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>
49 #include <asm/pgtable.h>
50 #include <asm/pgalloc.h>
51 #include <asm/processor.h>
52 #include <asm/smp_plat.h>
53 #include <asm/sections.h>
54 #include <asm/tlbflush.h>
55 #include <asm/ptrace.h>
58 #define CREATE_TRACE_POINTS
59 #include <trace/events/ipi.h>
61 DEFINE_PER_CPU_READ_MOSTLY(int, cpu_number);
62 EXPORT_PER_CPU_SYMBOL(cpu_number);
65 * as from 2.5, kernels no longer have an init_tasks structure
66 * so we need some other way of telling a new secondary core
67 * where to place its SVC stack
69 struct secondary_data secondary_data;
70 /* Number of CPUs which aren't online, but looping in kernel text. */
71 int cpus_stuck_in_kernel;
82 #ifdef CONFIG_ARM64_VHE
84 /* Whether the boot CPU is running in HYP mode or not*/
85 static bool boot_cpu_hyp_mode;
87 static inline void save_boot_cpu_run_el(void)
89 boot_cpu_hyp_mode = is_kernel_in_hyp_mode();
92 static inline bool is_boot_cpu_in_hyp_mode(void)
94 return boot_cpu_hyp_mode;
98 * Verify that a secondary CPU is running the kernel at the same
99 * EL as that of the boot CPU.
101 void verify_cpu_run_el(void)
103 bool in_el2 = is_kernel_in_hyp_mode();
104 bool boot_cpu_el2 = is_boot_cpu_in_hyp_mode();
106 if (in_el2 ^ boot_cpu_el2) {
107 pr_crit("CPU%d: mismatched Exception Level(EL%d) with boot CPU(EL%d)\n",
110 boot_cpu_el2 ? 2 : 1);
116 static inline void save_boot_cpu_run_el(void) {}
119 #ifdef CONFIG_HOTPLUG_CPU
120 static int op_cpu_kill(unsigned int cpu);
122 static inline int op_cpu_kill(unsigned int cpu)
130 * Boot a secondary CPU, and assign it the specified idle task.
131 * This also gives us the initial stack to use for this CPU.
133 static int boot_secondary(unsigned int cpu, struct task_struct *idle)
135 if (cpu_ops[cpu]->cpu_boot)
136 return cpu_ops[cpu]->cpu_boot(cpu);
141 static DECLARE_COMPLETION(cpu_running);
143 int __cpu_up(unsigned int cpu, struct task_struct *idle)
149 * We need to tell the secondary core where to find its stack and the
152 secondary_data.stack = task_stack_page(idle) + THREAD_START_SP;
153 update_cpu_boot_status(CPU_MMU_OFF);
154 __flush_dcache_area(&secondary_data, sizeof(secondary_data));
157 * Now bring the CPU into our world.
159 ret = boot_secondary(cpu, idle);
162 * CPU was successfully started, wait for it to come online or
165 wait_for_completion_timeout(&cpu_running,
166 msecs_to_jiffies(1000));
168 if (!cpu_online(cpu)) {
169 pr_crit("CPU%u: failed to come online\n", cpu);
173 pr_err("CPU%u: failed to boot: %d\n", cpu, ret);
176 secondary_data.stack = NULL;
177 status = READ_ONCE(secondary_data.status);
180 if (status == CPU_MMU_OFF)
181 status = READ_ONCE(__early_cpu_boot_status);
185 pr_err("CPU%u: failed in unknown state : 0x%lx\n",
189 if (!op_cpu_kill(cpu)) {
190 pr_crit("CPU%u: died during early boot\n", cpu);
194 pr_crit("CPU%u: may not have shut down cleanly\n", cpu);
195 case CPU_STUCK_IN_KERNEL:
196 pr_crit("CPU%u: is stuck in kernel\n", cpu);
197 cpus_stuck_in_kernel++;
199 case CPU_PANIC_KERNEL:
200 panic("CPU%u detected unsupported configuration\n", cpu);
208 * This is the secondary CPU boot entry. We're using this CPUs
209 * idle thread stack, but a set of temporary page tables.
211 asmlinkage void secondary_start_kernel(void)
213 struct mm_struct *mm = &init_mm;
216 cpu = task_cpu(current);
217 set_my_cpu_offset(per_cpu_offset(cpu));
220 * All kernel threads share the same mm context; grab a
221 * reference and switch to it.
223 atomic_inc(&mm->mm_count);
224 current->active_mm = mm;
227 * TTBR0 is only used for the identity mapping at this stage. Make it
228 * point to zero page to avoid speculatively fetching new entries.
230 cpu_uninstall_idmap();
233 trace_hardirqs_off();
236 * If the system has established the capabilities, make sure
237 * this CPU ticks all of those. If it doesn't, the CPU will
238 * fail to come online.
240 check_local_cpu_capabilities();
242 if (cpu_ops[cpu]->cpu_postboot)
243 cpu_ops[cpu]->cpu_postboot();
246 * Log the CPU info before it is marked online and might get read.
251 * Enable GIC and timers.
253 notify_cpu_starting(cpu);
255 store_cpu_topology(cpu);
258 * OK, now it's safe to let the boot CPU continue. Wait for
259 * the CPU migration code to notice that the CPU is online
260 * before we continue.
262 pr_info("CPU%u: Booted secondary processor [%08x]\n",
263 cpu, read_cpuid_id());
264 update_cpu_boot_status(CPU_BOOT_SUCCESS);
265 set_cpu_online(cpu, true);
266 complete(&cpu_running);
269 local_async_enable();
272 * OK, it's off to the idle thread for us
274 cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
277 #ifdef CONFIG_HOTPLUG_CPU
278 static int op_cpu_disable(unsigned int cpu)
281 * If we don't have a cpu_die method, abort before we reach the point
282 * of no return. CPU0 may not have an cpu_ops, so test for it.
284 if (!cpu_ops[cpu] || !cpu_ops[cpu]->cpu_die)
288 * We may need to abort a hot unplug for some other mechanism-specific
291 if (cpu_ops[cpu]->cpu_disable)
292 return cpu_ops[cpu]->cpu_disable(cpu);
298 * __cpu_disable runs on the processor to be shutdown.
300 int __cpu_disable(void)
302 unsigned int cpu = smp_processor_id();
305 ret = op_cpu_disable(cpu);
310 * Take this CPU offline. Once we clear this, we can't return,
311 * and we must not schedule until we're ready to give up the cpu.
313 set_cpu_online(cpu, false);
316 * OK - migrate IRQs away from this CPU
318 irq_migrate_all_off_this_cpu();
323 static int op_cpu_kill(unsigned int cpu)
326 * If we have no means of synchronising with the dying CPU, then assume
327 * that it is really dead. We can only wait for an arbitrary length of
328 * time and hope that it's dead, so let's skip the wait and just hope.
330 if (!cpu_ops[cpu]->cpu_kill)
333 return cpu_ops[cpu]->cpu_kill(cpu);
337 * called on the thread which is asking for a CPU to be shutdown -
338 * waits until shutdown has completed, or it is timed out.
340 void __cpu_die(unsigned int cpu)
344 if (!cpu_wait_death(cpu, 5)) {
345 pr_crit("CPU%u: cpu didn't die\n", cpu);
348 pr_notice("CPU%u: shutdown\n", cpu);
351 * Now that the dying CPU is beyond the point of no return w.r.t.
352 * in-kernel synchronisation, try to get the firwmare to help us to
353 * verify that it has really left the kernel before we consider
354 * clobbering anything it might still be using.
356 err = op_cpu_kill(cpu);
358 pr_warn("CPU%d may not have shut down cleanly: %d\n",
363 * Called from the idle thread for the CPU which has been shutdown.
365 * Note that we disable IRQs here, but do not re-enable them
366 * before returning to the caller. This is also the behaviour
367 * of the other hotplug-cpu capable cores, so presumably coming
368 * out of idle fixes this.
372 unsigned int cpu = smp_processor_id();
378 /* Tell __cpu_die() that this CPU is now safe to dispose of */
379 (void)cpu_report_death();
382 * Actually shutdown the CPU. This must never fail. The specific hotplug
383 * mechanism must perform all required cache maintenance to ensure that
384 * no dirty lines are lost in the process of shutting down the CPU.
386 cpu_ops[cpu]->cpu_die(cpu);
393 * Kill the calling secondary CPU, early in bringup before it is turned
396 void cpu_die_early(void)
398 int cpu = smp_processor_id();
400 pr_crit("CPU%d: will not boot\n", cpu);
402 /* Mark this CPU absent */
403 set_cpu_present(cpu, 0);
405 #ifdef CONFIG_HOTPLUG_CPU
406 update_cpu_boot_status(CPU_KILL_ME);
407 /* Check if we can park ourselves */
408 if (cpu_ops[cpu] && cpu_ops[cpu]->cpu_die)
409 cpu_ops[cpu]->cpu_die(cpu);
411 update_cpu_boot_status(CPU_STUCK_IN_KERNEL);
416 static void __init hyp_mode_check(void)
418 if (is_hyp_mode_available())
419 pr_info("CPU: All CPU(s) started at EL2\n");
420 else if (is_hyp_mode_mismatched())
421 WARN_TAINT(1, TAINT_CPU_OUT_OF_SPEC,
422 "CPU: CPUs started in inconsistent modes");
424 pr_info("CPU: All CPU(s) started at EL1\n");
427 void __init smp_cpus_done(unsigned int max_cpus)
429 pr_info("SMP: Total of %d processors activated.\n", num_online_cpus());
430 setup_cpu_features();
432 apply_alternatives_all();
435 void __init smp_prepare_boot_cpu(void)
437 set_my_cpu_offset(per_cpu_offset(smp_processor_id()));
439 * Initialise the static keys early as they may be enabled by the
443 cpuinfo_store_boot_cpu();
444 save_boot_cpu_run_el();
446 * Run the errata work around checks on the boot CPU, once we have
447 * initialised the cpu feature infrastructure from
448 * cpuinfo_store_boot_cpu() above.
450 update_cpu_errata_workarounds();
453 static u64 __init of_get_cpu_mpidr(struct device_node *dn)
459 * A cpu node with missing "reg" property is
460 * considered invalid to build a cpu_logical_map
463 cell = of_get_property(dn, "reg", NULL);
465 pr_err("%s: missing reg property\n", dn->full_name);
469 hwid = of_read_number(cell, of_n_addr_cells(dn));
471 * Non affinity bits must be set to 0 in the DT
473 if (hwid & ~MPIDR_HWID_BITMASK) {
474 pr_err("%s: invalid reg property\n", dn->full_name);
481 * Duplicate MPIDRs are a recipe for disaster. Scan all initialized
482 * entries and check for duplicates. If any is found just ignore the
483 * cpu. cpu_logical_map was initialized to INVALID_HWID to avoid
484 * matching valid MPIDR values.
486 static bool __init is_mpidr_duplicate(unsigned int cpu, u64 hwid)
490 for (i = 1; (i < cpu) && (i < NR_CPUS); i++)
491 if (cpu_logical_map(i) == hwid)
497 * Initialize cpu operations for a logical cpu and
498 * set it in the possible mask on success
500 static int __init smp_cpu_setup(int cpu)
502 if (cpu_read_ops(cpu))
505 if (cpu_ops[cpu]->cpu_init(cpu))
508 set_cpu_possible(cpu, true);
513 static bool bootcpu_valid __initdata;
514 static unsigned int cpu_count = 1;
518 * acpi_map_gic_cpu_interface - parse processor MADT entry
520 * Carry out sanity checks on MADT processor entry and initialize
521 * cpu_logical_map on success
524 acpi_map_gic_cpu_interface(struct acpi_madt_generic_interrupt *processor)
526 u64 hwid = processor->arm_mpidr;
528 if (!(processor->flags & ACPI_MADT_ENABLED)) {
529 pr_debug("skipping disabled CPU entry with 0x%llx MPIDR\n", hwid);
533 if (hwid & ~MPIDR_HWID_BITMASK || hwid == INVALID_HWID) {
534 pr_err("skipping CPU entry with invalid MPIDR 0x%llx\n", hwid);
538 if (is_mpidr_duplicate(cpu_count, hwid)) {
539 pr_err("duplicate CPU MPIDR 0x%llx in MADT\n", hwid);
543 /* Check if GICC structure of boot CPU is available in the MADT */
544 if (cpu_logical_map(0) == hwid) {
546 pr_err("duplicate boot CPU MPIDR: 0x%llx in MADT\n",
550 bootcpu_valid = true;
551 early_map_cpu_to_node(0, acpi_numa_get_nid(0, hwid));
555 if (cpu_count >= NR_CPUS)
558 /* map the logical cpu id to cpu MPIDR */
559 cpu_logical_map(cpu_count) = hwid;
562 * Set-up the ACPI parking protocol cpu entries
563 * while initializing the cpu_logical_map to
564 * avoid parsing MADT entries multiple times for
565 * nothing (ie a valid cpu_logical_map entry should
566 * contain a valid parking protocol data set to
567 * initialize the cpu if the parking protocol is
568 * the only available enable method).
570 acpi_set_mailbox_entry(cpu_count, processor);
572 early_map_cpu_to_node(cpu_count, acpi_numa_get_nid(cpu_count, hwid));
578 acpi_parse_gic_cpu_interface(struct acpi_subtable_header *header,
579 const unsigned long end)
581 struct acpi_madt_generic_interrupt *processor;
583 processor = (struct acpi_madt_generic_interrupt *)header;
584 if (BAD_MADT_GICC_ENTRY(processor, end))
587 acpi_table_print_madt_entry(header);
589 acpi_map_gic_cpu_interface(processor);
594 #define acpi_table_parse_madt(...) do { } while (0)
598 * Enumerate the possible CPU set from the device tree and build the
599 * cpu logical map array containing MPIDR values related to logical
600 * cpus. Assumes that cpu_logical_map(0) has already been initialized.
602 static void __init of_parse_and_init_cpus(void)
604 struct device_node *dn = NULL;
606 while ((dn = of_find_node_by_type(dn, "cpu"))) {
607 u64 hwid = of_get_cpu_mpidr(dn);
609 if (hwid == INVALID_HWID)
612 if (is_mpidr_duplicate(cpu_count, hwid)) {
613 pr_err("%s: duplicate cpu reg properties in the DT\n",
619 * The numbering scheme requires that the boot CPU
620 * must be assigned logical id 0. Record it so that
621 * the logical map built from DT is validated and can
624 if (hwid == cpu_logical_map(0)) {
626 pr_err("%s: duplicate boot cpu reg property in DT\n",
631 bootcpu_valid = true;
632 early_map_cpu_to_node(0, of_node_to_nid(dn));
635 * cpu_logical_map has already been
636 * initialized and the boot cpu doesn't need
637 * the enable-method so continue without
643 if (cpu_count >= NR_CPUS)
646 pr_debug("cpu logical map 0x%llx\n", hwid);
647 cpu_logical_map(cpu_count) = hwid;
649 early_map_cpu_to_node(cpu_count, of_node_to_nid(dn));
656 * Enumerate the possible CPU set from the device tree or ACPI and build the
657 * cpu logical map array containing MPIDR values related to logical
658 * cpus. Assumes that cpu_logical_map(0) has already been initialized.
660 void __init smp_init_cpus(void)
665 of_parse_and_init_cpus();
668 * do a walk of MADT to determine how many CPUs
669 * we have including disabled CPUs, and get information
670 * we need for SMP init
672 acpi_table_parse_madt(ACPI_MADT_TYPE_GENERIC_INTERRUPT,
673 acpi_parse_gic_cpu_interface, 0);
675 if (cpu_count > nr_cpu_ids)
676 pr_warn("Number of cores (%d) exceeds configured maximum of %d - clipping\n",
677 cpu_count, nr_cpu_ids);
679 if (!bootcpu_valid) {
680 pr_err("missing boot CPU MPIDR, not enabling secondaries\n");
685 * We need to set the cpu_logical_map entries before enabling
686 * the cpus so that cpu processor description entries (DT cpu nodes
687 * and ACPI MADT entries) can be retrieved by matching the cpu hwid
688 * with entries in cpu_logical_map while initializing the cpus.
689 * If the cpu set-up fails, invalidate the cpu_logical_map entry.
691 for (i = 1; i < nr_cpu_ids; i++) {
692 if (cpu_logical_map(i) != INVALID_HWID) {
693 if (smp_cpu_setup(i))
694 cpu_logical_map(i) = INVALID_HWID;
699 void __init smp_prepare_cpus(unsigned int max_cpus)
703 unsigned int this_cpu;
707 this_cpu = smp_processor_id();
708 store_cpu_topology(this_cpu);
709 numa_store_cpu_info(this_cpu);
712 * If UP is mandated by "nosmp" (which implies "maxcpus=0"), don't set
713 * secondary CPUs present.
719 * Initialise the present map (which describes the set of CPUs
720 * actually populated at the present time) and release the
721 * secondaries from the bootloader.
723 for_each_possible_cpu(cpu) {
725 per_cpu(cpu_number, cpu) = cpu;
727 if (cpu == smp_processor_id())
733 err = cpu_ops[cpu]->cpu_prepare(cpu);
737 set_cpu_present(cpu, true);
738 numa_store_cpu_info(cpu);
742 void (*__smp_cross_call)(const struct cpumask *, unsigned int);
744 void __init set_smp_cross_call(void (*fn)(const struct cpumask *, unsigned int))
746 __smp_cross_call = fn;
749 static const char *ipi_types[NR_IPI] __tracepoint_string = {
750 #define S(x,s) [x] = s
751 S(IPI_RESCHEDULE, "Rescheduling interrupts"),
752 S(IPI_CALL_FUNC, "Function call interrupts"),
753 S(IPI_CPU_STOP, "CPU stop interrupts"),
754 S(IPI_TIMER, "Timer broadcast interrupts"),
755 S(IPI_IRQ_WORK, "IRQ work interrupts"),
756 S(IPI_WAKEUP, "CPU wake-up interrupts"),
759 static void smp_cross_call(const struct cpumask *target, unsigned int ipinr)
761 trace_ipi_raise(target, ipi_types[ipinr]);
762 __smp_cross_call(target, ipinr);
765 void show_ipi_list(struct seq_file *p, int prec)
769 for (i = 0; i < NR_IPI; i++) {
770 seq_printf(p, "%*s%u:%s", prec - 1, "IPI", i,
771 prec >= 4 ? " " : "");
772 for_each_online_cpu(cpu)
773 seq_printf(p, "%10u ",
774 __get_irq_stat(cpu, ipi_irqs[i]));
775 seq_printf(p, " %s\n", ipi_types[i]);
779 u64 smp_irq_stat_cpu(unsigned int cpu)
784 for (i = 0; i < NR_IPI; i++)
785 sum += __get_irq_stat(cpu, ipi_irqs[i]);
790 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
792 smp_cross_call(mask, IPI_CALL_FUNC);
795 void arch_send_call_function_single_ipi(int cpu)
797 smp_cross_call(cpumask_of(cpu), IPI_CALL_FUNC);
800 #ifdef CONFIG_ARM64_ACPI_PARKING_PROTOCOL
801 void arch_send_wakeup_ipi_mask(const struct cpumask *mask)
803 smp_cross_call(mask, IPI_WAKEUP);
807 #ifdef CONFIG_IRQ_WORK
808 void arch_irq_work_raise(void)
810 if (__smp_cross_call)
811 smp_cross_call(cpumask_of(smp_processor_id()), IPI_IRQ_WORK);
816 * ipi_cpu_stop - handle IPI from smp_send_stop()
818 static void ipi_cpu_stop(unsigned int cpu)
820 set_cpu_online(cpu, false);
829 * Main handler for inter-processor interrupts
831 void handle_IPI(int ipinr, struct pt_regs *regs)
833 unsigned int cpu = smp_processor_id();
834 struct pt_regs *old_regs = set_irq_regs(regs);
836 if ((unsigned)ipinr < NR_IPI) {
837 trace_ipi_entry_rcuidle(ipi_types[ipinr]);
838 __inc_irq_stat(cpu, ipi_irqs[ipinr]);
848 generic_smp_call_function_interrupt();
858 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
861 tick_receive_broadcast();
866 #ifdef CONFIG_IRQ_WORK
874 #ifdef CONFIG_ARM64_ACPI_PARKING_PROTOCOL
876 WARN_ONCE(!acpi_parking_protocol_valid(cpu),
877 "CPU%u: Wake-up IPI outside the ACPI parking protocol\n",
883 pr_crit("CPU%u: Unknown IPI message 0x%x\n", cpu, ipinr);
887 if ((unsigned)ipinr < NR_IPI)
888 trace_ipi_exit_rcuidle(ipi_types[ipinr]);
889 set_irq_regs(old_regs);
892 void smp_send_reschedule(int cpu)
894 smp_cross_call(cpumask_of(cpu), IPI_RESCHEDULE);
897 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
898 void tick_broadcast(const struct cpumask *mask)
900 smp_cross_call(mask, IPI_TIMER);
904 void smp_send_stop(void)
906 unsigned long timeout;
908 if (num_online_cpus() > 1) {
911 cpumask_copy(&mask, cpu_online_mask);
912 cpumask_clear_cpu(smp_processor_id(), &mask);
914 if (system_state == SYSTEM_BOOTING ||
915 system_state == SYSTEM_RUNNING)
916 pr_crit("SMP: stopping secondary CPUs\n");
917 smp_cross_call(&mask, IPI_CPU_STOP);
920 /* Wait up to one second for other CPUs to stop */
921 timeout = USEC_PER_SEC;
922 while (num_online_cpus() > 1 && timeout--)
925 if (num_online_cpus() > 1)
926 pr_warning("SMP: failed to stop secondary CPUs %*pbl\n",
927 cpumask_pr_args(cpu_online_mask));
933 int setup_profiling_timer(unsigned int multiplier)
938 static bool have_cpu_die(void)
940 #ifdef CONFIG_HOTPLUG_CPU
941 int any_cpu = raw_smp_processor_id();
943 if (cpu_ops[any_cpu]->cpu_die)
949 bool cpus_are_stuck_in_kernel(void)
951 bool smp_spin_tables = (num_possible_cpus() > 1 && !have_cpu_die());
953 return !!cpus_stuck_in_kernel || smp_spin_tables;