xref: /openbmc/linux/arch/x86/kernel/apic/vector.c (revision 91cd9cb7ee1c081304d0e61f09e9faccb33d3df7)
1 /*
2  * Local APIC related interfaces to support IOAPIC, MSI, HT_IRQ etc.
3  *
4  * Copyright (C) 1997, 1998, 1999, 2000, 2009 Ingo Molnar, Hajnalka Szabo
5  *	Moved from arch/x86/kernel/apic/io_apic.c.
6  * Jiang Liu <jiang.liu@linux.intel.com>
7  *	Enable support of hierarchical irqdomains
8  *
9  * This program is free software; you can redistribute it and/or modify
10  * it under the terms of the GNU General Public License version 2 as
11  * published by the Free Software Foundation.
12  */
13 #include <linux/interrupt.h>
14 #include <linux/init.h>
15 #include <linux/compiler.h>
16 #include <linux/slab.h>
17 #include <asm/irqdomain.h>
18 #include <asm/hw_irq.h>
19 #include <asm/apic.h>
20 #include <asm/i8259.h>
21 #include <asm/desc.h>
22 #include <asm/irq_remapping.h>
23 
24 struct apic_chip_data {
25 	struct irq_cfg		cfg;
26 	cpumask_var_t		domain;
27 	cpumask_var_t		old_domain;
28 	u8			move_in_progress : 1;
29 };
30 
31 struct irq_domain *x86_vector_domain;
32 EXPORT_SYMBOL_GPL(x86_vector_domain);
33 static DEFINE_RAW_SPINLOCK(vector_lock);
34 static cpumask_var_t vector_cpumask, vector_searchmask, searched_cpumask;
35 static struct irq_chip lapic_controller;
36 #ifdef	CONFIG_X86_IO_APIC
37 static struct apic_chip_data *legacy_irq_data[NR_IRQS_LEGACY];
38 #endif
39 
40 void lock_vector_lock(void)
41 {
42 	/* Used to the online set of cpus does not change
43 	 * during assign_irq_vector.
44 	 */
45 	raw_spin_lock(&vector_lock);
46 }
47 
48 void unlock_vector_lock(void)
49 {
50 	raw_spin_unlock(&vector_lock);
51 }
52 
53 static struct apic_chip_data *apic_chip_data(struct irq_data *irq_data)
54 {
55 	if (!irq_data)
56 		return NULL;
57 
58 	while (irq_data->parent_data)
59 		irq_data = irq_data->parent_data;
60 
61 	return irq_data->chip_data;
62 }
63 
64 struct irq_cfg *irqd_cfg(struct irq_data *irq_data)
65 {
66 	struct apic_chip_data *data = apic_chip_data(irq_data);
67 
68 	return data ? &data->cfg : NULL;
69 }
70 EXPORT_SYMBOL_GPL(irqd_cfg);
71 
72 struct irq_cfg *irq_cfg(unsigned int irq)
73 {
74 	return irqd_cfg(irq_get_irq_data(irq));
75 }
76 
77 static struct apic_chip_data *alloc_apic_chip_data(int node)
78 {
79 	struct apic_chip_data *data;
80 
81 	data = kzalloc_node(sizeof(*data), GFP_KERNEL, node);
82 	if (!data)
83 		return NULL;
84 	if (!zalloc_cpumask_var_node(&data->domain, GFP_KERNEL, node))
85 		goto out_data;
86 	if (!zalloc_cpumask_var_node(&data->old_domain, GFP_KERNEL, node))
87 		goto out_domain;
88 	return data;
89 out_domain:
90 	free_cpumask_var(data->domain);
91 out_data:
92 	kfree(data);
93 	return NULL;
94 }
95 
96 static void free_apic_chip_data(struct apic_chip_data *data)
97 {
98 	if (data) {
99 		free_cpumask_var(data->domain);
100 		free_cpumask_var(data->old_domain);
101 		kfree(data);
102 	}
103 }
104 
105 static int __assign_irq_vector(int irq, struct apic_chip_data *d,
106 			       const struct cpumask *mask)
107 {
108 	/*
109 	 * NOTE! The local APIC isn't very good at handling
110 	 * multiple interrupts at the same interrupt level.
111 	 * As the interrupt level is determined by taking the
112 	 * vector number and shifting that right by 4, we
113 	 * want to spread these out a bit so that they don't
114 	 * all fall in the same interrupt level.
115 	 *
116 	 * Also, we've got to be careful not to trash gate
117 	 * 0x80, because int 0x80 is hm, kind of importantish. ;)
118 	 */
119 	static int current_vector = FIRST_EXTERNAL_VECTOR + VECTOR_OFFSET_START;
120 	static int current_offset = VECTOR_OFFSET_START % 16;
121 	int cpu, vector;
122 
123 	/*
124 	 * If there is still a move in progress or the previous move has not
125 	 * been cleaned up completely, tell the caller to come back later.
126 	 */
127 	if (d->move_in_progress ||
128 	    cpumask_intersects(d->old_domain, cpu_online_mask))
129 		return -EBUSY;
130 
131 	/* Only try and allocate irqs on cpus that are present */
132 	cpumask_clear(d->old_domain);
133 	cpumask_clear(searched_cpumask);
134 	cpu = cpumask_first_and(mask, cpu_online_mask);
135 	while (cpu < nr_cpu_ids) {
136 		int new_cpu, offset;
137 
138 		/* Get the possible target cpus for @mask/@cpu from the apic */
139 		apic->vector_allocation_domain(cpu, vector_cpumask, mask);
140 
141 		/*
142 		 * Clear the offline cpus from @vector_cpumask for searching
143 		 * and verify whether the result overlaps with @mask. If true,
144 		 * then the call to apic->cpu_mask_to_apicid() will
145 		 * succeed as well. If not, no point in trying to find a
146 		 * vector in this mask.
147 		 */
148 		cpumask_and(vector_searchmask, vector_cpumask, cpu_online_mask);
149 		if (!cpumask_intersects(vector_searchmask, mask))
150 			goto next_cpu;
151 
152 		if (cpumask_subset(vector_cpumask, d->domain)) {
153 			if (cpumask_equal(vector_cpumask, d->domain))
154 				goto success;
155 			/*
156 			 * Mark the cpus which are not longer in the mask for
157 			 * cleanup.
158 			 */
159 			cpumask_andnot(d->old_domain, d->domain, vector_cpumask);
160 			vector = d->cfg.vector;
161 			goto update;
162 		}
163 
164 		vector = current_vector;
165 		offset = current_offset;
166 next:
167 		vector += 16;
168 		if (vector >= first_system_vector) {
169 			offset = (offset + 1) % 16;
170 			vector = FIRST_EXTERNAL_VECTOR + offset;
171 		}
172 
173 		/* If the search wrapped around, try the next cpu */
174 		if (unlikely(current_vector == vector))
175 			goto next_cpu;
176 
177 		if (test_bit(vector, used_vectors))
178 			goto next;
179 
180 		for_each_cpu(new_cpu, vector_searchmask) {
181 			if (!IS_ERR_OR_NULL(per_cpu(vector_irq, new_cpu)[vector]))
182 				goto next;
183 		}
184 		/* Found one! */
185 		current_vector = vector;
186 		current_offset = offset;
187 		/* Schedule the old vector for cleanup on all cpus */
188 		if (d->cfg.vector)
189 			cpumask_copy(d->old_domain, d->domain);
190 		for_each_cpu(new_cpu, vector_searchmask)
191 			per_cpu(vector_irq, new_cpu)[vector] = irq_to_desc(irq);
192 		goto update;
193 
194 next_cpu:
195 		/*
196 		 * We exclude the current @vector_cpumask from the requested
197 		 * @mask and try again with the next online cpu in the
198 		 * result. We cannot modify @mask, so we use @vector_cpumask
199 		 * as a temporary buffer here as it will be reassigned when
200 		 * calling apic->vector_allocation_domain() above.
201 		 */
202 		cpumask_or(searched_cpumask, searched_cpumask, vector_cpumask);
203 		cpumask_andnot(vector_cpumask, mask, searched_cpumask);
204 		cpu = cpumask_first_and(vector_cpumask, cpu_online_mask);
205 		continue;
206 	}
207 	return -ENOSPC;
208 
209 update:
210 	/*
211 	 * Exclude offline cpus from the cleanup mask and set the
212 	 * move_in_progress flag when the result is not empty.
213 	 */
214 	cpumask_and(d->old_domain, d->old_domain, cpu_online_mask);
215 	d->move_in_progress = !cpumask_empty(d->old_domain);
216 	d->cfg.old_vector = d->move_in_progress ? d->cfg.vector : 0;
217 	d->cfg.vector = vector;
218 	cpumask_copy(d->domain, vector_cpumask);
219 success:
220 	/*
221 	 * Cache destination APIC IDs into cfg->dest_apicid. This cannot fail
222 	 * as we already established, that mask & d->domain & cpu_online_mask
223 	 * is not empty.
224 	 *
225 	 * vector_searchmask is a subset of d->domain and has the offline
226 	 * cpus masked out.
227 	 */
228 	cpumask_and(vector_searchmask, vector_searchmask, mask);
229 	BUG_ON(apic->cpu_mask_to_apicid(vector_searchmask, &d->cfg.dest_apicid));
230 	return 0;
231 }
232 
233 static int assign_irq_vector(int irq, struct apic_chip_data *data,
234 			     const struct cpumask *mask)
235 {
236 	int err;
237 	unsigned long flags;
238 
239 	raw_spin_lock_irqsave(&vector_lock, flags);
240 	err = __assign_irq_vector(irq, data, mask);
241 	raw_spin_unlock_irqrestore(&vector_lock, flags);
242 	return err;
243 }
244 
245 static int assign_irq_vector_policy(int irq, int node,
246 				    struct apic_chip_data *data,
247 				    struct irq_alloc_info *info)
248 {
249 	if (info && info->mask)
250 		return assign_irq_vector(irq, data, info->mask);
251 	if (node != NUMA_NO_NODE &&
252 	    assign_irq_vector(irq, data, cpumask_of_node(node)) == 0)
253 		return 0;
254 	return assign_irq_vector(irq, data, apic->target_cpus());
255 }
256 
257 static void clear_irq_vector(int irq, struct apic_chip_data *data)
258 {
259 	struct irq_desc *desc;
260 	int cpu, vector;
261 
262 	if (!data->cfg.vector)
263 		return;
264 
265 	vector = data->cfg.vector;
266 	for_each_cpu_and(cpu, data->domain, cpu_online_mask)
267 		per_cpu(vector_irq, cpu)[vector] = VECTOR_UNUSED;
268 
269 	data->cfg.vector = 0;
270 	cpumask_clear(data->domain);
271 
272 	/*
273 	 * If move is in progress or the old_domain mask is not empty,
274 	 * i.e. the cleanup IPI has not been processed yet, we need to remove
275 	 * the old references to desc from all cpus vector tables.
276 	 */
277 	if (!data->move_in_progress && cpumask_empty(data->old_domain))
278 		return;
279 
280 	desc = irq_to_desc(irq);
281 	for_each_cpu_and(cpu, data->old_domain, cpu_online_mask) {
282 		for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS;
283 		     vector++) {
284 			if (per_cpu(vector_irq, cpu)[vector] != desc)
285 				continue;
286 			per_cpu(vector_irq, cpu)[vector] = VECTOR_UNUSED;
287 			break;
288 		}
289 	}
290 	data->move_in_progress = 0;
291 }
292 
293 void init_irq_alloc_info(struct irq_alloc_info *info,
294 			 const struct cpumask *mask)
295 {
296 	memset(info, 0, sizeof(*info));
297 	info->mask = mask;
298 }
299 
300 void copy_irq_alloc_info(struct irq_alloc_info *dst, struct irq_alloc_info *src)
301 {
302 	if (src)
303 		*dst = *src;
304 	else
305 		memset(dst, 0, sizeof(*dst));
306 }
307 
308 static void x86_vector_free_irqs(struct irq_domain *domain,
309 				 unsigned int virq, unsigned int nr_irqs)
310 {
311 	struct apic_chip_data *apic_data;
312 	struct irq_data *irq_data;
313 	unsigned long flags;
314 	int i;
315 
316 	for (i = 0; i < nr_irqs; i++) {
317 		irq_data = irq_domain_get_irq_data(x86_vector_domain, virq + i);
318 		if (irq_data && irq_data->chip_data) {
319 			raw_spin_lock_irqsave(&vector_lock, flags);
320 			clear_irq_vector(virq + i, irq_data->chip_data);
321 			apic_data = irq_data->chip_data;
322 			irq_domain_reset_irq_data(irq_data);
323 			raw_spin_unlock_irqrestore(&vector_lock, flags);
324 			free_apic_chip_data(apic_data);
325 #ifdef	CONFIG_X86_IO_APIC
326 			if (virq + i < nr_legacy_irqs())
327 				legacy_irq_data[virq + i] = NULL;
328 #endif
329 		}
330 	}
331 }
332 
333 static int x86_vector_alloc_irqs(struct irq_domain *domain, unsigned int virq,
334 				 unsigned int nr_irqs, void *arg)
335 {
336 	struct irq_alloc_info *info = arg;
337 	struct apic_chip_data *data;
338 	struct irq_data *irq_data;
339 	int i, err, node;
340 
341 	if (disable_apic)
342 		return -ENXIO;
343 
344 	/* Currently vector allocator can't guarantee contiguous allocations */
345 	if ((info->flags & X86_IRQ_ALLOC_CONTIGUOUS_VECTORS) && nr_irqs > 1)
346 		return -ENOSYS;
347 
348 	for (i = 0; i < nr_irqs; i++) {
349 		irq_data = irq_domain_get_irq_data(domain, virq + i);
350 		BUG_ON(!irq_data);
351 		node = irq_data_get_node(irq_data);
352 #ifdef	CONFIG_X86_IO_APIC
353 		if (virq + i < nr_legacy_irqs() && legacy_irq_data[virq + i])
354 			data = legacy_irq_data[virq + i];
355 		else
356 #endif
357 			data = alloc_apic_chip_data(node);
358 		if (!data) {
359 			err = -ENOMEM;
360 			goto error;
361 		}
362 
363 		irq_data->chip = &lapic_controller;
364 		irq_data->chip_data = data;
365 		irq_data->hwirq = virq + i;
366 		err = assign_irq_vector_policy(virq + i, node, data, info);
367 		if (err)
368 			goto error;
369 	}
370 
371 	return 0;
372 
373 error:
374 	x86_vector_free_irqs(domain, virq, i + 1);
375 	return err;
376 }
377 
378 static const struct irq_domain_ops x86_vector_domain_ops = {
379 	.alloc	= x86_vector_alloc_irqs,
380 	.free	= x86_vector_free_irqs,
381 };
382 
383 int __init arch_probe_nr_irqs(void)
384 {
385 	int nr;
386 
387 	if (nr_irqs > (NR_VECTORS * nr_cpu_ids))
388 		nr_irqs = NR_VECTORS * nr_cpu_ids;
389 
390 	nr = (gsi_top + nr_legacy_irqs()) + 8 * nr_cpu_ids;
391 #if defined(CONFIG_PCI_MSI) || defined(CONFIG_HT_IRQ)
392 	/*
393 	 * for MSI and HT dyn irq
394 	 */
395 	if (gsi_top <= NR_IRQS_LEGACY)
396 		nr +=  8 * nr_cpu_ids;
397 	else
398 		nr += gsi_top * 16;
399 #endif
400 	if (nr < nr_irqs)
401 		nr_irqs = nr;
402 
403 	/*
404 	 * We don't know if PIC is present at this point so we need to do
405 	 * probe() to get the right number of legacy IRQs.
406 	 */
407 	return legacy_pic->probe();
408 }
409 
410 #ifdef	CONFIG_X86_IO_APIC
411 static void init_legacy_irqs(void)
412 {
413 	int i, node = cpu_to_node(0);
414 	struct apic_chip_data *data;
415 
416 	/*
417 	 * For legacy IRQ's, start with assigning irq0 to irq15 to
418 	 * ISA_IRQ_VECTOR(i) for all cpu's.
419 	 */
420 	for (i = 0; i < nr_legacy_irqs(); i++) {
421 		data = legacy_irq_data[i] = alloc_apic_chip_data(node);
422 		BUG_ON(!data);
423 
424 		data->cfg.vector = ISA_IRQ_VECTOR(i);
425 		cpumask_setall(data->domain);
426 		irq_set_chip_data(i, data);
427 	}
428 }
429 #else
430 static void init_legacy_irqs(void) { }
431 #endif
432 
433 int __init arch_early_irq_init(void)
434 {
435 	struct fwnode_handle *fn;
436 
437 	init_legacy_irqs();
438 
439 	fn = irq_domain_alloc_named_fwnode("VECTOR");
440 	BUG_ON(!fn);
441 	x86_vector_domain = irq_domain_create_tree(fn, &x86_vector_domain_ops,
442 						   NULL);
443 	BUG_ON(x86_vector_domain == NULL);
444 	irq_domain_free_fwnode(fn);
445 	irq_set_default_host(x86_vector_domain);
446 
447 	arch_init_msi_domain(x86_vector_domain);
448 	arch_init_htirq_domain(x86_vector_domain);
449 
450 	BUG_ON(!alloc_cpumask_var(&vector_cpumask, GFP_KERNEL));
451 	BUG_ON(!alloc_cpumask_var(&vector_searchmask, GFP_KERNEL));
452 	BUG_ON(!alloc_cpumask_var(&searched_cpumask, GFP_KERNEL));
453 
454 	return arch_early_ioapic_init();
455 }
456 
457 /* Initialize vector_irq on a new cpu */
458 static void __setup_vector_irq(int cpu)
459 {
460 	struct apic_chip_data *data;
461 	struct irq_desc *desc;
462 	int irq, vector;
463 
464 	/* Mark the inuse vectors */
465 	for_each_irq_desc(irq, desc) {
466 		struct irq_data *idata = irq_desc_get_irq_data(desc);
467 
468 		data = apic_chip_data(idata);
469 		if (!data || !cpumask_test_cpu(cpu, data->domain))
470 			continue;
471 		vector = data->cfg.vector;
472 		per_cpu(vector_irq, cpu)[vector] = desc;
473 	}
474 	/* Mark the free vectors */
475 	for (vector = 0; vector < NR_VECTORS; ++vector) {
476 		desc = per_cpu(vector_irq, cpu)[vector];
477 		if (IS_ERR_OR_NULL(desc))
478 			continue;
479 
480 		data = apic_chip_data(irq_desc_get_irq_data(desc));
481 		if (!cpumask_test_cpu(cpu, data->domain))
482 			per_cpu(vector_irq, cpu)[vector] = VECTOR_UNUSED;
483 	}
484 }
485 
486 /*
487  * Setup the vector to irq mappings. Must be called with vector_lock held.
488  */
489 void setup_vector_irq(int cpu)
490 {
491 	int irq;
492 
493 	lockdep_assert_held(&vector_lock);
494 	/*
495 	 * On most of the platforms, legacy PIC delivers the interrupts on the
496 	 * boot cpu. But there are certain platforms where PIC interrupts are
497 	 * delivered to multiple cpu's. If the legacy IRQ is handled by the
498 	 * legacy PIC, for the new cpu that is coming online, setup the static
499 	 * legacy vector to irq mapping:
500 	 */
501 	for (irq = 0; irq < nr_legacy_irqs(); irq++)
502 		per_cpu(vector_irq, cpu)[ISA_IRQ_VECTOR(irq)] = irq_to_desc(irq);
503 
504 	__setup_vector_irq(cpu);
505 }
506 
507 static int apic_retrigger_irq(struct irq_data *irq_data)
508 {
509 	struct apic_chip_data *data = apic_chip_data(irq_data);
510 	unsigned long flags;
511 	int cpu;
512 
513 	raw_spin_lock_irqsave(&vector_lock, flags);
514 	cpu = cpumask_first_and(data->domain, cpu_online_mask);
515 	apic->send_IPI_mask(cpumask_of(cpu), data->cfg.vector);
516 	raw_spin_unlock_irqrestore(&vector_lock, flags);
517 
518 	return 1;
519 }
520 
521 void apic_ack_edge(struct irq_data *data)
522 {
523 	irq_complete_move(irqd_cfg(data));
524 	irq_move_irq(data);
525 	ack_APIC_irq();
526 }
527 
528 static int apic_set_affinity(struct irq_data *irq_data,
529 			     const struct cpumask *dest, bool force)
530 {
531 	struct apic_chip_data *data = irq_data->chip_data;
532 	int err, irq = irq_data->irq;
533 
534 	if (!IS_ENABLED(CONFIG_SMP))
535 		return -EPERM;
536 
537 	if (!cpumask_intersects(dest, cpu_online_mask))
538 		return -EINVAL;
539 
540 	err = assign_irq_vector(irq, data, dest);
541 	return err ? err : IRQ_SET_MASK_OK;
542 }
543 
544 static struct irq_chip lapic_controller = {
545 	.name			= "APIC",
546 	.irq_ack		= apic_ack_edge,
547 	.irq_set_affinity	= apic_set_affinity,
548 	.irq_retrigger		= apic_retrigger_irq,
549 };
550 
551 #ifdef CONFIG_SMP
552 static void __send_cleanup_vector(struct apic_chip_data *data)
553 {
554 	raw_spin_lock(&vector_lock);
555 	cpumask_and(data->old_domain, data->old_domain, cpu_online_mask);
556 	data->move_in_progress = 0;
557 	if (!cpumask_empty(data->old_domain))
558 		apic->send_IPI_mask(data->old_domain, IRQ_MOVE_CLEANUP_VECTOR);
559 	raw_spin_unlock(&vector_lock);
560 }
561 
562 void send_cleanup_vector(struct irq_cfg *cfg)
563 {
564 	struct apic_chip_data *data;
565 
566 	data = container_of(cfg, struct apic_chip_data, cfg);
567 	if (data->move_in_progress)
568 		__send_cleanup_vector(data);
569 }
570 
571 asmlinkage __visible void __irq_entry smp_irq_move_cleanup_interrupt(void)
572 {
573 	unsigned vector, me;
574 
575 	entering_ack_irq();
576 
577 	/* Prevent vectors vanishing under us */
578 	raw_spin_lock(&vector_lock);
579 
580 	me = smp_processor_id();
581 	for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
582 		struct apic_chip_data *data;
583 		struct irq_desc *desc;
584 		unsigned int irr;
585 
586 	retry:
587 		desc = __this_cpu_read(vector_irq[vector]);
588 		if (IS_ERR_OR_NULL(desc))
589 			continue;
590 
591 		if (!raw_spin_trylock(&desc->lock)) {
592 			raw_spin_unlock(&vector_lock);
593 			cpu_relax();
594 			raw_spin_lock(&vector_lock);
595 			goto retry;
596 		}
597 
598 		data = apic_chip_data(irq_desc_get_irq_data(desc));
599 		if (!data)
600 			goto unlock;
601 
602 		/*
603 		 * Nothing to cleanup if irq migration is in progress
604 		 * or this cpu is not set in the cleanup mask.
605 		 */
606 		if (data->move_in_progress ||
607 		    !cpumask_test_cpu(me, data->old_domain))
608 			goto unlock;
609 
610 		/*
611 		 * We have two cases to handle here:
612 		 * 1) vector is unchanged but the target mask got reduced
613 		 * 2) vector and the target mask has changed
614 		 *
615 		 * #1 is obvious, but in #2 we have two vectors with the same
616 		 * irq descriptor: the old and the new vector. So we need to
617 		 * make sure that we only cleanup the old vector. The new
618 		 * vector has the current @vector number in the config and
619 		 * this cpu is part of the target mask. We better leave that
620 		 * one alone.
621 		 */
622 		if (vector == data->cfg.vector &&
623 		    cpumask_test_cpu(me, data->domain))
624 			goto unlock;
625 
626 		irr = apic_read(APIC_IRR + (vector / 32 * 0x10));
627 		/*
628 		 * Check if the vector that needs to be cleanedup is
629 		 * registered at the cpu's IRR. If so, then this is not
630 		 * the best time to clean it up. Lets clean it up in the
631 		 * next attempt by sending another IRQ_MOVE_CLEANUP_VECTOR
632 		 * to myself.
633 		 */
634 		if (irr  & (1 << (vector % 32))) {
635 			apic->send_IPI_self(IRQ_MOVE_CLEANUP_VECTOR);
636 			goto unlock;
637 		}
638 		__this_cpu_write(vector_irq[vector], VECTOR_UNUSED);
639 		cpumask_clear_cpu(me, data->old_domain);
640 unlock:
641 		raw_spin_unlock(&desc->lock);
642 	}
643 
644 	raw_spin_unlock(&vector_lock);
645 
646 	exiting_irq();
647 }
648 
649 static void __irq_complete_move(struct irq_cfg *cfg, unsigned vector)
650 {
651 	unsigned me;
652 	struct apic_chip_data *data;
653 
654 	data = container_of(cfg, struct apic_chip_data, cfg);
655 	if (likely(!data->move_in_progress))
656 		return;
657 
658 	me = smp_processor_id();
659 	if (vector == data->cfg.vector && cpumask_test_cpu(me, data->domain))
660 		__send_cleanup_vector(data);
661 }
662 
663 void irq_complete_move(struct irq_cfg *cfg)
664 {
665 	__irq_complete_move(cfg, ~get_irq_regs()->orig_ax);
666 }
667 
668 /*
669  * Called from fixup_irqs() with @desc->lock held and interrupts disabled.
670  */
671 void irq_force_complete_move(struct irq_desc *desc)
672 {
673 	struct irq_data *irqdata;
674 	struct apic_chip_data *data;
675 	struct irq_cfg *cfg;
676 	unsigned int cpu;
677 
678 	/*
679 	 * The function is called for all descriptors regardless of which
680 	 * irqdomain they belong to. For example if an IRQ is provided by
681 	 * an irq_chip as part of a GPIO driver, the chip data for that
682 	 * descriptor is specific to the irq_chip in question.
683 	 *
684 	 * Check first that the chip_data is what we expect
685 	 * (apic_chip_data) before touching it any further.
686 	 */
687 	irqdata = irq_domain_get_irq_data(x86_vector_domain,
688 					  irq_desc_get_irq(desc));
689 	if (!irqdata)
690 		return;
691 
692 	data = apic_chip_data(irqdata);
693 	cfg = data ? &data->cfg : NULL;
694 
695 	if (!cfg)
696 		return;
697 
698 	/*
699 	 * This is tricky. If the cleanup of @data->old_domain has not been
700 	 * done yet, then the following setaffinity call will fail with
701 	 * -EBUSY. This can leave the interrupt in a stale state.
702 	 *
703 	 * All CPUs are stuck in stop machine with interrupts disabled so
704 	 * calling __irq_complete_move() would be completely pointless.
705 	 */
706 	raw_spin_lock(&vector_lock);
707 	/*
708 	 * Clean out all offline cpus (including the outgoing one) from the
709 	 * old_domain mask.
710 	 */
711 	cpumask_and(data->old_domain, data->old_domain, cpu_online_mask);
712 
713 	/*
714 	 * If move_in_progress is cleared and the old_domain mask is empty,
715 	 * then there is nothing to cleanup. fixup_irqs() will take care of
716 	 * the stale vectors on the outgoing cpu.
717 	 */
718 	if (!data->move_in_progress && cpumask_empty(data->old_domain)) {
719 		raw_spin_unlock(&vector_lock);
720 		return;
721 	}
722 
723 	/*
724 	 * 1) The interrupt is in move_in_progress state. That means that we
725 	 *    have not seen an interrupt since the io_apic was reprogrammed to
726 	 *    the new vector.
727 	 *
728 	 * 2) The interrupt has fired on the new vector, but the cleanup IPIs
729 	 *    have not been processed yet.
730 	 */
731 	if (data->move_in_progress) {
732 		/*
733 		 * In theory there is a race:
734 		 *
735 		 * set_ioapic(new_vector) <-- Interrupt is raised before update
736 		 *			      is effective, i.e. it's raised on
737 		 *			      the old vector.
738 		 *
739 		 * So if the target cpu cannot handle that interrupt before
740 		 * the old vector is cleaned up, we get a spurious interrupt
741 		 * and in the worst case the ioapic irq line becomes stale.
742 		 *
743 		 * But in case of cpu hotplug this should be a non issue
744 		 * because if the affinity update happens right before all
745 		 * cpus rendevouz in stop machine, there is no way that the
746 		 * interrupt can be blocked on the target cpu because all cpus
747 		 * loops first with interrupts enabled in stop machine, so the
748 		 * old vector is not yet cleaned up when the interrupt fires.
749 		 *
750 		 * So the only way to run into this issue is if the delivery
751 		 * of the interrupt on the apic/system bus would be delayed
752 		 * beyond the point where the target cpu disables interrupts
753 		 * in stop machine. I doubt that it can happen, but at least
754 		 * there is a theroretical chance. Virtualization might be
755 		 * able to expose this, but AFAICT the IOAPIC emulation is not
756 		 * as stupid as the real hardware.
757 		 *
758 		 * Anyway, there is nothing we can do about that at this point
759 		 * w/o refactoring the whole fixup_irq() business completely.
760 		 * We print at least the irq number and the old vector number,
761 		 * so we have the necessary information when a problem in that
762 		 * area arises.
763 		 */
764 		pr_warn("IRQ fixup: irq %d move in progress, old vector %d\n",
765 			irqdata->irq, cfg->old_vector);
766 	}
767 	/*
768 	 * If old_domain is not empty, then other cpus still have the irq
769 	 * descriptor set in their vector array. Clean it up.
770 	 */
771 	for_each_cpu(cpu, data->old_domain)
772 		per_cpu(vector_irq, cpu)[cfg->old_vector] = VECTOR_UNUSED;
773 
774 	/* Cleanup the left overs of the (half finished) move */
775 	cpumask_clear(data->old_domain);
776 	data->move_in_progress = 0;
777 	raw_spin_unlock(&vector_lock);
778 }
779 #endif
780 
781 static void __init print_APIC_field(int base)
782 {
783 	int i;
784 
785 	printk(KERN_DEBUG);
786 
787 	for (i = 0; i < 8; i++)
788 		pr_cont("%08x", apic_read(base + i*0x10));
789 
790 	pr_cont("\n");
791 }
792 
793 static void __init print_local_APIC(void *dummy)
794 {
795 	unsigned int i, v, ver, maxlvt;
796 	u64 icr;
797 
798 	pr_debug("printing local APIC contents on CPU#%d/%d:\n",
799 		 smp_processor_id(), hard_smp_processor_id());
800 	v = apic_read(APIC_ID);
801 	pr_info("... APIC ID:      %08x (%01x)\n", v, read_apic_id());
802 	v = apic_read(APIC_LVR);
803 	pr_info("... APIC VERSION: %08x\n", v);
804 	ver = GET_APIC_VERSION(v);
805 	maxlvt = lapic_get_maxlvt();
806 
807 	v = apic_read(APIC_TASKPRI);
808 	pr_debug("... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
809 
810 	/* !82489DX */
811 	if (APIC_INTEGRATED(ver)) {
812 		if (!APIC_XAPIC(ver)) {
813 			v = apic_read(APIC_ARBPRI);
814 			pr_debug("... APIC ARBPRI: %08x (%02x)\n",
815 				 v, v & APIC_ARBPRI_MASK);
816 		}
817 		v = apic_read(APIC_PROCPRI);
818 		pr_debug("... APIC PROCPRI: %08x\n", v);
819 	}
820 
821 	/*
822 	 * Remote read supported only in the 82489DX and local APIC for
823 	 * Pentium processors.
824 	 */
825 	if (!APIC_INTEGRATED(ver) || maxlvt == 3) {
826 		v = apic_read(APIC_RRR);
827 		pr_debug("... APIC RRR: %08x\n", v);
828 	}
829 
830 	v = apic_read(APIC_LDR);
831 	pr_debug("... APIC LDR: %08x\n", v);
832 	if (!x2apic_enabled()) {
833 		v = apic_read(APIC_DFR);
834 		pr_debug("... APIC DFR: %08x\n", v);
835 	}
836 	v = apic_read(APIC_SPIV);
837 	pr_debug("... APIC SPIV: %08x\n", v);
838 
839 	pr_debug("... APIC ISR field:\n");
840 	print_APIC_field(APIC_ISR);
841 	pr_debug("... APIC TMR field:\n");
842 	print_APIC_field(APIC_TMR);
843 	pr_debug("... APIC IRR field:\n");
844 	print_APIC_field(APIC_IRR);
845 
846 	/* !82489DX */
847 	if (APIC_INTEGRATED(ver)) {
848 		/* Due to the Pentium erratum 3AP. */
849 		if (maxlvt > 3)
850 			apic_write(APIC_ESR, 0);
851 
852 		v = apic_read(APIC_ESR);
853 		pr_debug("... APIC ESR: %08x\n", v);
854 	}
855 
856 	icr = apic_icr_read();
857 	pr_debug("... APIC ICR: %08x\n", (u32)icr);
858 	pr_debug("... APIC ICR2: %08x\n", (u32)(icr >> 32));
859 
860 	v = apic_read(APIC_LVTT);
861 	pr_debug("... APIC LVTT: %08x\n", v);
862 
863 	if (maxlvt > 3) {
864 		/* PC is LVT#4. */
865 		v = apic_read(APIC_LVTPC);
866 		pr_debug("... APIC LVTPC: %08x\n", v);
867 	}
868 	v = apic_read(APIC_LVT0);
869 	pr_debug("... APIC LVT0: %08x\n", v);
870 	v = apic_read(APIC_LVT1);
871 	pr_debug("... APIC LVT1: %08x\n", v);
872 
873 	if (maxlvt > 2) {
874 		/* ERR is LVT#3. */
875 		v = apic_read(APIC_LVTERR);
876 		pr_debug("... APIC LVTERR: %08x\n", v);
877 	}
878 
879 	v = apic_read(APIC_TMICT);
880 	pr_debug("... APIC TMICT: %08x\n", v);
881 	v = apic_read(APIC_TMCCT);
882 	pr_debug("... APIC TMCCT: %08x\n", v);
883 	v = apic_read(APIC_TDCR);
884 	pr_debug("... APIC TDCR: %08x\n", v);
885 
886 	if (boot_cpu_has(X86_FEATURE_EXTAPIC)) {
887 		v = apic_read(APIC_EFEAT);
888 		maxlvt = (v >> 16) & 0xff;
889 		pr_debug("... APIC EFEAT: %08x\n", v);
890 		v = apic_read(APIC_ECTRL);
891 		pr_debug("... APIC ECTRL: %08x\n", v);
892 		for (i = 0; i < maxlvt; i++) {
893 			v = apic_read(APIC_EILVTn(i));
894 			pr_debug("... APIC EILVT%d: %08x\n", i, v);
895 		}
896 	}
897 	pr_cont("\n");
898 }
899 
900 static void __init print_local_APICs(int maxcpu)
901 {
902 	int cpu;
903 
904 	if (!maxcpu)
905 		return;
906 
907 	preempt_disable();
908 	for_each_online_cpu(cpu) {
909 		if (cpu >= maxcpu)
910 			break;
911 		smp_call_function_single(cpu, print_local_APIC, NULL, 1);
912 	}
913 	preempt_enable();
914 }
915 
916 static void __init print_PIC(void)
917 {
918 	unsigned int v;
919 	unsigned long flags;
920 
921 	if (!nr_legacy_irqs())
922 		return;
923 
924 	pr_debug("\nprinting PIC contents\n");
925 
926 	raw_spin_lock_irqsave(&i8259A_lock, flags);
927 
928 	v = inb(0xa1) << 8 | inb(0x21);
929 	pr_debug("... PIC  IMR: %04x\n", v);
930 
931 	v = inb(0xa0) << 8 | inb(0x20);
932 	pr_debug("... PIC  IRR: %04x\n", v);
933 
934 	outb(0x0b, 0xa0);
935 	outb(0x0b, 0x20);
936 	v = inb(0xa0) << 8 | inb(0x20);
937 	outb(0x0a, 0xa0);
938 	outb(0x0a, 0x20);
939 
940 	raw_spin_unlock_irqrestore(&i8259A_lock, flags);
941 
942 	pr_debug("... PIC  ISR: %04x\n", v);
943 
944 	v = inb(0x4d1) << 8 | inb(0x4d0);
945 	pr_debug("... PIC ELCR: %04x\n", v);
946 }
947 
948 static int show_lapic __initdata = 1;
949 static __init int setup_show_lapic(char *arg)
950 {
951 	int num = -1;
952 
953 	if (strcmp(arg, "all") == 0) {
954 		show_lapic = CONFIG_NR_CPUS;
955 	} else {
956 		get_option(&arg, &num);
957 		if (num >= 0)
958 			show_lapic = num;
959 	}
960 
961 	return 1;
962 }
963 __setup("show_lapic=", setup_show_lapic);
964 
965 static int __init print_ICs(void)
966 {
967 	if (apic_verbosity == APIC_QUIET)
968 		return 0;
969 
970 	print_PIC();
971 
972 	/* don't print out if apic is not there */
973 	if (!boot_cpu_has(X86_FEATURE_APIC) && !apic_from_smp_config())
974 		return 0;
975 
976 	print_local_APICs(show_lapic);
977 	print_IO_APICs();
978 
979 	return 0;
980 }
981 
982 late_initcall(print_ICs);
983