdef CONFIG_ARM_MPU
150 secondary_data
.mpu_rgn_info
= &mpu_rgn_info
;
154 secondary_data
.pgdir
= virt_to_phys(idmap_pgd
);
155 secondary_data
.swapper_pg_dir
= get_arch_pgd(swapper_pg_dir
);
157 sync_cache_w(&secondary_data
);
160 * Now bring the CPU into our world.
162 ret
= smp_ops
.smp_boot_secondary(cpu
, idle
);
165 * CPU was successfully started, wait for it
166 * to come online or time out.
168 wait_for_completion_timeout(&cpu_running
,
169 msecs_to_jiffies(1000));
171 if (!cpu_online(cpu
)) {
172 pr_crit("CPU%u: failed to come online\n", cpu
);
176 pr_err("CPU%u: failed to boot: %d\n", cpu
, ret
);
180 memset(&secondary_data
, 0, sizeof(secondary_data
));
184 /* platform specific SMP operations */
185 void __init
smp_init_cpus(void)
187 if (smp_ops
.smp_init_cpus
)
188 smp_ops
.smp_init_cpus();
191 int platform_can_secondary_boot(void)
193 return !!smp_ops
.smp_boot_secondary
;
196 int platform_can_cpu_hotplug(void)
198 #ifdef CONFIG_HOTPLUG_CPU
199 if (smp_ops
.cpu_kill
)
206 #ifdef CONFIG_HOTPLUG_CPU
207 static int platform_cpu_kill(unsigned int cpu
)
209 if (smp_ops
.cpu_kill
)
210 return smp_ops
.cpu_kill(cpu
);
214 static int platform_cpu_disable(unsigned int cpu
)
216 if (smp_ops
.cpu_disable
)
217 return smp_ops
.cpu_disable(cpu
);
222 int platform_can_hotplug_cpu(unsigned int cpu
)
224 /* cpu_die must be specified to support hotplug */
225 if (!smp_ops
.cpu_die
)
228 if (smp_ops
.cpu_can_disable
)
229 return smp_ops
.cpu_can_disable(cpu
);
232 * By default, allow disabling all CPUs except the first one,
233 * since this is special on a lot of platforms, e.g. because
234 * of clock tick interrupts.
240 * __cpu_disable runs on the processor to be shutdown.
242 int __cpu_disable(void)
244 unsigned int cpu
= smp_processor_id();
247 ret
= platform_cpu_disable(cpu
);
251 #ifdef CONFIG_GENERIC_ARCH_TOPOLOGY
252 remove_cpu_topology(cpu
);
256 * Take this CPU offline. Once we clear this, we can't return,
257 * and we must not schedule until we're ready to give up the cpu.
259 set_cpu_online(cpu
, false);
263 * OK - migrate IRQs away from this CPU
265 irq_migrate_all_off_this_cpu();
268 * Flush user cache and TLB mappings, and then remove this CPU
269 * from the vm mask set of all processes.
271 * Caches are flushed to the Level of Unification Inner Shareable
272 * to write-back dirty lines to unified caches shared by all CPUs.
275 local_flush_tlb_all();
281 * called on the thread which is asking for a CPU to be shutdown -
282 * waits until shutdown has completed, or it is timed out.
284 void __cpu_die(unsigned int cpu
)
286 if (!cpu_wait_death(cpu
, 5)) {
287 pr_err("CPU%u: cpu didn't die\n", cpu
);
290 pr_debug("CPU%u: shutdown\n", cpu
);
292 clear_tasks_mm_cpumask(cpu
);
294 * platform_cpu_kill() is generally expected to do the powering off
295 * and/or cutting of clocks to the dying CPU. Optionally, this may
296 * be done by the CPU which is dying in preference to supporting
297 * this call, but that means there is _no_ synchronisation between
298 * the requesting CPU and the dying CPU actually losing power.
300 if (!platform_cpu_kill(cpu
))
301 pr_err("CPU%u: unable to kill\n", cpu
);
305 * Called from the idle thread for the CPU which has been shutdown.
307 * Note that we disable IRQs here, but do not re-enable them
308 * before returning to the caller. This is also the behaviour
309 * of the other hotplug-cpu capable cores, so presumably coming
310 * out of idle fixes this.
312 void arch_cpu_idle_dead(void)
314 unsigned int cpu
= smp_processor_id();
321 * Flush the data out of the L1 cache for this CPU. This must be
322 * before the completion to ensure that data is safely written out
323 * before platform_cpu_kill() gets called - which may disable
324 * *this* CPU and power down its cache.
329 * Tell __cpu_die() that this CPU is now safe to dispose of. Once
330 * this returns, power and/or clocks can be removed at any point
331 * from this CPU and its cache by platform_cpu_kill().
333 (void)cpu_report_death();
336 * Ensure that the cache lines associated with that completion are
337 * written out. This covers the case where _this_ CPU is doing the
338 * powering down, to ensure that the completion is visible to the
339 * CPU waiting for this one.
344 * The actual CPU shutdown procedure is at least platform (if not
345 * CPU) specific. This may remove power, or it may simply spin.
347 * Platforms are generally expected *NOT* to return from this call,
348 * although there are some which do because they have no way to
349 * power down the CPU. These platforms are the _only_ reason we
350 * have a return path which uses the fragment of assembly below.
352 * The return path should not be used for platforms which can
356 smp_ops
.cpu_die(cpu
);
358 pr_warn("CPU%u: smp_ops.cpu_die() returned, trying to resuscitate\n",
362 * Do not return to the idle loop - jump back to the secondary
363 * cpu initialisation. There's some initialisation which needs
364 * to be repeated to undo the effects of taking the CPU offline.
366 __asm__("mov sp, %0\n"
368 " b secondary_start_kernel"
370 : "r" (task_stack_page(current
) + THREAD_SIZE
- 8));
372 #endif /* CONFIG_HOTPLUG_CPU */
375 * Called by both boot and secondaries to move global data into
376 * per-processor storage.
378 static void smp_store_cpu_info(unsigned int cpuid
)
380 struct cpuinfo_arm
*cpu_info
= &per_cpu(cpu_data
, cpuid
);
382 cpu_info
->loops_per_jiffy
= loops_per_jiffy
;
383 cpu_info
->cpuid
= read_cpuid_id();
385 store_cpu_topology(cpuid
);
386 check_cpu_icache_size(cpuid
);
390 * This is the secondary CPU boot entry. We're using this CPUs
391 * idle thread stack, but a set of temporary page tables.
393 asmlinkage
void secondary_start_kernel(void)
395 struct mm_struct
*mm
= &init_mm
;
398 secondary_biglittle_init();
401 * The identity mapping is uncached (strongly ordered), so
402 * switch away from it before attempting any exclusive accesses.
404 cpu_switch_mm(mm
->pgd
, mm
);
405 local_flush_bp_all();
406 enter_lazy_tlb(mm
, current
);
407 local_flush_tlb_all();
410 * All kernel threads share the same mm context; grab a
411 * reference and switch to it.
413 cpu
= smp_processor_id();
415 current
->active_mm
= mm
;
416 cpumask_set_cpu(cpu
, mm_cpumask(mm
));
421 setup_vectors_base();
423 pr_debug("CPU%u: Booted secondary processor\n", cpu
);
426 trace_hardirqs_off();
429 * Give the platform a chance to do its own initialisation.
431 if (smp_ops
.smp_secondary_init
)
432 smp_ops
.smp_secondary_init(cpu
);
434 notify_cpu_starting(cpu
);
440 smp_store_cpu_info(cpu
);
443 * OK, now it's safe to let the boot CPU continue. Wait for
444 * the CPU migration code to notice that the CPU is online
445 * before we continue - which happens after __cpu_up returns.
447 set_cpu_online(cpu
, true);
451 complete(&cpu_running
);
458 * OK, it's off to the idle thread for us
460 cpu_startup_entry(CPUHP_AP_ONLINE_IDLE
);
463 void __init
smp_cpus_done(unsigned int max_cpus
)
466 unsigned long bogosum
= 0;
468 for_each_online_cpu(cpu
)
469 bogosum
+= per_cpu(cpu_data
, cpu
).loops_per_jiffy
;
471 printk(KERN_INFO
"SMP: Total of %d processors activated "
472 "(%lu.%02lu BogoMIPS).\n",
474 bogosum
/ (500000/HZ
),
475 (bogosum
/ (5000/HZ
)) % 100);
480 void __init
smp_prepare_boot_cpu(void)
482 set_my_cpu_offset(per_cpu_offset(smp_processor_id()));
485 void __init
smp_prepare_cpus(unsigned int max_cpus
)
487 unsigned int ncores
= num_possible_cpus();
491 smp_store_cpu_info(smp_processor_id());
494 * are we trying to boot more cores than exist?
496 if (max_cpus
> ncores
)
498 if (ncores
> 1 && max_cpus
) {
500 * Initialise the present map, which describes the set of CPUs
501 * actually populated at the present time. A platform should
502 * re-initialize the map in the platforms smp_prepare_cpus()
503 * if present != possible (e.g. physical hotplug).
505 init_cpu_present(cpu_possible_mask
);
508 * Initialise the SCU if there are more than one CPU
509 * and let them know where to start.
511 if (smp_ops
.smp_prepare_cpus
)
512 smp_ops
.smp_prepare_cpus(max_cpus
);
516 static const char *ipi_types
[NR_IPI
] __tracepoint_string
= {
517 #define S(x,s) [x] = s
518 S(IPI_WAKEUP
, "CPU wakeup interrupts"),
519 S(IPI_TIMER
, "Timer broadcast interrupts"),
520 S(IPI_RESCHEDULE
, "Rescheduling interrupts"),
521 S(IPI_CALL_FUNC
, "Function call interrupts"),
522 S(IPI_CPU_STOP
, "CPU stop interrupts"),
523 S(IPI_IRQ_WORK
, "IRQ work interrupts"),
524 S(IPI_COMPLETION
, "completion interrupts"),
527 static void smp_cross_call(const struct cpumask
*target
, unsigned int ipinr
);
529 void show_ipi_list(struct seq_file
*p
, int prec
)
533 for (i
= 0; i
< NR_IPI
; i
++) {
534 unsigned int irq
= irq_desc_get_irq(ipi_desc
[i
]);
535 seq_printf(p
, "%*s%u: ", prec
- 1, "IPI", i
);
537 for_each_online_cpu(cpu
)
538 seq_printf(p
, "%10u ", kstat_irqs_cpu(irq
, cpu
));
540 seq_printf(p
, " %s\n", ipi_types
[i
]);
544 void arch_send_call_function_ipi_mask(const struct cpumask
*mask
)
546 smp_cross_call(mask
, IPI_CALL_FUNC
);
549 void arch_send_wakeup_ipi_mask(const struct cpumask
*mask
)
551 smp_cross_call(mask
, IPI_WAKEUP
);
554 void arch_send_call_function_single_ipi(int cpu
)
556 smp_cross_call(cpumask_of(cpu
), IPI_CALL_FUNC
);
559 #ifdef CONFIG_IRQ_WORK
560 void arch_irq_work_raise(void)
562 if (arch_irq_work_has_interrupt())
563 smp_cross_call(cpumask_of(smp_processor_id()), IPI_IRQ_WORK
);
567 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
568 void tick_broadcast(const struct cpumask
*mask
)
570 smp_cross_call(mask
, IPI_TIMER
);
574 static DEFINE_RAW_SPINLOCK(stop_lock
);
577 * ipi_cpu_stop - handle IPI from smp_send_stop()
579 static void ipi_cpu_stop(unsigned int cpu
)
581 if (system_state
<= SYSTEM_RUNNING
) {
582 raw_spin_lock(&stop_lock
);
583 pr_crit("CPU%u: stopping\n", cpu
);
585 raw_spin_unlock(&stop_lock
);
588 set_cpu_online(cpu
, false);
599 static DEFINE_PER_CPU(struct completion
*, cpu_completion
);
601 int register_ipi_completion(struct completion
*completion
, int cpu
)
603 per_cpu(cpu_completion
, cpu
) = completion
;
604 return IPI_COMPLETION
;
607 static void ipi_complete(unsigned int cpu
)
609 complete(per_cpu(cpu_completion
, cpu
));
613 * Main handler for inter-processor interrupts
615 asmlinkage
void __exception_irq_entry
do_IPI(int ipinr
, struct pt_regs
*regs
)
617 handle_IPI(ipinr
, regs
);
620 static void do_handle_IPI(int ipinr
)
622 unsigned int cpu
= smp_processor_id();
624 if ((unsigned)ipinr
< NR_IPI
)
625 trace_ipi_entry_rcuidle(ipi_types
[ipinr
]);
631 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
633 tick_receive_broadcast();
642 generic_smp_call_function_interrupt();
649 #ifdef CONFIG_IRQ_WORK
659 case IPI_CPU_BACKTRACE
:
661 nmi_cpu_backtrace(get_irq_regs());
666 pr_crit("CPU%u: Unknown IPI message 0x%x\n",
671 if ((unsigned)ipinr
< NR_IPI
)
672 trace_ipi_exit_rcuidle(ipi_types
[ipinr
]);
675 /* Legacy version, should go away once all irqchips have been converted */
676 void handle_IPI(int ipinr
, struct pt_regs
*regs
)
678 struct pt_regs
*old_regs
= set_irq_regs(regs
);
681 do_handle_IPI(ipinr
);
684 set_irq_regs(old_regs
);
687 static irqreturn_t
ipi_handler(int irq
, void *data
)
689 do_handle_IPI(irq
- ipi_irq_base
);
693 static void smp_cross_call(const struct cpumask
*target
, unsigned int ipinr
)
695 trace_ipi_raise_rcuidle(target
, ipi_types
[ipinr
]);
696 __ipi_send_mask(ipi_desc
[ipinr
], target
);
699 static void ipi_setup(int cpu
)
703 if (WARN_ON_ONCE(!ipi_irq_base
))
706 for (i
= 0; i
< nr_ipi
; i
++)
707 enable_percpu_irq(ipi_irq_base
+ i
, 0);
710 static void ipi_teardown(int cpu
)
714 if (WARN_ON_ONCE(!ipi_irq_base
))
717 for (i
= 0; i
< nr_ipi
; i
++)
718 disable_percpu_irq(ipi_irq_base
+ i
);
721 void __init
set_smp_ipi_range(int ipi_base
, int n
)
725 WARN_ON(n
< MAX_IPI
);
726 nr_ipi
= min(n
, MAX_IPI
);
728 for (i
= 0; i
< nr_ipi
; i
++) {
731 err
= request_percpu_irq(ipi_base
+ i
, ipi_handler
,
735 ipi_desc
[i
] = irq_to_desc(ipi_base
+ i
);
736 irq_set_status_flags(ipi_base
+ i
, IRQ_HIDDEN
);
739 ipi_irq_base
= ipi_base
;
741 /* Setup the boot CPU immediately */
742 ipi_setup(smp_processor_id());
745 void smp_send_reschedule(int cpu
)
747 smp_cross_call(cpumask_of(cpu
), IPI_RESCHEDULE
);
750 void smp_send_stop(void)
752 unsigned long timeout
;
755 cpumask_copy(&mask
, cpu_online_mask
);
756 cpumask_clear_cpu(smp_processor_id(), &mask
);
757 if (!cpumask_empty(&mask
))
758 smp_cross_call(&mask
, IPI_CPU_STOP
);
760 /* Wait up to one second for other CPUs to stop */
761 timeout
= USEC_PER_SEC
;
762 while (num_online_cpus() > 1 && timeout
--)
765 if (num_online_cpus() > 1)
766 pr_warn("SMP: failed to stop secondary CPUs\n");
769 /* In case panic() and panic() called at the same time on CPU1 and CPU2,
770 * and CPU 1 calls panic_smp_self_stop() before crash_smp_send_stop()
771 * CPU1 can't receive the ipi irqs from CPU2, CPU1 will be always online,
772 * kdump fails. So split out the panic_smp_self_stop() and add
773 * set_cpu_online(smp_processor_id(), false).
775 void panic_smp_self_stop(void)
777 pr_debug("CPU %u will stop doing anything useful since another CPU has paniced\n",
779 set_cpu_online(smp_processor_id(), false);
787 int setup_profiling_timer(unsigned int multiplier
)
792 #ifdef CONFIG_CPU_FREQ
794 static DEFINE_PER_CPU(unsigned long, l_p_j_ref
);
795 static DEFINE_PER_CPU(unsigned long, l_p_j_ref_freq
);
796 static unsigned long global_l_p_j_ref
;
797 static unsigned long global_l_p_j_ref_freq
;
799 static int cpufreq_callback(struct notifier_block
*nb
,
800 unsigned long val
, void *data
)
802 struct cpufreq_freqs
*freq
= data
;
803 struct cpumask
*cpus
= freq
->policy
->cpus
;
804 int cpu
, first
= cpumask_first(cpus
);
807 if (freq
->flags
& CPUFREQ_CONST_LOOPS
)
810 if (!per_cpu(l_p_j_ref
, first
)) {
811 for_each_cpu(cpu
, cpus
) {
812 per_cpu(l_p_j_ref
, cpu
) =
813 per_cpu(cpu_data
, cpu
).loops_per_jiffy
;
814 per_cpu(l_p_j_ref_freq
, cpu
) = freq
->old
;
817 if (!global_l_p_j_ref
) {
818 global_l_p_j_ref
= loops_per_jiffy
;
819 global_l_p_j_ref_freq
= freq
->old
;
823 if ((val
== CPUFREQ_PRECHANGE
&& freq
->old
< freq
->new) ||
824 (val
== CPUFREQ_POSTCHANGE
&& freq
->old
> freq
->new)) {
825 loops_per_jiffy
= cpufreq_scale(global_l_p_j_ref
,
826 global_l_p_j_ref_freq
,
829 lpj
= cpufreq_scale(per_cpu(l_p_j_ref
, first
),
830 per_cpu(l_p_j_ref_freq
, first
), freq
->new);
831 for_each_cpu(cpu
, cpus
)
832 per_cpu(cpu_data
, cpu
).loops_per_jiffy
= lpj
;
837 static struct notifier_block cpufreq_notifier
= {
838 .notifier_call
= cpufreq_callback
,
841 static int __init
register_cpufreq_notifier(void)
843 return cpufreq_register_notifier(&cpufreq_notifier
,
844 CPUFREQ_TRANSITION_NOTIFIER
);
846 core_initcall(register_cpufreq_notifier
);
850 static void raise_nmi(cpumask_t
*mask
)
852 __ipi_send_mask(ipi_desc
[IPI_CPU_BACKTRACE
], mask
);
855 void arch_trigger_cpumask_backtrace(const cpumask_t
*mask
, bool exclude_self
)
857 nmi_trigger_cpumask_backtrace(mask
, exclude_self
, raise_nmi
);