2 * SMP initialisation and IPI support
3 * Based on arch/arm/kernel/smp.c
5 * Copyright (C) 2012 ARM Ltd.
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
16 * You should have received a copy of the GNU General Public License
17 * along with this program. If not, see <http://www.gnu.org/licenses/>.
20 #include <linux/acpi.h>
21 #include <linux/delay.h>
22 #include <linux/init.h>
23 #include <linux/spinlock.h>
24 #include <linux/sched.h>
25 #include <linux/interrupt.h>
26 #include <linux/cache.h>
27 #include <linux/profile.h>
28 #include <linux/errno.h>
30 #include <linux/err.h>
31 #include <linux/cpu.h>
32 #include <linux/smp.h>
33 #include <linux/seq_file.h>
34 #include <linux/irq.h>
35 #include <linux/percpu.h>
36 #include <linux/clockchips.h>
37 #include <linux/completion.h>
39 #include <linux/irq_work.h>
41 #include <asm/alternative.h>
42 #include <asm/atomic.h>
43 #include <asm/cacheflush.h>
45 #include <asm/cputype.h>
46 #include <asm/cpu_ops.h>
47 #include <asm/mmu_context.h>
48 #include <asm/pgtable.h>
49 #include <asm/pgalloc.h>
50 #include <asm/processor.h>
51 #include <asm/smp_plat.h>
52 #include <asm/sections.h>
53 #include <asm/tlbflush.h>
54 #include <asm/ptrace.h>
57 #define CREATE_TRACE_POINTS
58 #include <trace/events/ipi.h>
61 * as from 2.5, kernels no longer have an init_tasks structure
62 * so we need some other way of telling a new secondary core
63 * where to place its SVC stack
65 struct secondary_data secondary_data;
66 /* Number of CPUs which aren't online, but looping in kernel text. */
67 int cpus_stuck_in_kernel;
78 #ifdef CONFIG_HOTPLUG_CPU
79 static int op_cpu_kill(unsigned int cpu);
81 static inline int op_cpu_kill(unsigned int cpu)
89 * Boot a secondary CPU, and assign it the specified idle task.
90 * This also gives us the initial stack to use for this CPU.
92 static int boot_secondary(unsigned int cpu, struct task_struct *idle)
94 if (cpu_ops[cpu]->cpu_boot)
95 return cpu_ops[cpu]->cpu_boot(cpu);
100 static DECLARE_COMPLETION(cpu_running);
102 int __cpu_up(unsigned int cpu, struct task_struct *idle)
108 * We need to tell the secondary core where to find its stack and the
111 secondary_data.stack = task_stack_page(idle) + THREAD_START_SP;
112 update_cpu_boot_status(CPU_MMU_OFF);
113 __flush_dcache_area(&secondary_data, sizeof(secondary_data));
116 * Now bring the CPU into our world.
118 ret = boot_secondary(cpu, idle);
121 * CPU was successfully started, wait for it to come online or
124 wait_for_completion_timeout(&cpu_running,
125 msecs_to_jiffies(1000));
127 if (!cpu_online(cpu)) {
128 pr_crit("CPU%u: failed to come online\n", cpu);
132 pr_err("CPU%u: failed to boot: %d\n", cpu, ret);
135 secondary_data.stack = NULL;
136 status = READ_ONCE(secondary_data.status);
139 if (status == CPU_MMU_OFF)
140 status = READ_ONCE(__early_cpu_boot_status);
144 pr_err("CPU%u: failed in unknown state : 0x%lx\n",
148 if (!op_cpu_kill(cpu)) {
149 pr_crit("CPU%u: died during early boot\n", cpu);
153 pr_crit("CPU%u: may not have shut down cleanly\n", cpu);
154 case CPU_STUCK_IN_KERNEL:
155 pr_crit("CPU%u: is stuck in kernel\n", cpu);
156 cpus_stuck_in_kernel++;
158 case CPU_PANIC_KERNEL:
159 panic("CPU%u detected unsupported configuration\n", cpu);
166 static void smp_store_cpu_info(unsigned int cpuid)
168 store_cpu_topology(cpuid);
172 * This is the secondary CPU boot entry. We're using this CPUs
173 * idle thread stack, but a set of temporary page tables.
175 asmlinkage void secondary_start_kernel(void)
177 struct mm_struct *mm = &init_mm;
178 unsigned int cpu = smp_processor_id();
181 * All kernel threads share the same mm context; grab a
182 * reference and switch to it.
184 atomic_inc(&mm->mm_count);
185 current->active_mm = mm;
187 set_my_cpu_offset(per_cpu_offset(smp_processor_id()));
190 * TTBR0 is only used for the identity mapping at this stage. Make it
191 * point to zero page to avoid speculatively fetching new entries.
193 cpu_uninstall_idmap();
196 trace_hardirqs_off();
199 * If the system has established the capabilities, make sure
200 * this CPU ticks all of those. If it doesn't, the CPU will
201 * fail to come online.
203 verify_local_cpu_capabilities();
205 if (cpu_ops[cpu]->cpu_postboot)
206 cpu_ops[cpu]->cpu_postboot();
209 * Log the CPU info before it is marked online and might get read.
214 * Enable GIC and timers.
216 notify_cpu_starting(cpu);
218 smp_store_cpu_info(cpu);
221 * OK, now it's safe to let the boot CPU continue. Wait for
222 * the CPU migration code to notice that the CPU is online
223 * before we continue.
225 pr_info("CPU%u: Booted secondary processor [%08x]\n",
226 cpu, read_cpuid_id());
227 update_cpu_boot_status(CPU_BOOT_SUCCESS);
228 /* Make sure the status update is visible before we complete */
230 set_cpu_online(cpu, true);
231 complete(&cpu_running);
235 local_async_enable();
238 * OK, it's off to the idle thread for us
240 cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
243 #ifdef CONFIG_HOTPLUG_CPU
244 static int op_cpu_disable(unsigned int cpu)
247 * If we don't have a cpu_die method, abort before we reach the point
248 * of no return. CPU0 may not have an cpu_ops, so test for it.
250 if (!cpu_ops[cpu] || !cpu_ops[cpu]->cpu_die)
254 * We may need to abort a hot unplug for some other mechanism-specific
257 if (cpu_ops[cpu]->cpu_disable)
258 return cpu_ops[cpu]->cpu_disable(cpu);
264 * __cpu_disable runs on the processor to be shutdown.
266 int __cpu_disable(void)
268 unsigned int cpu = smp_processor_id();
271 ret = op_cpu_disable(cpu);
276 * Take this CPU offline. Once we clear this, we can't return,
277 * and we must not schedule until we're ready to give up the cpu.
279 set_cpu_online(cpu, false);
282 * OK - migrate IRQs away from this CPU
284 irq_migrate_all_off_this_cpu();
289 static int op_cpu_kill(unsigned int cpu)
292 * If we have no means of synchronising with the dying CPU, then assume
293 * that it is really dead. We can only wait for an arbitrary length of
294 * time and hope that it's dead, so let's skip the wait and just hope.
296 if (!cpu_ops[cpu]->cpu_kill)
299 return cpu_ops[cpu]->cpu_kill(cpu);
303 * called on the thread which is asking for a CPU to be shutdown -
304 * waits until shutdown has completed, or it is timed out.
306 void __cpu_die(unsigned int cpu)
310 if (!cpu_wait_death(cpu, 5)) {
311 pr_crit("CPU%u: cpu didn't die\n", cpu);
314 pr_notice("CPU%u: shutdown\n", cpu);
317 * Now that the dying CPU is beyond the point of no return w.r.t.
318 * in-kernel synchronisation, try to get the firwmare to help us to
319 * verify that it has really left the kernel before we consider
320 * clobbering anything it might still be using.
322 err = op_cpu_kill(cpu);
324 pr_warn("CPU%d may not have shut down cleanly: %d\n",
329 * Called from the idle thread for the CPU which has been shutdown.
331 * Note that we disable IRQs here, but do not re-enable them
332 * before returning to the caller. This is also the behaviour
333 * of the other hotplug-cpu capable cores, so presumably coming
334 * out of idle fixes this.
338 unsigned int cpu = smp_processor_id();
344 /* Tell __cpu_die() that this CPU is now safe to dispose of */
345 (void)cpu_report_death();
348 * Actually shutdown the CPU. This must never fail. The specific hotplug
349 * mechanism must perform all required cache maintenance to ensure that
350 * no dirty lines are lost in the process of shutting down the CPU.
352 cpu_ops[cpu]->cpu_die(cpu);
359 * Kill the calling secondary CPU, early in bringup before it is turned
362 void cpu_die_early(void)
364 int cpu = smp_processor_id();
366 pr_crit("CPU%d: will not boot\n", cpu);
368 /* Mark this CPU absent */
369 set_cpu_present(cpu, 0);
371 #ifdef CONFIG_HOTPLUG_CPU
372 update_cpu_boot_status(CPU_KILL_ME);
373 /* Check if we can park ourselves */
374 if (cpu_ops[cpu] && cpu_ops[cpu]->cpu_die)
375 cpu_ops[cpu]->cpu_die(cpu);
377 update_cpu_boot_status(CPU_STUCK_IN_KERNEL);
382 static void __init hyp_mode_check(void)
384 if (is_hyp_mode_available())
385 pr_info("CPU: All CPU(s) started at EL2\n");
386 else if (is_hyp_mode_mismatched())
387 WARN_TAINT(1, TAINT_CPU_OUT_OF_SPEC,
388 "CPU: CPUs started in inconsistent modes");
390 pr_info("CPU: All CPU(s) started at EL1\n");
393 void __init smp_cpus_done(unsigned int max_cpus)
395 pr_info("SMP: Total of %d processors activated.\n", num_online_cpus());
396 setup_cpu_features();
398 apply_alternatives_all();
401 void __init smp_prepare_boot_cpu(void)
403 cpuinfo_store_boot_cpu();
404 set_my_cpu_offset(per_cpu_offset(smp_processor_id()));
407 static u64 __init of_get_cpu_mpidr(struct device_node *dn)
413 * A cpu node with missing "reg" property is
414 * considered invalid to build a cpu_logical_map
417 cell = of_get_property(dn, "reg", NULL);
419 pr_err("%s: missing reg property\n", dn->full_name);
423 hwid = of_read_number(cell, of_n_addr_cells(dn));
425 * Non affinity bits must be set to 0 in the DT
427 if (hwid & ~MPIDR_HWID_BITMASK) {
428 pr_err("%s: invalid reg property\n", dn->full_name);
435 * Duplicate MPIDRs are a recipe for disaster. Scan all initialized
436 * entries and check for duplicates. If any is found just ignore the
437 * cpu. cpu_logical_map was initialized to INVALID_HWID to avoid
438 * matching valid MPIDR values.
440 static bool __init is_mpidr_duplicate(unsigned int cpu, u64 hwid)
444 for (i = 1; (i < cpu) && (i < NR_CPUS); i++)
445 if (cpu_logical_map(i) == hwid)
451 * Initialize cpu operations for a logical cpu and
452 * set it in the possible mask on success
454 static int __init smp_cpu_setup(int cpu)
456 if (cpu_read_ops(cpu))
459 if (cpu_ops[cpu]->cpu_init(cpu))
462 set_cpu_possible(cpu, true);
467 static bool bootcpu_valid __initdata;
468 static unsigned int cpu_count = 1;
472 * acpi_map_gic_cpu_interface - parse processor MADT entry
474 * Carry out sanity checks on MADT processor entry and initialize
475 * cpu_logical_map on success
478 acpi_map_gic_cpu_interface(struct acpi_madt_generic_interrupt *processor)
480 u64 hwid = processor->arm_mpidr;
482 if (!(processor->flags & ACPI_MADT_ENABLED)) {
483 pr_debug("skipping disabled CPU entry with 0x%llx MPIDR\n", hwid);
487 if (hwid & ~MPIDR_HWID_BITMASK || hwid == INVALID_HWID) {
488 pr_err("skipping CPU entry with invalid MPIDR 0x%llx\n", hwid);
492 if (is_mpidr_duplicate(cpu_count, hwid)) {
493 pr_err("duplicate CPU MPIDR 0x%llx in MADT\n", hwid);
497 /* Check if GICC structure of boot CPU is available in the MADT */
498 if (cpu_logical_map(0) == hwid) {
500 pr_err("duplicate boot CPU MPIDR: 0x%llx in MADT\n",
504 bootcpu_valid = true;
508 if (cpu_count >= NR_CPUS)
511 /* map the logical cpu id to cpu MPIDR */
512 cpu_logical_map(cpu_count) = hwid;
515 * Set-up the ACPI parking protocol cpu entries
516 * while initializing the cpu_logical_map to
517 * avoid parsing MADT entries multiple times for
518 * nothing (ie a valid cpu_logical_map entry should
519 * contain a valid parking protocol data set to
520 * initialize the cpu if the parking protocol is
521 * the only available enable method).
523 acpi_set_mailbox_entry(cpu_count, processor);
529 acpi_parse_gic_cpu_interface(struct acpi_subtable_header *header,
530 const unsigned long end)
532 struct acpi_madt_generic_interrupt *processor;
534 processor = (struct acpi_madt_generic_interrupt *)header;
535 if (BAD_MADT_GICC_ENTRY(processor, end))
538 acpi_table_print_madt_entry(header);
540 acpi_map_gic_cpu_interface(processor);
545 #define acpi_table_parse_madt(...) do { } while (0)
549 * Enumerate the possible CPU set from the device tree and build the
550 * cpu logical map array containing MPIDR values related to logical
551 * cpus. Assumes that cpu_logical_map(0) has already been initialized.
553 static void __init of_parse_and_init_cpus(void)
555 struct device_node *dn = NULL;
557 while ((dn = of_find_node_by_type(dn, "cpu"))) {
558 u64 hwid = of_get_cpu_mpidr(dn);
560 if (hwid == INVALID_HWID)
563 if (is_mpidr_duplicate(cpu_count, hwid)) {
564 pr_err("%s: duplicate cpu reg properties in the DT\n",
570 * The numbering scheme requires that the boot CPU
571 * must be assigned logical id 0. Record it so that
572 * the logical map built from DT is validated and can
575 if (hwid == cpu_logical_map(0)) {
577 pr_err("%s: duplicate boot cpu reg property in DT\n",
582 bootcpu_valid = true;
585 * cpu_logical_map has already been
586 * initialized and the boot cpu doesn't need
587 * the enable-method so continue without
593 if (cpu_count >= NR_CPUS)
596 pr_debug("cpu logical map 0x%llx\n", hwid);
597 cpu_logical_map(cpu_count) = hwid;
604 * Enumerate the possible CPU set from the device tree or ACPI and build the
605 * cpu logical map array containing MPIDR values related to logical
606 * cpus. Assumes that cpu_logical_map(0) has already been initialized.
608 void __init smp_init_cpus(void)
613 of_parse_and_init_cpus();
616 * do a walk of MADT to determine how many CPUs
617 * we have including disabled CPUs, and get information
618 * we need for SMP init
620 acpi_table_parse_madt(ACPI_MADT_TYPE_GENERIC_INTERRUPT,
621 acpi_parse_gic_cpu_interface, 0);
623 if (cpu_count > NR_CPUS)
624 pr_warn("no. of cores (%d) greater than configured maximum of %d - clipping\n",
627 if (!bootcpu_valid) {
628 pr_err("missing boot CPU MPIDR, not enabling secondaries\n");
633 * We need to set the cpu_logical_map entries before enabling
634 * the cpus so that cpu processor description entries (DT cpu nodes
635 * and ACPI MADT entries) can be retrieved by matching the cpu hwid
636 * with entries in cpu_logical_map while initializing the cpus.
637 * If the cpu set-up fails, invalidate the cpu_logical_map entry.
639 for (i = 1; i < NR_CPUS; i++) {
640 if (cpu_logical_map(i) != INVALID_HWID) {
641 if (smp_cpu_setup(i))
642 cpu_logical_map(i) = INVALID_HWID;
647 void __init smp_prepare_cpus(unsigned int max_cpus)
650 unsigned int cpu, ncores = num_possible_cpus();
654 smp_store_cpu_info(smp_processor_id());
657 * are we trying to boot more cores than exist?
659 if (max_cpus > ncores)
662 /* Don't bother if we're effectively UP */
667 * Initialise the present map (which describes the set of CPUs
668 * actually populated at the present time) and release the
669 * secondaries from the bootloader.
671 * Make sure we online at most (max_cpus - 1) additional CPUs.
674 for_each_possible_cpu(cpu) {
678 if (cpu == smp_processor_id())
684 err = cpu_ops[cpu]->cpu_prepare(cpu);
688 set_cpu_present(cpu, true);
693 void (*__smp_cross_call)(const struct cpumask *, unsigned int);
695 void __init set_smp_cross_call(void (*fn)(const struct cpumask *, unsigned int))
697 __smp_cross_call = fn;
700 static const char *ipi_types[NR_IPI] __tracepoint_string = {
701 #define S(x,s) [x] = s
702 S(IPI_RESCHEDULE, "Rescheduling interrupts"),
703 S(IPI_CALL_FUNC, "Function call interrupts"),
704 S(IPI_CPU_STOP, "CPU stop interrupts"),
705 S(IPI_TIMER, "Timer broadcast interrupts"),
706 S(IPI_IRQ_WORK, "IRQ work interrupts"),
707 S(IPI_WAKEUP, "CPU wake-up interrupts"),
710 static void smp_cross_call(const struct cpumask *target, unsigned int ipinr)
712 trace_ipi_raise(target, ipi_types[ipinr]);
713 __smp_cross_call(target, ipinr);
716 void show_ipi_list(struct seq_file *p, int prec)
720 for (i = 0; i < NR_IPI; i++) {
721 seq_printf(p, "%*s%u:%s", prec - 1, "IPI", i,
722 prec >= 4 ? " " : "");
723 for_each_online_cpu(cpu)
724 seq_printf(p, "%10u ",
725 __get_irq_stat(cpu, ipi_irqs[i]));
726 seq_printf(p, " %s\n", ipi_types[i]);
730 u64 smp_irq_stat_cpu(unsigned int cpu)
735 for (i = 0; i < NR_IPI; i++)
736 sum += __get_irq_stat(cpu, ipi_irqs[i]);
741 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
743 smp_cross_call(mask, IPI_CALL_FUNC);
746 void arch_send_call_function_single_ipi(int cpu)
748 smp_cross_call(cpumask_of(cpu), IPI_CALL_FUNC);
751 #ifdef CONFIG_ARM64_ACPI_PARKING_PROTOCOL
752 void arch_send_wakeup_ipi_mask(const struct cpumask *mask)
754 smp_cross_call(mask, IPI_WAKEUP);
758 #ifdef CONFIG_IRQ_WORK
759 void arch_irq_work_raise(void)
761 if (__smp_cross_call)
762 smp_cross_call(cpumask_of(smp_processor_id()), IPI_IRQ_WORK);
766 static DEFINE_RAW_SPINLOCK(stop_lock);
769 * ipi_cpu_stop - handle IPI from smp_send_stop()
771 static void ipi_cpu_stop(unsigned int cpu)
773 if (system_state == SYSTEM_BOOTING ||
774 system_state == SYSTEM_RUNNING) {
775 raw_spin_lock(&stop_lock);
776 pr_crit("CPU%u: stopping\n", cpu);
778 raw_spin_unlock(&stop_lock);
781 set_cpu_online(cpu, false);
790 * Main handler for inter-processor interrupts
792 void handle_IPI(int ipinr, struct pt_regs *regs)
794 unsigned int cpu = smp_processor_id();
795 struct pt_regs *old_regs = set_irq_regs(regs);
797 if ((unsigned)ipinr < NR_IPI) {
798 trace_ipi_entry_rcuidle(ipi_types[ipinr]);
799 __inc_irq_stat(cpu, ipi_irqs[ipinr]);
809 generic_smp_call_function_interrupt();
819 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
822 tick_receive_broadcast();
827 #ifdef CONFIG_IRQ_WORK
835 #ifdef CONFIG_ARM64_ACPI_PARKING_PROTOCOL
837 WARN_ONCE(!acpi_parking_protocol_valid(cpu),
838 "CPU%u: Wake-up IPI outside the ACPI parking protocol\n",
844 pr_crit("CPU%u: Unknown IPI message 0x%x\n", cpu, ipinr);
848 if ((unsigned)ipinr < NR_IPI)
849 trace_ipi_exit_rcuidle(ipi_types[ipinr]);
850 set_irq_regs(old_regs);
853 void smp_send_reschedule(int cpu)
855 smp_cross_call(cpumask_of(cpu), IPI_RESCHEDULE);
858 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
859 void tick_broadcast(const struct cpumask *mask)
861 smp_cross_call(mask, IPI_TIMER);
865 void smp_send_stop(void)
867 unsigned long timeout;
869 if (num_online_cpus() > 1) {
872 cpumask_copy(&mask, cpu_online_mask);
873 cpumask_clear_cpu(smp_processor_id(), &mask);
875 smp_cross_call(&mask, IPI_CPU_STOP);
878 /* Wait up to one second for other CPUs to stop */
879 timeout = USEC_PER_SEC;
880 while (num_online_cpus() > 1 && timeout--)
883 if (num_online_cpus() > 1)
884 pr_warning("SMP: failed to stop secondary CPUs\n");
890 int setup_profiling_timer(unsigned int multiplier)