xref: /linux/arch/arm64/kernel/smp.c (revision 9a6b55ac)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * SMP initialisation and IPI support
4  * Based on arch/arm/kernel/smp.c
5  *
6  * Copyright (C) 2012 ARM Ltd.
7  */
8 
9 #include <linux/acpi.h>
10 #include <linux/arm_sdei.h>
11 #include <linux/delay.h>
12 #include <linux/init.h>
13 #include <linux/spinlock.h>
14 #include <linux/sched/mm.h>
15 #include <linux/sched/hotplug.h>
16 #include <linux/sched/task_stack.h>
17 #include <linux/interrupt.h>
18 #include <linux/cache.h>
19 #include <linux/profile.h>
20 #include <linux/errno.h>
21 #include <linux/mm.h>
22 #include <linux/err.h>
23 #include <linux/cpu.h>
24 #include <linux/smp.h>
25 #include <linux/seq_file.h>
26 #include <linux/irq.h>
27 #include <linux/irqchip/arm-gic-v3.h>
28 #include <linux/percpu.h>
29 #include <linux/clockchips.h>
30 #include <linux/completion.h>
31 #include <linux/of.h>
32 #include <linux/irq_work.h>
33 #include <linux/kexec.h>
34 #include <linux/kvm_host.h>
35 
36 #include <asm/alternative.h>
37 #include <asm/atomic.h>
38 #include <asm/cacheflush.h>
39 #include <asm/cpu.h>
40 #include <asm/cputype.h>
41 #include <asm/cpu_ops.h>
42 #include <asm/daifflags.h>
43 #include <asm/kvm_mmu.h>
44 #include <asm/mmu_context.h>
45 #include <asm/numa.h>
46 #include <asm/pgtable.h>
47 #include <asm/pgalloc.h>
48 #include <asm/processor.h>
49 #include <asm/smp_plat.h>
50 #include <asm/sections.h>
51 #include <asm/tlbflush.h>
52 #include <asm/ptrace.h>
53 #include <asm/virt.h>
54 
55 #define CREATE_TRACE_POINTS
56 #include <trace/events/ipi.h>
57 
58 DEFINE_PER_CPU_READ_MOSTLY(int, cpu_number);
59 EXPORT_PER_CPU_SYMBOL(cpu_number);
60 
61 /*
62  * as from 2.5, kernels no longer have an init_tasks structure
63  * so we need some other way of telling a new secondary core
64  * where to place its SVC stack
65  */
66 struct secondary_data secondary_data;
67 /* Number of CPUs which aren't online, but looping in kernel text. */
68 int cpus_stuck_in_kernel;
69 
70 enum ipi_msg_type {
71 	IPI_RESCHEDULE,
72 	IPI_CALL_FUNC,
73 	IPI_CPU_STOP,
74 	IPI_CPU_CRASH_STOP,
75 	IPI_TIMER,
76 	IPI_IRQ_WORK,
77 	IPI_WAKEUP
78 };
79 
80 #ifdef CONFIG_HOTPLUG_CPU
81 static int op_cpu_kill(unsigned int cpu);
82 #else
83 static inline int op_cpu_kill(unsigned int cpu)
84 {
85 	return -ENOSYS;
86 }
87 #endif
88 
89 
90 /*
91  * Boot a secondary CPU, and assign it the specified idle task.
92  * This also gives us the initial stack to use for this CPU.
93  */
94 static int boot_secondary(unsigned int cpu, struct task_struct *idle)
95 {
96 	if (cpu_ops[cpu]->cpu_boot)
97 		return cpu_ops[cpu]->cpu_boot(cpu);
98 
99 	return -EOPNOTSUPP;
100 }
101 
102 static DECLARE_COMPLETION(cpu_running);
103 
104 int __cpu_up(unsigned int cpu, struct task_struct *idle)
105 {
106 	int ret;
107 	long status;
108 
109 	/*
110 	 * We need to tell the secondary core where to find its stack and the
111 	 * page tables.
112 	 */
113 	secondary_data.task = idle;
114 	secondary_data.stack = task_stack_page(idle) + THREAD_SIZE;
115 	update_cpu_boot_status(CPU_MMU_OFF);
116 	__flush_dcache_area(&secondary_data, sizeof(secondary_data));
117 
118 	/*
119 	 * Now bring the CPU into our world.
120 	 */
121 	ret = boot_secondary(cpu, idle);
122 	if (ret == 0) {
123 		/*
124 		 * CPU was successfully started, wait for it to come online or
125 		 * time out.
126 		 */
127 		wait_for_completion_timeout(&cpu_running,
128 					    msecs_to_jiffies(5000));
129 
130 		if (!cpu_online(cpu)) {
131 			pr_crit("CPU%u: failed to come online\n", cpu);
132 			ret = -EIO;
133 		}
134 	} else {
135 		pr_err("CPU%u: failed to boot: %d\n", cpu, ret);
136 		return ret;
137 	}
138 
139 	secondary_data.task = NULL;
140 	secondary_data.stack = NULL;
141 	__flush_dcache_area(&secondary_data, sizeof(secondary_data));
142 	status = READ_ONCE(secondary_data.status);
143 	if (ret && status) {
144 
145 		if (status == CPU_MMU_OFF)
146 			status = READ_ONCE(__early_cpu_boot_status);
147 
148 		switch (status & CPU_BOOT_STATUS_MASK) {
149 		default:
150 			pr_err("CPU%u: failed in unknown state : 0x%lx\n",
151 					cpu, status);
152 			cpus_stuck_in_kernel++;
153 			break;
154 		case CPU_KILL_ME:
155 			if (!op_cpu_kill(cpu)) {
156 				pr_crit("CPU%u: died during early boot\n", cpu);
157 				break;
158 			}
159 			pr_crit("CPU%u: may not have shut down cleanly\n", cpu);
160 			/* Fall through */
161 		case CPU_STUCK_IN_KERNEL:
162 			pr_crit("CPU%u: is stuck in kernel\n", cpu);
163 			if (status & CPU_STUCK_REASON_52_BIT_VA)
164 				pr_crit("CPU%u: does not support 52-bit VAs\n", cpu);
165 			if (status & CPU_STUCK_REASON_NO_GRAN)
166 				pr_crit("CPU%u: does not support %luK granule \n", cpu, PAGE_SIZE / SZ_1K);
167 			cpus_stuck_in_kernel++;
168 			break;
169 		case CPU_PANIC_KERNEL:
170 			panic("CPU%u detected unsupported configuration\n", cpu);
171 		}
172 	}
173 
174 	return ret;
175 }
176 
177 static void init_gic_priority_masking(void)
178 {
179 	u32 cpuflags;
180 
181 	if (WARN_ON(!gic_enable_sre()))
182 		return;
183 
184 	cpuflags = read_sysreg(daif);
185 
186 	WARN_ON(!(cpuflags & PSR_I_BIT));
187 
188 	gic_write_pmr(GIC_PRIO_IRQON | GIC_PRIO_PSR_I_SET);
189 }
190 
191 /*
192  * This is the secondary CPU boot entry.  We're using this CPUs
193  * idle thread stack, but a set of temporary page tables.
194  */
195 asmlinkage notrace void secondary_start_kernel(void)
196 {
197 	u64 mpidr = read_cpuid_mpidr() & MPIDR_HWID_BITMASK;
198 	struct mm_struct *mm = &init_mm;
199 	unsigned int cpu;
200 
201 	cpu = task_cpu(current);
202 	set_my_cpu_offset(per_cpu_offset(cpu));
203 
204 	/*
205 	 * All kernel threads share the same mm context; grab a
206 	 * reference and switch to it.
207 	 */
208 	mmgrab(mm);
209 	current->active_mm = mm;
210 
211 	/*
212 	 * TTBR0 is only used for the identity mapping at this stage. Make it
213 	 * point to zero page to avoid speculatively fetching new entries.
214 	 */
215 	cpu_uninstall_idmap();
216 
217 	if (system_uses_irq_prio_masking())
218 		init_gic_priority_masking();
219 
220 	preempt_disable();
221 	trace_hardirqs_off();
222 
223 	/*
224 	 * If the system has established the capabilities, make sure
225 	 * this CPU ticks all of those. If it doesn't, the CPU will
226 	 * fail to come online.
227 	 */
228 	check_local_cpu_capabilities();
229 
230 	if (cpu_ops[cpu]->cpu_postboot)
231 		cpu_ops[cpu]->cpu_postboot();
232 
233 	/*
234 	 * Log the CPU info before it is marked online and might get read.
235 	 */
236 	cpuinfo_store_cpu();
237 
238 	/*
239 	 * Enable GIC and timers.
240 	 */
241 	notify_cpu_starting(cpu);
242 
243 	store_cpu_topology(cpu);
244 	numa_add_cpu(cpu);
245 
246 	/*
247 	 * OK, now it's safe to let the boot CPU continue.  Wait for
248 	 * the CPU migration code to notice that the CPU is online
249 	 * before we continue.
250 	 */
251 	pr_info("CPU%u: Booted secondary processor 0x%010lx [0x%08x]\n",
252 					 cpu, (unsigned long)mpidr,
253 					 read_cpuid_id());
254 	update_cpu_boot_status(CPU_BOOT_SUCCESS);
255 	set_cpu_online(cpu, true);
256 	complete(&cpu_running);
257 
258 	local_daif_restore(DAIF_PROCCTX);
259 
260 	/*
261 	 * OK, it's off to the idle thread for us
262 	 */
263 	cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
264 }
265 
266 #ifdef CONFIG_HOTPLUG_CPU
267 static int op_cpu_disable(unsigned int cpu)
268 {
269 	/*
270 	 * If we don't have a cpu_die method, abort before we reach the point
271 	 * of no return. CPU0 may not have an cpu_ops, so test for it.
272 	 */
273 	if (!cpu_ops[cpu] || !cpu_ops[cpu]->cpu_die)
274 		return -EOPNOTSUPP;
275 
276 	/*
277 	 * We may need to abort a hot unplug for some other mechanism-specific
278 	 * reason.
279 	 */
280 	if (cpu_ops[cpu]->cpu_disable)
281 		return cpu_ops[cpu]->cpu_disable(cpu);
282 
283 	return 0;
284 }
285 
286 /*
287  * __cpu_disable runs on the processor to be shutdown.
288  */
289 int __cpu_disable(void)
290 {
291 	unsigned int cpu = smp_processor_id();
292 	int ret;
293 
294 	ret = op_cpu_disable(cpu);
295 	if (ret)
296 		return ret;
297 
298 	remove_cpu_topology(cpu);
299 	numa_remove_cpu(cpu);
300 
301 	/*
302 	 * Take this CPU offline.  Once we clear this, we can't return,
303 	 * and we must not schedule until we're ready to give up the cpu.
304 	 */
305 	set_cpu_online(cpu, false);
306 
307 	/*
308 	 * OK - migrate IRQs away from this CPU
309 	 */
310 	irq_migrate_all_off_this_cpu();
311 
312 	return 0;
313 }
314 
315 static int op_cpu_kill(unsigned int cpu)
316 {
317 	/*
318 	 * If we have no means of synchronising with the dying CPU, then assume
319 	 * that it is really dead. We can only wait for an arbitrary length of
320 	 * time and hope that it's dead, so let's skip the wait and just hope.
321 	 */
322 	if (!cpu_ops[cpu]->cpu_kill)
323 		return 0;
324 
325 	return cpu_ops[cpu]->cpu_kill(cpu);
326 }
327 
328 /*
329  * called on the thread which is asking for a CPU to be shutdown -
330  * waits until shutdown has completed, or it is timed out.
331  */
332 void __cpu_die(unsigned int cpu)
333 {
334 	int err;
335 
336 	if (!cpu_wait_death(cpu, 5)) {
337 		pr_crit("CPU%u: cpu didn't die\n", cpu);
338 		return;
339 	}
340 	pr_notice("CPU%u: shutdown\n", cpu);
341 
342 	/*
343 	 * Now that the dying CPU is beyond the point of no return w.r.t.
344 	 * in-kernel synchronisation, try to get the firwmare to help us to
345 	 * verify that it has really left the kernel before we consider
346 	 * clobbering anything it might still be using.
347 	 */
348 	err = op_cpu_kill(cpu);
349 	if (err)
350 		pr_warn("CPU%d may not have shut down cleanly: %d\n", cpu, err);
351 }
352 
353 /*
354  * Called from the idle thread for the CPU which has been shutdown.
355  *
356  */
357 void cpu_die(void)
358 {
359 	unsigned int cpu = smp_processor_id();
360 
361 	idle_task_exit();
362 
363 	local_daif_mask();
364 
365 	/* Tell __cpu_die() that this CPU is now safe to dispose of */
366 	(void)cpu_report_death();
367 
368 	/*
369 	 * Actually shutdown the CPU. This must never fail. The specific hotplug
370 	 * mechanism must perform all required cache maintenance to ensure that
371 	 * no dirty lines are lost in the process of shutting down the CPU.
372 	 */
373 	cpu_ops[cpu]->cpu_die(cpu);
374 
375 	BUG();
376 }
377 #endif
378 
379 /*
380  * Kill the calling secondary CPU, early in bringup before it is turned
381  * online.
382  */
383 void cpu_die_early(void)
384 {
385 	int cpu = smp_processor_id();
386 
387 	pr_crit("CPU%d: will not boot\n", cpu);
388 
389 	/* Mark this CPU absent */
390 	set_cpu_present(cpu, 0);
391 
392 #ifdef CONFIG_HOTPLUG_CPU
393 	update_cpu_boot_status(CPU_KILL_ME);
394 	/* Check if we can park ourselves */
395 	if (cpu_ops[cpu] && cpu_ops[cpu]->cpu_die)
396 		cpu_ops[cpu]->cpu_die(cpu);
397 #endif
398 	update_cpu_boot_status(CPU_STUCK_IN_KERNEL);
399 
400 	cpu_park_loop();
401 }
402 
403 static void __init hyp_mode_check(void)
404 {
405 	if (is_hyp_mode_available())
406 		pr_info("CPU: All CPU(s) started at EL2\n");
407 	else if (is_hyp_mode_mismatched())
408 		WARN_TAINT(1, TAINT_CPU_OUT_OF_SPEC,
409 			   "CPU: CPUs started in inconsistent modes");
410 	else
411 		pr_info("CPU: All CPU(s) started at EL1\n");
412 	if (IS_ENABLED(CONFIG_KVM_ARM_HOST))
413 		kvm_compute_layout();
414 }
415 
416 void __init smp_cpus_done(unsigned int max_cpus)
417 {
418 	pr_info("SMP: Total of %d processors activated.\n", num_online_cpus());
419 	setup_cpu_features();
420 	hyp_mode_check();
421 	apply_alternatives_all();
422 	mark_linear_text_alias_ro();
423 }
424 
425 void __init smp_prepare_boot_cpu(void)
426 {
427 	set_my_cpu_offset(per_cpu_offset(smp_processor_id()));
428 	cpuinfo_store_boot_cpu();
429 
430 	/*
431 	 * We now know enough about the boot CPU to apply the
432 	 * alternatives that cannot wait until interrupt handling
433 	 * and/or scheduling is enabled.
434 	 */
435 	apply_boot_alternatives();
436 
437 	/* Conditionally switch to GIC PMR for interrupt masking */
438 	if (system_uses_irq_prio_masking())
439 		init_gic_priority_masking();
440 }
441 
442 static u64 __init of_get_cpu_mpidr(struct device_node *dn)
443 {
444 	const __be32 *cell;
445 	u64 hwid;
446 
447 	/*
448 	 * A cpu node with missing "reg" property is
449 	 * considered invalid to build a cpu_logical_map
450 	 * entry.
451 	 */
452 	cell = of_get_property(dn, "reg", NULL);
453 	if (!cell) {
454 		pr_err("%pOF: missing reg property\n", dn);
455 		return INVALID_HWID;
456 	}
457 
458 	hwid = of_read_number(cell, of_n_addr_cells(dn));
459 	/*
460 	 * Non affinity bits must be set to 0 in the DT
461 	 */
462 	if (hwid & ~MPIDR_HWID_BITMASK) {
463 		pr_err("%pOF: invalid reg property\n", dn);
464 		return INVALID_HWID;
465 	}
466 	return hwid;
467 }
468 
469 /*
470  * Duplicate MPIDRs are a recipe for disaster. Scan all initialized
471  * entries and check for duplicates. If any is found just ignore the
472  * cpu. cpu_logical_map was initialized to INVALID_HWID to avoid
473  * matching valid MPIDR values.
474  */
475 static bool __init is_mpidr_duplicate(unsigned int cpu, u64 hwid)
476 {
477 	unsigned int i;
478 
479 	for (i = 1; (i < cpu) && (i < NR_CPUS); i++)
480 		if (cpu_logical_map(i) == hwid)
481 			return true;
482 	return false;
483 }
484 
485 /*
486  * Initialize cpu operations for a logical cpu and
487  * set it in the possible mask on success
488  */
489 static int __init smp_cpu_setup(int cpu)
490 {
491 	if (cpu_read_ops(cpu))
492 		return -ENODEV;
493 
494 	if (cpu_ops[cpu]->cpu_init(cpu))
495 		return -ENODEV;
496 
497 	set_cpu_possible(cpu, true);
498 
499 	return 0;
500 }
501 
502 static bool bootcpu_valid __initdata;
503 static unsigned int cpu_count = 1;
504 
505 #ifdef CONFIG_ACPI
506 static struct acpi_madt_generic_interrupt cpu_madt_gicc[NR_CPUS];
507 
508 struct acpi_madt_generic_interrupt *acpi_cpu_get_madt_gicc(int cpu)
509 {
510 	return &cpu_madt_gicc[cpu];
511 }
512 
513 /*
514  * acpi_map_gic_cpu_interface - parse processor MADT entry
515  *
516  * Carry out sanity checks on MADT processor entry and initialize
517  * cpu_logical_map on success
518  */
519 static void __init
520 acpi_map_gic_cpu_interface(struct acpi_madt_generic_interrupt *processor)
521 {
522 	u64 hwid = processor->arm_mpidr;
523 
524 	if (!(processor->flags & ACPI_MADT_ENABLED)) {
525 		pr_debug("skipping disabled CPU entry with 0x%llx MPIDR\n", hwid);
526 		return;
527 	}
528 
529 	if (hwid & ~MPIDR_HWID_BITMASK || hwid == INVALID_HWID) {
530 		pr_err("skipping CPU entry with invalid MPIDR 0x%llx\n", hwid);
531 		return;
532 	}
533 
534 	if (is_mpidr_duplicate(cpu_count, hwid)) {
535 		pr_err("duplicate CPU MPIDR 0x%llx in MADT\n", hwid);
536 		return;
537 	}
538 
539 	/* Check if GICC structure of boot CPU is available in the MADT */
540 	if (cpu_logical_map(0) == hwid) {
541 		if (bootcpu_valid) {
542 			pr_err("duplicate boot CPU MPIDR: 0x%llx in MADT\n",
543 			       hwid);
544 			return;
545 		}
546 		bootcpu_valid = true;
547 		cpu_madt_gicc[0] = *processor;
548 		return;
549 	}
550 
551 	if (cpu_count >= NR_CPUS)
552 		return;
553 
554 	/* map the logical cpu id to cpu MPIDR */
555 	cpu_logical_map(cpu_count) = hwid;
556 
557 	cpu_madt_gicc[cpu_count] = *processor;
558 
559 	/*
560 	 * Set-up the ACPI parking protocol cpu entries
561 	 * while initializing the cpu_logical_map to
562 	 * avoid parsing MADT entries multiple times for
563 	 * nothing (ie a valid cpu_logical_map entry should
564 	 * contain a valid parking protocol data set to
565 	 * initialize the cpu if the parking protocol is
566 	 * the only available enable method).
567 	 */
568 	acpi_set_mailbox_entry(cpu_count, processor);
569 
570 	cpu_count++;
571 }
572 
573 static int __init
574 acpi_parse_gic_cpu_interface(union acpi_subtable_headers *header,
575 			     const unsigned long end)
576 {
577 	struct acpi_madt_generic_interrupt *processor;
578 
579 	processor = (struct acpi_madt_generic_interrupt *)header;
580 	if (BAD_MADT_GICC_ENTRY(processor, end))
581 		return -EINVAL;
582 
583 	acpi_table_print_madt_entry(&header->common);
584 
585 	acpi_map_gic_cpu_interface(processor);
586 
587 	return 0;
588 }
589 
590 static void __init acpi_parse_and_init_cpus(void)
591 {
592 	int i;
593 
594 	/*
595 	 * do a walk of MADT to determine how many CPUs
596 	 * we have including disabled CPUs, and get information
597 	 * we need for SMP init.
598 	 */
599 	acpi_table_parse_madt(ACPI_MADT_TYPE_GENERIC_INTERRUPT,
600 				      acpi_parse_gic_cpu_interface, 0);
601 
602 	/*
603 	 * In ACPI, SMP and CPU NUMA information is provided in separate
604 	 * static tables, namely the MADT and the SRAT.
605 	 *
606 	 * Thus, it is simpler to first create the cpu logical map through
607 	 * an MADT walk and then map the logical cpus to their node ids
608 	 * as separate steps.
609 	 */
610 	acpi_map_cpus_to_nodes();
611 
612 	for (i = 0; i < nr_cpu_ids; i++)
613 		early_map_cpu_to_node(i, acpi_numa_get_nid(i));
614 }
615 #else
616 #define acpi_parse_and_init_cpus(...)	do { } while (0)
617 #endif
618 
619 /*
620  * Enumerate the possible CPU set from the device tree and build the
621  * cpu logical map array containing MPIDR values related to logical
622  * cpus. Assumes that cpu_logical_map(0) has already been initialized.
623  */
624 static void __init of_parse_and_init_cpus(void)
625 {
626 	struct device_node *dn;
627 
628 	for_each_of_cpu_node(dn) {
629 		u64 hwid = of_get_cpu_mpidr(dn);
630 
631 		if (hwid == INVALID_HWID)
632 			goto next;
633 
634 		if (is_mpidr_duplicate(cpu_count, hwid)) {
635 			pr_err("%pOF: duplicate cpu reg properties in the DT\n",
636 				dn);
637 			goto next;
638 		}
639 
640 		/*
641 		 * The numbering scheme requires that the boot CPU
642 		 * must be assigned logical id 0. Record it so that
643 		 * the logical map built from DT is validated and can
644 		 * be used.
645 		 */
646 		if (hwid == cpu_logical_map(0)) {
647 			if (bootcpu_valid) {
648 				pr_err("%pOF: duplicate boot cpu reg property in DT\n",
649 					dn);
650 				goto next;
651 			}
652 
653 			bootcpu_valid = true;
654 			early_map_cpu_to_node(0, of_node_to_nid(dn));
655 
656 			/*
657 			 * cpu_logical_map has already been
658 			 * initialized and the boot cpu doesn't need
659 			 * the enable-method so continue without
660 			 * incrementing cpu.
661 			 */
662 			continue;
663 		}
664 
665 		if (cpu_count >= NR_CPUS)
666 			goto next;
667 
668 		pr_debug("cpu logical map 0x%llx\n", hwid);
669 		cpu_logical_map(cpu_count) = hwid;
670 
671 		early_map_cpu_to_node(cpu_count, of_node_to_nid(dn));
672 next:
673 		cpu_count++;
674 	}
675 }
676 
677 /*
678  * Enumerate the possible CPU set from the device tree or ACPI and build the
679  * cpu logical map array containing MPIDR values related to logical
680  * cpus. Assumes that cpu_logical_map(0) has already been initialized.
681  */
682 void __init smp_init_cpus(void)
683 {
684 	int i;
685 
686 	if (acpi_disabled)
687 		of_parse_and_init_cpus();
688 	else
689 		acpi_parse_and_init_cpus();
690 
691 	if (cpu_count > nr_cpu_ids)
692 		pr_warn("Number of cores (%d) exceeds configured maximum of %u - clipping\n",
693 			cpu_count, nr_cpu_ids);
694 
695 	if (!bootcpu_valid) {
696 		pr_err("missing boot CPU MPIDR, not enabling secondaries\n");
697 		return;
698 	}
699 
700 	/*
701 	 * We need to set the cpu_logical_map entries before enabling
702 	 * the cpus so that cpu processor description entries (DT cpu nodes
703 	 * and ACPI MADT entries) can be retrieved by matching the cpu hwid
704 	 * with entries in cpu_logical_map while initializing the cpus.
705 	 * If the cpu set-up fails, invalidate the cpu_logical_map entry.
706 	 */
707 	for (i = 1; i < nr_cpu_ids; i++) {
708 		if (cpu_logical_map(i) != INVALID_HWID) {
709 			if (smp_cpu_setup(i))
710 				cpu_logical_map(i) = INVALID_HWID;
711 		}
712 	}
713 }
714 
715 void __init smp_prepare_cpus(unsigned int max_cpus)
716 {
717 	int err;
718 	unsigned int cpu;
719 	unsigned int this_cpu;
720 
721 	init_cpu_topology();
722 
723 	this_cpu = smp_processor_id();
724 	store_cpu_topology(this_cpu);
725 	numa_store_cpu_info(this_cpu);
726 	numa_add_cpu(this_cpu);
727 
728 	/*
729 	 * If UP is mandated by "nosmp" (which implies "maxcpus=0"), don't set
730 	 * secondary CPUs present.
731 	 */
732 	if (max_cpus == 0)
733 		return;
734 
735 	/*
736 	 * Initialise the present map (which describes the set of CPUs
737 	 * actually populated at the present time) and release the
738 	 * secondaries from the bootloader.
739 	 */
740 	for_each_possible_cpu(cpu) {
741 
742 		per_cpu(cpu_number, cpu) = cpu;
743 
744 		if (cpu == smp_processor_id())
745 			continue;
746 
747 		if (!cpu_ops[cpu])
748 			continue;
749 
750 		err = cpu_ops[cpu]->cpu_prepare(cpu);
751 		if (err)
752 			continue;
753 
754 		set_cpu_present(cpu, true);
755 		numa_store_cpu_info(cpu);
756 	}
757 }
758 
759 void (*__smp_cross_call)(const struct cpumask *, unsigned int);
760 
761 void __init set_smp_cross_call(void (*fn)(const struct cpumask *, unsigned int))
762 {
763 	__smp_cross_call = fn;
764 }
765 
766 static const char *ipi_types[NR_IPI] __tracepoint_string = {
767 #define S(x,s)	[x] = s
768 	S(IPI_RESCHEDULE, "Rescheduling interrupts"),
769 	S(IPI_CALL_FUNC, "Function call interrupts"),
770 	S(IPI_CPU_STOP, "CPU stop interrupts"),
771 	S(IPI_CPU_CRASH_STOP, "CPU stop (for crash dump) interrupts"),
772 	S(IPI_TIMER, "Timer broadcast interrupts"),
773 	S(IPI_IRQ_WORK, "IRQ work interrupts"),
774 	S(IPI_WAKEUP, "CPU wake-up interrupts"),
775 };
776 
777 static void smp_cross_call(const struct cpumask *target, unsigned int ipinr)
778 {
779 	trace_ipi_raise(target, ipi_types[ipinr]);
780 	__smp_cross_call(target, ipinr);
781 }
782 
783 void show_ipi_list(struct seq_file *p, int prec)
784 {
785 	unsigned int cpu, i;
786 
787 	for (i = 0; i < NR_IPI; i++) {
788 		seq_printf(p, "%*s%u:%s", prec - 1, "IPI", i,
789 			   prec >= 4 ? " " : "");
790 		for_each_online_cpu(cpu)
791 			seq_printf(p, "%10u ",
792 				   __get_irq_stat(cpu, ipi_irqs[i]));
793 		seq_printf(p, "      %s\n", ipi_types[i]);
794 	}
795 }
796 
797 u64 smp_irq_stat_cpu(unsigned int cpu)
798 {
799 	u64 sum = 0;
800 	int i;
801 
802 	for (i = 0; i < NR_IPI; i++)
803 		sum += __get_irq_stat(cpu, ipi_irqs[i]);
804 
805 	return sum;
806 }
807 
808 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
809 {
810 	smp_cross_call(mask, IPI_CALL_FUNC);
811 }
812 
813 void arch_send_call_function_single_ipi(int cpu)
814 {
815 	smp_cross_call(cpumask_of(cpu), IPI_CALL_FUNC);
816 }
817 
818 #ifdef CONFIG_ARM64_ACPI_PARKING_PROTOCOL
819 void arch_send_wakeup_ipi_mask(const struct cpumask *mask)
820 {
821 	smp_cross_call(mask, IPI_WAKEUP);
822 }
823 #endif
824 
825 #ifdef CONFIG_IRQ_WORK
826 void arch_irq_work_raise(void)
827 {
828 	if (__smp_cross_call)
829 		smp_cross_call(cpumask_of(smp_processor_id()), IPI_IRQ_WORK);
830 }
831 #endif
832 
833 static void local_cpu_stop(void)
834 {
835 	set_cpu_online(smp_processor_id(), false);
836 
837 	local_daif_mask();
838 	sdei_mask_local_cpu();
839 	cpu_park_loop();
840 }
841 
842 /*
843  * We need to implement panic_smp_self_stop() for parallel panic() calls, so
844  * that cpu_online_mask gets correctly updated and smp_send_stop() can skip
845  * CPUs that have already stopped themselves.
846  */
847 void panic_smp_self_stop(void)
848 {
849 	local_cpu_stop();
850 }
851 
852 #ifdef CONFIG_KEXEC_CORE
853 static atomic_t waiting_for_crash_ipi = ATOMIC_INIT(0);
854 #endif
855 
856 static void ipi_cpu_crash_stop(unsigned int cpu, struct pt_regs *regs)
857 {
858 #ifdef CONFIG_KEXEC_CORE
859 	crash_save_cpu(regs, cpu);
860 
861 	atomic_dec(&waiting_for_crash_ipi);
862 
863 	local_irq_disable();
864 	sdei_mask_local_cpu();
865 
866 #ifdef CONFIG_HOTPLUG_CPU
867 	if (cpu_ops[cpu]->cpu_die)
868 		cpu_ops[cpu]->cpu_die(cpu);
869 #endif
870 
871 	/* just in case */
872 	cpu_park_loop();
873 #endif
874 }
875 
876 /*
877  * Main handler for inter-processor interrupts
878  */
879 void handle_IPI(int ipinr, struct pt_regs *regs)
880 {
881 	unsigned int cpu = smp_processor_id();
882 	struct pt_regs *old_regs = set_irq_regs(regs);
883 
884 	if ((unsigned)ipinr < NR_IPI) {
885 		trace_ipi_entry_rcuidle(ipi_types[ipinr]);
886 		__inc_irq_stat(cpu, ipi_irqs[ipinr]);
887 	}
888 
889 	switch (ipinr) {
890 	case IPI_RESCHEDULE:
891 		scheduler_ipi();
892 		break;
893 
894 	case IPI_CALL_FUNC:
895 		irq_enter();
896 		generic_smp_call_function_interrupt();
897 		irq_exit();
898 		break;
899 
900 	case IPI_CPU_STOP:
901 		irq_enter();
902 		local_cpu_stop();
903 		irq_exit();
904 		break;
905 
906 	case IPI_CPU_CRASH_STOP:
907 		if (IS_ENABLED(CONFIG_KEXEC_CORE)) {
908 			irq_enter();
909 			ipi_cpu_crash_stop(cpu, regs);
910 
911 			unreachable();
912 		}
913 		break;
914 
915 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
916 	case IPI_TIMER:
917 		irq_enter();
918 		tick_receive_broadcast();
919 		irq_exit();
920 		break;
921 #endif
922 
923 #ifdef CONFIG_IRQ_WORK
924 	case IPI_IRQ_WORK:
925 		irq_enter();
926 		irq_work_run();
927 		irq_exit();
928 		break;
929 #endif
930 
931 #ifdef CONFIG_ARM64_ACPI_PARKING_PROTOCOL
932 	case IPI_WAKEUP:
933 		WARN_ONCE(!acpi_parking_protocol_valid(cpu),
934 			  "CPU%u: Wake-up IPI outside the ACPI parking protocol\n",
935 			  cpu);
936 		break;
937 #endif
938 
939 	default:
940 		pr_crit("CPU%u: Unknown IPI message 0x%x\n", cpu, ipinr);
941 		break;
942 	}
943 
944 	if ((unsigned)ipinr < NR_IPI)
945 		trace_ipi_exit_rcuidle(ipi_types[ipinr]);
946 	set_irq_regs(old_regs);
947 }
948 
949 void smp_send_reschedule(int cpu)
950 {
951 	smp_cross_call(cpumask_of(cpu), IPI_RESCHEDULE);
952 }
953 
954 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
955 void tick_broadcast(const struct cpumask *mask)
956 {
957 	smp_cross_call(mask, IPI_TIMER);
958 }
959 #endif
960 
961 void smp_send_stop(void)
962 {
963 	unsigned long timeout;
964 
965 	if (num_online_cpus() > 1) {
966 		cpumask_t mask;
967 
968 		cpumask_copy(&mask, cpu_online_mask);
969 		cpumask_clear_cpu(smp_processor_id(), &mask);
970 
971 		if (system_state <= SYSTEM_RUNNING)
972 			pr_crit("SMP: stopping secondary CPUs\n");
973 		smp_cross_call(&mask, IPI_CPU_STOP);
974 	}
975 
976 	/* Wait up to one second for other CPUs to stop */
977 	timeout = USEC_PER_SEC;
978 	while (num_online_cpus() > 1 && timeout--)
979 		udelay(1);
980 
981 	if (num_online_cpus() > 1)
982 		pr_warn("SMP: failed to stop secondary CPUs %*pbl\n",
983 			cpumask_pr_args(cpu_online_mask));
984 
985 	sdei_mask_local_cpu();
986 }
987 
988 #ifdef CONFIG_KEXEC_CORE
989 void crash_smp_send_stop(void)
990 {
991 	static int cpus_stopped;
992 	cpumask_t mask;
993 	unsigned long timeout;
994 
995 	/*
996 	 * This function can be called twice in panic path, but obviously
997 	 * we execute this only once.
998 	 */
999 	if (cpus_stopped)
1000 		return;
1001 
1002 	cpus_stopped = 1;
1003 
1004 	if (num_online_cpus() == 1) {
1005 		sdei_mask_local_cpu();
1006 		return;
1007 	}
1008 
1009 	cpumask_copy(&mask, cpu_online_mask);
1010 	cpumask_clear_cpu(smp_processor_id(), &mask);
1011 
1012 	atomic_set(&waiting_for_crash_ipi, num_online_cpus() - 1);
1013 
1014 	pr_crit("SMP: stopping secondary CPUs\n");
1015 	smp_cross_call(&mask, IPI_CPU_CRASH_STOP);
1016 
1017 	/* Wait up to one second for other CPUs to stop */
1018 	timeout = USEC_PER_SEC;
1019 	while ((atomic_read(&waiting_for_crash_ipi) > 0) && timeout--)
1020 		udelay(1);
1021 
1022 	if (atomic_read(&waiting_for_crash_ipi) > 0)
1023 		pr_warn("SMP: failed to stop secondary CPUs %*pbl\n",
1024 			cpumask_pr_args(&mask));
1025 
1026 	sdei_mask_local_cpu();
1027 }
1028 
1029 bool smp_crash_stop_failed(void)
1030 {
1031 	return (atomic_read(&waiting_for_crash_ipi) > 0);
1032 }
1033 #endif
1034 
1035 /*
1036  * not supported here
1037  */
1038 int setup_profiling_timer(unsigned int multiplier)
1039 {
1040 	return -EINVAL;
1041 }
1042 
1043 static bool have_cpu_die(void)
1044 {
1045 #ifdef CONFIG_HOTPLUG_CPU
1046 	int any_cpu = raw_smp_processor_id();
1047 
1048 	if (cpu_ops[any_cpu] && cpu_ops[any_cpu]->cpu_die)
1049 		return true;
1050 #endif
1051 	return false;
1052 }
1053 
1054 bool cpus_are_stuck_in_kernel(void)
1055 {
1056 	bool smp_spin_tables = (num_possible_cpus() > 1 && !have_cpu_die());
1057 
1058 	return !!cpus_stuck_in_kernel || smp_spin_tables;
1059 }
1060