xref: /openbmc/linux/arch/arm64/kernel/smp.c (revision 1a18374f)
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 
35 #include <asm/alternative.h>
36 #include <asm/atomic.h>
37 #include <asm/cacheflush.h>
38 #include <asm/cpu.h>
39 #include <asm/cputype.h>
40 #include <asm/cpu_ops.h>
41 #include <asm/daifflags.h>
42 #include <asm/mmu_context.h>
43 #include <asm/numa.h>
44 #include <asm/pgtable.h>
45 #include <asm/pgalloc.h>
46 #include <asm/processor.h>
47 #include <asm/smp_plat.h>
48 #include <asm/sections.h>
49 #include <asm/tlbflush.h>
50 #include <asm/ptrace.h>
51 #include <asm/virt.h>
52 
53 #define CREATE_TRACE_POINTS
54 #include <trace/events/ipi.h>
55 
56 DEFINE_PER_CPU_READ_MOSTLY(int, cpu_number);
57 EXPORT_PER_CPU_SYMBOL(cpu_number);
58 
59 /*
60  * as from 2.5, kernels no longer have an init_tasks structure
61  * so we need some other way of telling a new secondary core
62  * where to place its SVC stack
63  */
64 struct secondary_data secondary_data;
65 /* Number of CPUs which aren't online, but looping in kernel text. */
66 int cpus_stuck_in_kernel;
67 
68 enum ipi_msg_type {
69 	IPI_RESCHEDULE,
70 	IPI_CALL_FUNC,
71 	IPI_CPU_STOP,
72 	IPI_CPU_CRASH_STOP,
73 	IPI_TIMER,
74 	IPI_IRQ_WORK,
75 	IPI_WAKEUP
76 };
77 
78 #ifdef CONFIG_HOTPLUG_CPU
79 static int op_cpu_kill(unsigned int cpu);
80 #else
81 static inline int op_cpu_kill(unsigned int cpu)
82 {
83 	return -ENOSYS;
84 }
85 #endif
86 
87 
88 /*
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.
91  */
92 static int boot_secondary(unsigned int cpu, struct task_struct *idle)
93 {
94 	if (cpu_ops[cpu]->cpu_boot)
95 		return cpu_ops[cpu]->cpu_boot(cpu);
96 
97 	return -EOPNOTSUPP;
98 }
99 
100 static DECLARE_COMPLETION(cpu_running);
101 
102 int __cpu_up(unsigned int cpu, struct task_struct *idle)
103 {
104 	int ret;
105 	long status;
106 
107 	/*
108 	 * We need to tell the secondary core where to find its stack and the
109 	 * page tables.
110 	 */
111 	secondary_data.task = idle;
112 	secondary_data.stack = task_stack_page(idle) + THREAD_SIZE;
113 	update_cpu_boot_status(CPU_MMU_OFF);
114 	__flush_dcache_area(&secondary_data, sizeof(secondary_data));
115 
116 	/*
117 	 * Now bring the CPU into our world.
118 	 */
119 	ret = boot_secondary(cpu, idle);
120 	if (ret == 0) {
121 		/*
122 		 * CPU was successfully started, wait for it to come online or
123 		 * time out.
124 		 */
125 		wait_for_completion_timeout(&cpu_running,
126 					    msecs_to_jiffies(5000));
127 
128 		if (!cpu_online(cpu)) {
129 			pr_crit("CPU%u: failed to come online\n", cpu);
130 			ret = -EIO;
131 		}
132 	} else {
133 		pr_err("CPU%u: failed to boot: %d\n", cpu, ret);
134 		return ret;
135 	}
136 
137 	secondary_data.task = NULL;
138 	secondary_data.stack = NULL;
139 	__flush_dcache_area(&secondary_data, sizeof(secondary_data));
140 	status = READ_ONCE(secondary_data.status);
141 	if (ret && status) {
142 
143 		if (status == CPU_MMU_OFF)
144 			status = READ_ONCE(__early_cpu_boot_status);
145 
146 		switch (status & CPU_BOOT_STATUS_MASK) {
147 		default:
148 			pr_err("CPU%u: failed in unknown state : 0x%lx\n",
149 					cpu, status);
150 			cpus_stuck_in_kernel++;
151 			break;
152 		case CPU_KILL_ME:
153 			if (!op_cpu_kill(cpu)) {
154 				pr_crit("CPU%u: died during early boot\n", cpu);
155 				break;
156 			}
157 			pr_crit("CPU%u: may not have shut down cleanly\n", cpu);
158 			/* Fall through */
159 		case CPU_STUCK_IN_KERNEL:
160 			pr_crit("CPU%u: is stuck in kernel\n", cpu);
161 			if (status & CPU_STUCK_REASON_52_BIT_VA)
162 				pr_crit("CPU%u: does not support 52-bit VAs\n", cpu);
163 			if (status & CPU_STUCK_REASON_NO_GRAN)
164 				pr_crit("CPU%u: does not support %luK granule \n", cpu, PAGE_SIZE / SZ_1K);
165 			cpus_stuck_in_kernel++;
166 			break;
167 		case CPU_PANIC_KERNEL:
168 			panic("CPU%u detected unsupported configuration\n", cpu);
169 		}
170 	}
171 
172 	return ret;
173 }
174 
175 static void init_gic_priority_masking(void)
176 {
177 	u32 cpuflags;
178 
179 	if (WARN_ON(!gic_enable_sre()))
180 		return;
181 
182 	cpuflags = read_sysreg(daif);
183 
184 	WARN_ON(!(cpuflags & PSR_I_BIT));
185 
186 	gic_write_pmr(GIC_PRIO_IRQON | GIC_PRIO_PSR_I_SET);
187 }
188 
189 /*
190  * This is the secondary CPU boot entry.  We're using this CPUs
191  * idle thread stack, but a set of temporary page tables.
192  */
193 asmlinkage notrace void secondary_start_kernel(void)
194 {
195 	u64 mpidr = read_cpuid_mpidr() & MPIDR_HWID_BITMASK;
196 	struct mm_struct *mm = &init_mm;
197 	unsigned int cpu;
198 
199 	cpu = task_cpu(current);
200 	set_my_cpu_offset(per_cpu_offset(cpu));
201 
202 	/*
203 	 * All kernel threads share the same mm context; grab a
204 	 * reference and switch to it.
205 	 */
206 	mmgrab(mm);
207 	current->active_mm = mm;
208 
209 	/*
210 	 * TTBR0 is only used for the identity mapping at this stage. Make it
211 	 * point to zero page to avoid speculatively fetching new entries.
212 	 */
213 	cpu_uninstall_idmap();
214 
215 	if (system_uses_irq_prio_masking())
216 		init_gic_priority_masking();
217 
218 	preempt_disable();
219 	trace_hardirqs_off();
220 
221 	/*
222 	 * If the system has established the capabilities, make sure
223 	 * this CPU ticks all of those. If it doesn't, the CPU will
224 	 * fail to come online.
225 	 */
226 	check_local_cpu_capabilities();
227 
228 	if (cpu_ops[cpu]->cpu_postboot)
229 		cpu_ops[cpu]->cpu_postboot();
230 
231 	/*
232 	 * Log the CPU info before it is marked online and might get read.
233 	 */
234 	cpuinfo_store_cpu();
235 
236 	/*
237 	 * Enable GIC and timers.
238 	 */
239 	notify_cpu_starting(cpu);
240 
241 	store_cpu_topology(cpu);
242 	numa_add_cpu(cpu);
243 
244 	/*
245 	 * OK, now it's safe to let the boot CPU continue.  Wait for
246 	 * the CPU migration code to notice that the CPU is online
247 	 * before we continue.
248 	 */
249 	pr_info("CPU%u: Booted secondary processor 0x%010lx [0x%08x]\n",
250 					 cpu, (unsigned long)mpidr,
251 					 read_cpuid_id());
252 	update_cpu_boot_status(CPU_BOOT_SUCCESS);
253 	set_cpu_online(cpu, true);
254 	complete(&cpu_running);
255 
256 	local_daif_restore(DAIF_PROCCTX);
257 
258 	/*
259 	 * OK, it's off to the idle thread for us
260 	 */
261 	cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
262 }
263 
264 #ifdef CONFIG_HOTPLUG_CPU
265 static int op_cpu_disable(unsigned int cpu)
266 {
267 	/*
268 	 * If we don't have a cpu_die method, abort before we reach the point
269 	 * of no return. CPU0 may not have an cpu_ops, so test for it.
270 	 */
271 	if (!cpu_ops[cpu] || !cpu_ops[cpu]->cpu_die)
272 		return -EOPNOTSUPP;
273 
274 	/*
275 	 * We may need to abort a hot unplug for some other mechanism-specific
276 	 * reason.
277 	 */
278 	if (cpu_ops[cpu]->cpu_disable)
279 		return cpu_ops[cpu]->cpu_disable(cpu);
280 
281 	return 0;
282 }
283 
284 /*
285  * __cpu_disable runs on the processor to be shutdown.
286  */
287 int __cpu_disable(void)
288 {
289 	unsigned int cpu = smp_processor_id();
290 	int ret;
291 
292 	ret = op_cpu_disable(cpu);
293 	if (ret)
294 		return ret;
295 
296 	remove_cpu_topology(cpu);
297 	numa_remove_cpu(cpu);
298 
299 	/*
300 	 * Take this CPU offline.  Once we clear this, we can't return,
301 	 * and we must not schedule until we're ready to give up the cpu.
302 	 */
303 	set_cpu_online(cpu, false);
304 
305 	/*
306 	 * OK - migrate IRQs away from this CPU
307 	 */
308 	irq_migrate_all_off_this_cpu();
309 
310 	return 0;
311 }
312 
313 static int op_cpu_kill(unsigned int cpu)
314 {
315 	/*
316 	 * If we have no means of synchronising with the dying CPU, then assume
317 	 * that it is really dead. We can only wait for an arbitrary length of
318 	 * time and hope that it's dead, so let's skip the wait and just hope.
319 	 */
320 	if (!cpu_ops[cpu]->cpu_kill)
321 		return 0;
322 
323 	return cpu_ops[cpu]->cpu_kill(cpu);
324 }
325 
326 /*
327  * called on the thread which is asking for a CPU to be shutdown -
328  * waits until shutdown has completed, or it is timed out.
329  */
330 void __cpu_die(unsigned int cpu)
331 {
332 	int err;
333 
334 	if (!cpu_wait_death(cpu, 5)) {
335 		pr_crit("CPU%u: cpu didn't die\n", cpu);
336 		return;
337 	}
338 	pr_notice("CPU%u: shutdown\n", cpu);
339 
340 	/*
341 	 * Now that the dying CPU is beyond the point of no return w.r.t.
342 	 * in-kernel synchronisation, try to get the firwmare to help us to
343 	 * verify that it has really left the kernel before we consider
344 	 * clobbering anything it might still be using.
345 	 */
346 	err = op_cpu_kill(cpu);
347 	if (err)
348 		pr_warn("CPU%d may not have shut down cleanly: %d\n", cpu, err);
349 }
350 
351 /*
352  * Called from the idle thread for the CPU which has been shutdown.
353  *
354  */
355 void cpu_die(void)
356 {
357 	unsigned int cpu = smp_processor_id();
358 
359 	idle_task_exit();
360 
361 	local_daif_mask();
362 
363 	/* Tell __cpu_die() that this CPU is now safe to dispose of */
364 	(void)cpu_report_death();
365 
366 	/*
367 	 * Actually shutdown the CPU. This must never fail. The specific hotplug
368 	 * mechanism must perform all required cache maintenance to ensure that
369 	 * no dirty lines are lost in the process of shutting down the CPU.
370 	 */
371 	cpu_ops[cpu]->cpu_die(cpu);
372 
373 	BUG();
374 }
375 #endif
376 
377 /*
378  * Kill the calling secondary CPU, early in bringup before it is turned
379  * online.
380  */
381 void cpu_die_early(void)
382 {
383 	int cpu = smp_processor_id();
384 
385 	pr_crit("CPU%d: will not boot\n", cpu);
386 
387 	/* Mark this CPU absent */
388 	set_cpu_present(cpu, 0);
389 
390 #ifdef CONFIG_HOTPLUG_CPU
391 	update_cpu_boot_status(CPU_KILL_ME);
392 	/* Check if we can park ourselves */
393 	if (cpu_ops[cpu] && cpu_ops[cpu]->cpu_die)
394 		cpu_ops[cpu]->cpu_die(cpu);
395 #endif
396 	update_cpu_boot_status(CPU_STUCK_IN_KERNEL);
397 
398 	cpu_park_loop();
399 }
400 
401 static void __init hyp_mode_check(void)
402 {
403 	if (is_hyp_mode_available())
404 		pr_info("CPU: All CPU(s) started at EL2\n");
405 	else if (is_hyp_mode_mismatched())
406 		WARN_TAINT(1, TAINT_CPU_OUT_OF_SPEC,
407 			   "CPU: CPUs started in inconsistent modes");
408 	else
409 		pr_info("CPU: All CPU(s) started at EL1\n");
410 }
411 
412 void __init smp_cpus_done(unsigned int max_cpus)
413 {
414 	pr_info("SMP: Total of %d processors activated.\n", num_online_cpus());
415 	setup_cpu_features();
416 	hyp_mode_check();
417 	apply_alternatives_all();
418 	mark_linear_text_alias_ro();
419 }
420 
421 void __init smp_prepare_boot_cpu(void)
422 {
423 	set_my_cpu_offset(per_cpu_offset(smp_processor_id()));
424 	cpuinfo_store_boot_cpu();
425 
426 	/*
427 	 * We now know enough about the boot CPU to apply the
428 	 * alternatives that cannot wait until interrupt handling
429 	 * and/or scheduling is enabled.
430 	 */
431 	apply_boot_alternatives();
432 
433 	/* Conditionally switch to GIC PMR for interrupt masking */
434 	if (system_uses_irq_prio_masking())
435 		init_gic_priority_masking();
436 }
437 
438 static u64 __init of_get_cpu_mpidr(struct device_node *dn)
439 {
440 	const __be32 *cell;
441 	u64 hwid;
442 
443 	/*
444 	 * A cpu node with missing "reg" property is
445 	 * considered invalid to build a cpu_logical_map
446 	 * entry.
447 	 */
448 	cell = of_get_property(dn, "reg", NULL);
449 	if (!cell) {
450 		pr_err("%pOF: missing reg property\n", dn);
451 		return INVALID_HWID;
452 	}
453 
454 	hwid = of_read_number(cell, of_n_addr_cells(dn));
455 	/*
456 	 * Non affinity bits must be set to 0 in the DT
457 	 */
458 	if (hwid & ~MPIDR_HWID_BITMASK) {
459 		pr_err("%pOF: invalid reg property\n", dn);
460 		return INVALID_HWID;
461 	}
462 	return hwid;
463 }
464 
465 /*
466  * Duplicate MPIDRs are a recipe for disaster. Scan all initialized
467  * entries and check for duplicates. If any is found just ignore the
468  * cpu. cpu_logical_map was initialized to INVALID_HWID to avoid
469  * matching valid MPIDR values.
470  */
471 static bool __init is_mpidr_duplicate(unsigned int cpu, u64 hwid)
472 {
473 	unsigned int i;
474 
475 	for (i = 1; (i < cpu) && (i < NR_CPUS); i++)
476 		if (cpu_logical_map(i) == hwid)
477 			return true;
478 	return false;
479 }
480 
481 /*
482  * Initialize cpu operations for a logical cpu and
483  * set it in the possible mask on success
484  */
485 static int __init smp_cpu_setup(int cpu)
486 {
487 	if (cpu_read_ops(cpu))
488 		return -ENODEV;
489 
490 	if (cpu_ops[cpu]->cpu_init(cpu))
491 		return -ENODEV;
492 
493 	set_cpu_possible(cpu, true);
494 
495 	return 0;
496 }
497 
498 static bool bootcpu_valid __initdata;
499 static unsigned int cpu_count = 1;
500 
501 #ifdef CONFIG_ACPI
502 static struct acpi_madt_generic_interrupt cpu_madt_gicc[NR_CPUS];
503 
504 struct acpi_madt_generic_interrupt *acpi_cpu_get_madt_gicc(int cpu)
505 {
506 	return &cpu_madt_gicc[cpu];
507 }
508 
509 /*
510  * acpi_map_gic_cpu_interface - parse processor MADT entry
511  *
512  * Carry out sanity checks on MADT processor entry and initialize
513  * cpu_logical_map on success
514  */
515 static void __init
516 acpi_map_gic_cpu_interface(struct acpi_madt_generic_interrupt *processor)
517 {
518 	u64 hwid = processor->arm_mpidr;
519 
520 	if (!(processor->flags & ACPI_MADT_ENABLED)) {
521 		pr_debug("skipping disabled CPU entry with 0x%llx MPIDR\n", hwid);
522 		return;
523 	}
524 
525 	if (hwid & ~MPIDR_HWID_BITMASK || hwid == INVALID_HWID) {
526 		pr_err("skipping CPU entry with invalid MPIDR 0x%llx\n", hwid);
527 		return;
528 	}
529 
530 	if (is_mpidr_duplicate(cpu_count, hwid)) {
531 		pr_err("duplicate CPU MPIDR 0x%llx in MADT\n", hwid);
532 		return;
533 	}
534 
535 	/* Check if GICC structure of boot CPU is available in the MADT */
536 	if (cpu_logical_map(0) == hwid) {
537 		if (bootcpu_valid) {
538 			pr_err("duplicate boot CPU MPIDR: 0x%llx in MADT\n",
539 			       hwid);
540 			return;
541 		}
542 		bootcpu_valid = true;
543 		cpu_madt_gicc[0] = *processor;
544 		return;
545 	}
546 
547 	if (cpu_count >= NR_CPUS)
548 		return;
549 
550 	/* map the logical cpu id to cpu MPIDR */
551 	cpu_logical_map(cpu_count) = hwid;
552 
553 	cpu_madt_gicc[cpu_count] = *processor;
554 
555 	/*
556 	 * Set-up the ACPI parking protocol cpu entries
557 	 * while initializing the cpu_logical_map to
558 	 * avoid parsing MADT entries multiple times for
559 	 * nothing (ie a valid cpu_logical_map entry should
560 	 * contain a valid parking protocol data set to
561 	 * initialize the cpu if the parking protocol is
562 	 * the only available enable method).
563 	 */
564 	acpi_set_mailbox_entry(cpu_count, processor);
565 
566 	cpu_count++;
567 }
568 
569 static int __init
570 acpi_parse_gic_cpu_interface(union acpi_subtable_headers *header,
571 			     const unsigned long end)
572 {
573 	struct acpi_madt_generic_interrupt *processor;
574 
575 	processor = (struct acpi_madt_generic_interrupt *)header;
576 	if (BAD_MADT_GICC_ENTRY(processor, end))
577 		return -EINVAL;
578 
579 	acpi_table_print_madt_entry(&header->common);
580 
581 	acpi_map_gic_cpu_interface(processor);
582 
583 	return 0;
584 }
585 
586 static void __init acpi_parse_and_init_cpus(void)
587 {
588 	int i;
589 
590 	/*
591 	 * do a walk of MADT to determine how many CPUs
592 	 * we have including disabled CPUs, and get information
593 	 * we need for SMP init.
594 	 */
595 	acpi_table_parse_madt(ACPI_MADT_TYPE_GENERIC_INTERRUPT,
596 				      acpi_parse_gic_cpu_interface, 0);
597 
598 	/*
599 	 * In ACPI, SMP and CPU NUMA information is provided in separate
600 	 * static tables, namely the MADT and the SRAT.
601 	 *
602 	 * Thus, it is simpler to first create the cpu logical map through
603 	 * an MADT walk and then map the logical cpus to their node ids
604 	 * as separate steps.
605 	 */
606 	acpi_map_cpus_to_nodes();
607 
608 	for (i = 0; i < nr_cpu_ids; i++)
609 		early_map_cpu_to_node(i, acpi_numa_get_nid(i));
610 }
611 #else
612 #define acpi_parse_and_init_cpus(...)	do { } while (0)
613 #endif
614 
615 /*
616  * Enumerate the possible CPU set from the device tree and build the
617  * cpu logical map array containing MPIDR values related to logical
618  * cpus. Assumes that cpu_logical_map(0) has already been initialized.
619  */
620 static void __init of_parse_and_init_cpus(void)
621 {
622 	struct device_node *dn;
623 
624 	for_each_of_cpu_node(dn) {
625 		u64 hwid = of_get_cpu_mpidr(dn);
626 
627 		if (hwid == INVALID_HWID)
628 			goto next;
629 
630 		if (is_mpidr_duplicate(cpu_count, hwid)) {
631 			pr_err("%pOF: duplicate cpu reg properties in the DT\n",
632 				dn);
633 			goto next;
634 		}
635 
636 		/*
637 		 * The numbering scheme requires that the boot CPU
638 		 * must be assigned logical id 0. Record it so that
639 		 * the logical map built from DT is validated and can
640 		 * be used.
641 		 */
642 		if (hwid == cpu_logical_map(0)) {
643 			if (bootcpu_valid) {
644 				pr_err("%pOF: duplicate boot cpu reg property in DT\n",
645 					dn);
646 				goto next;
647 			}
648 
649 			bootcpu_valid = true;
650 			early_map_cpu_to_node(0, of_node_to_nid(dn));
651 
652 			/*
653 			 * cpu_logical_map has already been
654 			 * initialized and the boot cpu doesn't need
655 			 * the enable-method so continue without
656 			 * incrementing cpu.
657 			 */
658 			continue;
659 		}
660 
661 		if (cpu_count >= NR_CPUS)
662 			goto next;
663 
664 		pr_debug("cpu logical map 0x%llx\n", hwid);
665 		cpu_logical_map(cpu_count) = hwid;
666 
667 		early_map_cpu_to_node(cpu_count, of_node_to_nid(dn));
668 next:
669 		cpu_count++;
670 	}
671 }
672 
673 /*
674  * Enumerate the possible CPU set from the device tree or ACPI and build the
675  * cpu logical map array containing MPIDR values related to logical
676  * cpus. Assumes that cpu_logical_map(0) has already been initialized.
677  */
678 void __init smp_init_cpus(void)
679 {
680 	int i;
681 
682 	if (acpi_disabled)
683 		of_parse_and_init_cpus();
684 	else
685 		acpi_parse_and_init_cpus();
686 
687 	if (cpu_count > nr_cpu_ids)
688 		pr_warn("Number of cores (%d) exceeds configured maximum of %u - clipping\n",
689 			cpu_count, nr_cpu_ids);
690 
691 	if (!bootcpu_valid) {
692 		pr_err("missing boot CPU MPIDR, not enabling secondaries\n");
693 		return;
694 	}
695 
696 	/*
697 	 * We need to set the cpu_logical_map entries before enabling
698 	 * the cpus so that cpu processor description entries (DT cpu nodes
699 	 * and ACPI MADT entries) can be retrieved by matching the cpu hwid
700 	 * with entries in cpu_logical_map while initializing the cpus.
701 	 * If the cpu set-up fails, invalidate the cpu_logical_map entry.
702 	 */
703 	for (i = 1; i < nr_cpu_ids; i++) {
704 		if (cpu_logical_map(i) != INVALID_HWID) {
705 			if (smp_cpu_setup(i))
706 				cpu_logical_map(i) = INVALID_HWID;
707 		}
708 	}
709 }
710 
711 void __init smp_prepare_cpus(unsigned int max_cpus)
712 {
713 	int err;
714 	unsigned int cpu;
715 	unsigned int this_cpu;
716 
717 	init_cpu_topology();
718 
719 	this_cpu = smp_processor_id();
720 	store_cpu_topology(this_cpu);
721 	numa_store_cpu_info(this_cpu);
722 	numa_add_cpu(this_cpu);
723 
724 	/*
725 	 * If UP is mandated by "nosmp" (which implies "maxcpus=0"), don't set
726 	 * secondary CPUs present.
727 	 */
728 	if (max_cpus == 0)
729 		return;
730 
731 	/*
732 	 * Initialise the present map (which describes the set of CPUs
733 	 * actually populated at the present time) and release the
734 	 * secondaries from the bootloader.
735 	 */
736 	for_each_possible_cpu(cpu) {
737 
738 		per_cpu(cpu_number, cpu) = cpu;
739 
740 		if (cpu == smp_processor_id())
741 			continue;
742 
743 		if (!cpu_ops[cpu])
744 			continue;
745 
746 		err = cpu_ops[cpu]->cpu_prepare(cpu);
747 		if (err)
748 			continue;
749 
750 		set_cpu_present(cpu, true);
751 		numa_store_cpu_info(cpu);
752 	}
753 }
754 
755 void (*__smp_cross_call)(const struct cpumask *, unsigned int);
756 
757 void __init set_smp_cross_call(void (*fn)(const struct cpumask *, unsigned int))
758 {
759 	__smp_cross_call = fn;
760 }
761 
762 static const char *ipi_types[NR_IPI] __tracepoint_string = {
763 #define S(x,s)	[x] = s
764 	S(IPI_RESCHEDULE, "Rescheduling interrupts"),
765 	S(IPI_CALL_FUNC, "Function call interrupts"),
766 	S(IPI_CPU_STOP, "CPU stop interrupts"),
767 	S(IPI_CPU_CRASH_STOP, "CPU stop (for crash dump) interrupts"),
768 	S(IPI_TIMER, "Timer broadcast interrupts"),
769 	S(IPI_IRQ_WORK, "IRQ work interrupts"),
770 	S(IPI_WAKEUP, "CPU wake-up interrupts"),
771 };
772 
773 static void smp_cross_call(const struct cpumask *target, unsigned int ipinr)
774 {
775 	trace_ipi_raise(target, ipi_types[ipinr]);
776 	__smp_cross_call(target, ipinr);
777 }
778 
779 void show_ipi_list(struct seq_file *p, int prec)
780 {
781 	unsigned int cpu, i;
782 
783 	for (i = 0; i < NR_IPI; i++) {
784 		seq_printf(p, "%*s%u:%s", prec - 1, "IPI", i,
785 			   prec >= 4 ? " " : "");
786 		for_each_online_cpu(cpu)
787 			seq_printf(p, "%10u ",
788 				   __get_irq_stat(cpu, ipi_irqs[i]));
789 		seq_printf(p, "      %s\n", ipi_types[i]);
790 	}
791 }
792 
793 u64 smp_irq_stat_cpu(unsigned int cpu)
794 {
795 	u64 sum = 0;
796 	int i;
797 
798 	for (i = 0; i < NR_IPI; i++)
799 		sum += __get_irq_stat(cpu, ipi_irqs[i]);
800 
801 	return sum;
802 }
803 
804 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
805 {
806 	smp_cross_call(mask, IPI_CALL_FUNC);
807 }
808 
809 void arch_send_call_function_single_ipi(int cpu)
810 {
811 	smp_cross_call(cpumask_of(cpu), IPI_CALL_FUNC);
812 }
813 
814 #ifdef CONFIG_ARM64_ACPI_PARKING_PROTOCOL
815 void arch_send_wakeup_ipi_mask(const struct cpumask *mask)
816 {
817 	smp_cross_call(mask, IPI_WAKEUP);
818 }
819 #endif
820 
821 #ifdef CONFIG_IRQ_WORK
822 void arch_irq_work_raise(void)
823 {
824 	if (__smp_cross_call)
825 		smp_cross_call(cpumask_of(smp_processor_id()), IPI_IRQ_WORK);
826 }
827 #endif
828 
829 static void local_cpu_stop(void)
830 {
831 	set_cpu_online(smp_processor_id(), false);
832 
833 	local_daif_mask();
834 	sdei_mask_local_cpu();
835 	cpu_park_loop();
836 }
837 
838 /*
839  * We need to implement panic_smp_self_stop() for parallel panic() calls, so
840  * that cpu_online_mask gets correctly updated and smp_send_stop() can skip
841  * CPUs that have already stopped themselves.
842  */
843 void panic_smp_self_stop(void)
844 {
845 	local_cpu_stop();
846 }
847 
848 #ifdef CONFIG_KEXEC_CORE
849 static atomic_t waiting_for_crash_ipi = ATOMIC_INIT(0);
850 #endif
851 
852 static void ipi_cpu_crash_stop(unsigned int cpu, struct pt_regs *regs)
853 {
854 #ifdef CONFIG_KEXEC_CORE
855 	crash_save_cpu(regs, cpu);
856 
857 	atomic_dec(&waiting_for_crash_ipi);
858 
859 	local_irq_disable();
860 	sdei_mask_local_cpu();
861 
862 #ifdef CONFIG_HOTPLUG_CPU
863 	if (cpu_ops[cpu]->cpu_die)
864 		cpu_ops[cpu]->cpu_die(cpu);
865 #endif
866 
867 	/* just in case */
868 	cpu_park_loop();
869 #endif
870 }
871 
872 /*
873  * Main handler for inter-processor interrupts
874  */
875 void handle_IPI(int ipinr, struct pt_regs *regs)
876 {
877 	unsigned int cpu = smp_processor_id();
878 	struct pt_regs *old_regs = set_irq_regs(regs);
879 
880 	if ((unsigned)ipinr < NR_IPI) {
881 		trace_ipi_entry_rcuidle(ipi_types[ipinr]);
882 		__inc_irq_stat(cpu, ipi_irqs[ipinr]);
883 	}
884 
885 	switch (ipinr) {
886 	case IPI_RESCHEDULE:
887 		scheduler_ipi();
888 		break;
889 
890 	case IPI_CALL_FUNC:
891 		irq_enter();
892 		generic_smp_call_function_interrupt();
893 		irq_exit();
894 		break;
895 
896 	case IPI_CPU_STOP:
897 		irq_enter();
898 		local_cpu_stop();
899 		irq_exit();
900 		break;
901 
902 	case IPI_CPU_CRASH_STOP:
903 		if (IS_ENABLED(CONFIG_KEXEC_CORE)) {
904 			irq_enter();
905 			ipi_cpu_crash_stop(cpu, regs);
906 
907 			unreachable();
908 		}
909 		break;
910 
911 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
912 	case IPI_TIMER:
913 		irq_enter();
914 		tick_receive_broadcast();
915 		irq_exit();
916 		break;
917 #endif
918 
919 #ifdef CONFIG_IRQ_WORK
920 	case IPI_IRQ_WORK:
921 		irq_enter();
922 		irq_work_run();
923 		irq_exit();
924 		break;
925 #endif
926 
927 #ifdef CONFIG_ARM64_ACPI_PARKING_PROTOCOL
928 	case IPI_WAKEUP:
929 		WARN_ONCE(!acpi_parking_protocol_valid(cpu),
930 			  "CPU%u: Wake-up IPI outside the ACPI parking protocol\n",
931 			  cpu);
932 		break;
933 #endif
934 
935 	default:
936 		pr_crit("CPU%u: Unknown IPI message 0x%x\n", cpu, ipinr);
937 		break;
938 	}
939 
940 	if ((unsigned)ipinr < NR_IPI)
941 		trace_ipi_exit_rcuidle(ipi_types[ipinr]);
942 	set_irq_regs(old_regs);
943 }
944 
945 void smp_send_reschedule(int cpu)
946 {
947 	smp_cross_call(cpumask_of(cpu), IPI_RESCHEDULE);
948 }
949 
950 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
951 void tick_broadcast(const struct cpumask *mask)
952 {
953 	smp_cross_call(mask, IPI_TIMER);
954 }
955 #endif
956 
957 void smp_send_stop(void)
958 {
959 	unsigned long timeout;
960 
961 	if (num_online_cpus() > 1) {
962 		cpumask_t mask;
963 
964 		cpumask_copy(&mask, cpu_online_mask);
965 		cpumask_clear_cpu(smp_processor_id(), &mask);
966 
967 		if (system_state <= SYSTEM_RUNNING)
968 			pr_crit("SMP: stopping secondary CPUs\n");
969 		smp_cross_call(&mask, IPI_CPU_STOP);
970 	}
971 
972 	/* Wait up to one second for other CPUs to stop */
973 	timeout = USEC_PER_SEC;
974 	while (num_online_cpus() > 1 && timeout--)
975 		udelay(1);
976 
977 	if (num_online_cpus() > 1)
978 		pr_warn("SMP: failed to stop secondary CPUs %*pbl\n",
979 			cpumask_pr_args(cpu_online_mask));
980 
981 	sdei_mask_local_cpu();
982 }
983 
984 #ifdef CONFIG_KEXEC_CORE
985 void crash_smp_send_stop(void)
986 {
987 	static int cpus_stopped;
988 	cpumask_t mask;
989 	unsigned long timeout;
990 
991 	/*
992 	 * This function can be called twice in panic path, but obviously
993 	 * we execute this only once.
994 	 */
995 	if (cpus_stopped)
996 		return;
997 
998 	cpus_stopped = 1;
999 
1000 	if (num_online_cpus() == 1) {
1001 		sdei_mask_local_cpu();
1002 		return;
1003 	}
1004 
1005 	cpumask_copy(&mask, cpu_online_mask);
1006 	cpumask_clear_cpu(smp_processor_id(), &mask);
1007 
1008 	atomic_set(&waiting_for_crash_ipi, num_online_cpus() - 1);
1009 
1010 	pr_crit("SMP: stopping secondary CPUs\n");
1011 	smp_cross_call(&mask, IPI_CPU_CRASH_STOP);
1012 
1013 	/* Wait up to one second for other CPUs to stop */
1014 	timeout = USEC_PER_SEC;
1015 	while ((atomic_read(&waiting_for_crash_ipi) > 0) && timeout--)
1016 		udelay(1);
1017 
1018 	if (atomic_read(&waiting_for_crash_ipi) > 0)
1019 		pr_warn("SMP: failed to stop secondary CPUs %*pbl\n",
1020 			cpumask_pr_args(&mask));
1021 
1022 	sdei_mask_local_cpu();
1023 }
1024 
1025 bool smp_crash_stop_failed(void)
1026 {
1027 	return (atomic_read(&waiting_for_crash_ipi) > 0);
1028 }
1029 #endif
1030 
1031 /*
1032  * not supported here
1033  */
1034 int setup_profiling_timer(unsigned int multiplier)
1035 {
1036 	return -EINVAL;
1037 }
1038 
1039 static bool have_cpu_die(void)
1040 {
1041 #ifdef CONFIG_HOTPLUG_CPU
1042 	int any_cpu = raw_smp_processor_id();
1043 
1044 	if (cpu_ops[any_cpu] && cpu_ops[any_cpu]->cpu_die)
1045 		return true;
1046 #endif
1047 	return false;
1048 }
1049 
1050 bool cpus_are_stuck_in_kernel(void)
1051 {
1052 	bool smp_spin_tables = (num_possible_cpus() > 1 && !have_cpu_die());
1053 
1054 	return !!cpus_stuck_in_kernel || smp_spin_tables;
1055 }
1056