xref: /openbmc/linux/arch/x86/kernel/apic/vector.c (revision 3e26a691)
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_and() 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 	BUG_ON(apic->cpu_mask_to_apicid_and(mask, d->domain,
226 					    &d->cfg.dest_apicid));
227 	return 0;
228 }
229 
230 static int assign_irq_vector(int irq, struct apic_chip_data *data,
231 			     const struct cpumask *mask)
232 {
233 	int err;
234 	unsigned long flags;
235 
236 	raw_spin_lock_irqsave(&vector_lock, flags);
237 	err = __assign_irq_vector(irq, data, mask);
238 	raw_spin_unlock_irqrestore(&vector_lock, flags);
239 	return err;
240 }
241 
242 static int assign_irq_vector_policy(int irq, int node,
243 				    struct apic_chip_data *data,
244 				    struct irq_alloc_info *info)
245 {
246 	if (info && info->mask)
247 		return assign_irq_vector(irq, data, info->mask);
248 	if (node != NUMA_NO_NODE &&
249 	    assign_irq_vector(irq, data, cpumask_of_node(node)) == 0)
250 		return 0;
251 	return assign_irq_vector(irq, data, apic->target_cpus());
252 }
253 
254 static void clear_irq_vector(int irq, struct apic_chip_data *data)
255 {
256 	struct irq_desc *desc;
257 	int cpu, vector;
258 
259 	BUG_ON(!data->cfg.vector);
260 
261 	vector = data->cfg.vector;
262 	for_each_cpu_and(cpu, data->domain, cpu_online_mask)
263 		per_cpu(vector_irq, cpu)[vector] = VECTOR_UNUSED;
264 
265 	data->cfg.vector = 0;
266 	cpumask_clear(data->domain);
267 
268 	/*
269 	 * If move is in progress or the old_domain mask is not empty,
270 	 * i.e. the cleanup IPI has not been processed yet, we need to remove
271 	 * the old references to desc from all cpus vector tables.
272 	 */
273 	if (!data->move_in_progress && cpumask_empty(data->old_domain))
274 		return;
275 
276 	desc = irq_to_desc(irq);
277 	for_each_cpu_and(cpu, data->old_domain, cpu_online_mask) {
278 		for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS;
279 		     vector++) {
280 			if (per_cpu(vector_irq, cpu)[vector] != desc)
281 				continue;
282 			per_cpu(vector_irq, cpu)[vector] = VECTOR_UNUSED;
283 			break;
284 		}
285 	}
286 	data->move_in_progress = 0;
287 }
288 
289 void init_irq_alloc_info(struct irq_alloc_info *info,
290 			 const struct cpumask *mask)
291 {
292 	memset(info, 0, sizeof(*info));
293 	info->mask = mask;
294 }
295 
296 void copy_irq_alloc_info(struct irq_alloc_info *dst, struct irq_alloc_info *src)
297 {
298 	if (src)
299 		*dst = *src;
300 	else
301 		memset(dst, 0, sizeof(*dst));
302 }
303 
304 static void x86_vector_free_irqs(struct irq_domain *domain,
305 				 unsigned int virq, unsigned int nr_irqs)
306 {
307 	struct apic_chip_data *apic_data;
308 	struct irq_data *irq_data;
309 	unsigned long flags;
310 	int i;
311 
312 	for (i = 0; i < nr_irqs; i++) {
313 		irq_data = irq_domain_get_irq_data(x86_vector_domain, virq + i);
314 		if (irq_data && irq_data->chip_data) {
315 			raw_spin_lock_irqsave(&vector_lock, flags);
316 			clear_irq_vector(virq + i, irq_data->chip_data);
317 			apic_data = irq_data->chip_data;
318 			irq_domain_reset_irq_data(irq_data);
319 			raw_spin_unlock_irqrestore(&vector_lock, flags);
320 			free_apic_chip_data(apic_data);
321 #ifdef	CONFIG_X86_IO_APIC
322 			if (virq + i < nr_legacy_irqs())
323 				legacy_irq_data[virq + i] = NULL;
324 #endif
325 		}
326 	}
327 }
328 
329 static int x86_vector_alloc_irqs(struct irq_domain *domain, unsigned int virq,
330 				 unsigned int nr_irqs, void *arg)
331 {
332 	struct irq_alloc_info *info = arg;
333 	struct apic_chip_data *data;
334 	struct irq_data *irq_data;
335 	int i, err, node;
336 
337 	if (disable_apic)
338 		return -ENXIO;
339 
340 	/* Currently vector allocator can't guarantee contiguous allocations */
341 	if ((info->flags & X86_IRQ_ALLOC_CONTIGUOUS_VECTORS) && nr_irqs > 1)
342 		return -ENOSYS;
343 
344 	for (i = 0; i < nr_irqs; i++) {
345 		irq_data = irq_domain_get_irq_data(domain, virq + i);
346 		BUG_ON(!irq_data);
347 		node = irq_data_get_node(irq_data);
348 #ifdef	CONFIG_X86_IO_APIC
349 		if (virq + i < nr_legacy_irqs() && legacy_irq_data[virq + i])
350 			data = legacy_irq_data[virq + i];
351 		else
352 #endif
353 			data = alloc_apic_chip_data(node);
354 		if (!data) {
355 			err = -ENOMEM;
356 			goto error;
357 		}
358 
359 		irq_data->chip = &lapic_controller;
360 		irq_data->chip_data = data;
361 		irq_data->hwirq = virq + i;
362 		err = assign_irq_vector_policy(virq + i, node, data, info);
363 		if (err)
364 			goto error;
365 	}
366 
367 	return 0;
368 
369 error:
370 	x86_vector_free_irqs(domain, virq, i + 1);
371 	return err;
372 }
373 
374 static const struct irq_domain_ops x86_vector_domain_ops = {
375 	.alloc	= x86_vector_alloc_irqs,
376 	.free	= x86_vector_free_irqs,
377 };
378 
379 int __init arch_probe_nr_irqs(void)
380 {
381 	int nr;
382 
383 	if (nr_irqs > (NR_VECTORS * nr_cpu_ids))
384 		nr_irqs = NR_VECTORS * nr_cpu_ids;
385 
386 	nr = (gsi_top + nr_legacy_irqs()) + 8 * nr_cpu_ids;
387 #if defined(CONFIG_PCI_MSI) || defined(CONFIG_HT_IRQ)
388 	/*
389 	 * for MSI and HT dyn irq
390 	 */
391 	if (gsi_top <= NR_IRQS_LEGACY)
392 		nr +=  8 * nr_cpu_ids;
393 	else
394 		nr += gsi_top * 16;
395 #endif
396 	if (nr < nr_irqs)
397 		nr_irqs = nr;
398 
399 	/*
400 	 * We don't know if PIC is present at this point so we need to do
401 	 * probe() to get the right number of legacy IRQs.
402 	 */
403 	return legacy_pic->probe();
404 }
405 
406 #ifdef	CONFIG_X86_IO_APIC
407 static void init_legacy_irqs(void)
408 {
409 	int i, node = cpu_to_node(0);
410 	struct apic_chip_data *data;
411 
412 	/*
413 	 * For legacy IRQ's, start with assigning irq0 to irq15 to
414 	 * ISA_IRQ_VECTOR(i) for all cpu's.
415 	 */
416 	for (i = 0; i < nr_legacy_irqs(); i++) {
417 		data = legacy_irq_data[i] = alloc_apic_chip_data(node);
418 		BUG_ON(!data);
419 
420 		data->cfg.vector = ISA_IRQ_VECTOR(i);
421 		cpumask_setall(data->domain);
422 		irq_set_chip_data(i, data);
423 	}
424 }
425 #else
426 static void init_legacy_irqs(void) { }
427 #endif
428 
429 int __init arch_early_irq_init(void)
430 {
431 	init_legacy_irqs();
432 
433 	x86_vector_domain = irq_domain_add_tree(NULL, &x86_vector_domain_ops,
434 						NULL);
435 	BUG_ON(x86_vector_domain == NULL);
436 	irq_set_default_host(x86_vector_domain);
437 
438 	arch_init_msi_domain(x86_vector_domain);
439 	arch_init_htirq_domain(x86_vector_domain);
440 
441 	BUG_ON(!alloc_cpumask_var(&vector_cpumask, GFP_KERNEL));
442 	BUG_ON(!alloc_cpumask_var(&vector_searchmask, GFP_KERNEL));
443 	BUG_ON(!alloc_cpumask_var(&searched_cpumask, GFP_KERNEL));
444 
445 	return arch_early_ioapic_init();
446 }
447 
448 /* Initialize vector_irq on a new cpu */
449 static void __setup_vector_irq(int cpu)
450 {
451 	struct apic_chip_data *data;
452 	struct irq_desc *desc;
453 	int irq, vector;
454 
455 	/* Mark the inuse vectors */
456 	for_each_irq_desc(irq, desc) {
457 		struct irq_data *idata = irq_desc_get_irq_data(desc);
458 
459 		data = apic_chip_data(idata);
460 		if (!data || !cpumask_test_cpu(cpu, data->domain))
461 			continue;
462 		vector = data->cfg.vector;
463 		per_cpu(vector_irq, cpu)[vector] = desc;
464 	}
465 	/* Mark the free vectors */
466 	for (vector = 0; vector < NR_VECTORS; ++vector) {
467 		desc = per_cpu(vector_irq, cpu)[vector];
468 		if (IS_ERR_OR_NULL(desc))
469 			continue;
470 
471 		data = apic_chip_data(irq_desc_get_irq_data(desc));
472 		if (!cpumask_test_cpu(cpu, data->domain))
473 			per_cpu(vector_irq, cpu)[vector] = VECTOR_UNUSED;
474 	}
475 }
476 
477 /*
478  * Setup the vector to irq mappings. Must be called with vector_lock held.
479  */
480 void setup_vector_irq(int cpu)
481 {
482 	int irq;
483 
484 	lockdep_assert_held(&vector_lock);
485 	/*
486 	 * On most of the platforms, legacy PIC delivers the interrupts on the
487 	 * boot cpu. But there are certain platforms where PIC interrupts are
488 	 * delivered to multiple cpu's. If the legacy IRQ is handled by the
489 	 * legacy PIC, for the new cpu that is coming online, setup the static
490 	 * legacy vector to irq mapping:
491 	 */
492 	for (irq = 0; irq < nr_legacy_irqs(); irq++)
493 		per_cpu(vector_irq, cpu)[ISA_IRQ_VECTOR(irq)] = irq_to_desc(irq);
494 
495 	__setup_vector_irq(cpu);
496 }
497 
498 static int apic_retrigger_irq(struct irq_data *irq_data)
499 {
500 	struct apic_chip_data *data = apic_chip_data(irq_data);
501 	unsigned long flags;
502 	int cpu;
503 
504 	raw_spin_lock_irqsave(&vector_lock, flags);
505 	cpu = cpumask_first_and(data->domain, cpu_online_mask);
506 	apic->send_IPI_mask(cpumask_of(cpu), data->cfg.vector);
507 	raw_spin_unlock_irqrestore(&vector_lock, flags);
508 
509 	return 1;
510 }
511 
512 void apic_ack_edge(struct irq_data *data)
513 {
514 	irq_complete_move(irqd_cfg(data));
515 	irq_move_irq(data);
516 	ack_APIC_irq();
517 }
518 
519 static int apic_set_affinity(struct irq_data *irq_data,
520 			     const struct cpumask *dest, bool force)
521 {
522 	struct apic_chip_data *data = irq_data->chip_data;
523 	int err, irq = irq_data->irq;
524 
525 	if (!config_enabled(CONFIG_SMP))
526 		return -EPERM;
527 
528 	if (!cpumask_intersects(dest, cpu_online_mask))
529 		return -EINVAL;
530 
531 	err = assign_irq_vector(irq, data, dest);
532 	return err ? err : IRQ_SET_MASK_OK;
533 }
534 
535 static struct irq_chip lapic_controller = {
536 	.irq_ack		= apic_ack_edge,
537 	.irq_set_affinity	= apic_set_affinity,
538 	.irq_retrigger		= apic_retrigger_irq,
539 };
540 
541 #ifdef CONFIG_SMP
542 static void __send_cleanup_vector(struct apic_chip_data *data)
543 {
544 	raw_spin_lock(&vector_lock);
545 	cpumask_and(data->old_domain, data->old_domain, cpu_online_mask);
546 	data->move_in_progress = 0;
547 	if (!cpumask_empty(data->old_domain))
548 		apic->send_IPI_mask(data->old_domain, IRQ_MOVE_CLEANUP_VECTOR);
549 	raw_spin_unlock(&vector_lock);
550 }
551 
552 void send_cleanup_vector(struct irq_cfg *cfg)
553 {
554 	struct apic_chip_data *data;
555 
556 	data = container_of(cfg, struct apic_chip_data, cfg);
557 	if (data->move_in_progress)
558 		__send_cleanup_vector(data);
559 }
560 
561 asmlinkage __visible void smp_irq_move_cleanup_interrupt(void)
562 {
563 	unsigned vector, me;
564 
565 	entering_ack_irq();
566 
567 	/* Prevent vectors vanishing under us */
568 	raw_spin_lock(&vector_lock);
569 
570 	me = smp_processor_id();
571 	for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
572 		struct apic_chip_data *data;
573 		struct irq_desc *desc;
574 		unsigned int irr;
575 
576 	retry:
577 		desc = __this_cpu_read(vector_irq[vector]);
578 		if (IS_ERR_OR_NULL(desc))
579 			continue;
580 
581 		if (!raw_spin_trylock(&desc->lock)) {
582 			raw_spin_unlock(&vector_lock);
583 			cpu_relax();
584 			raw_spin_lock(&vector_lock);
585 			goto retry;
586 		}
587 
588 		data = apic_chip_data(irq_desc_get_irq_data(desc));
589 		if (!data)
590 			goto unlock;
591 
592 		/*
593 		 * Nothing to cleanup if irq migration is in progress
594 		 * or this cpu is not set in the cleanup mask.
595 		 */
596 		if (data->move_in_progress ||
597 		    !cpumask_test_cpu(me, data->old_domain))
598 			goto unlock;
599 
600 		/*
601 		 * We have two cases to handle here:
602 		 * 1) vector is unchanged but the target mask got reduced
603 		 * 2) vector and the target mask has changed
604 		 *
605 		 * #1 is obvious, but in #2 we have two vectors with the same
606 		 * irq descriptor: the old and the new vector. So we need to
607 		 * make sure that we only cleanup the old vector. The new
608 		 * vector has the current @vector number in the config and
609 		 * this cpu is part of the target mask. We better leave that
610 		 * one alone.
611 		 */
612 		if (vector == data->cfg.vector &&
613 		    cpumask_test_cpu(me, data->domain))
614 			goto unlock;
615 
616 		irr = apic_read(APIC_IRR + (vector / 32 * 0x10));
617 		/*
618 		 * Check if the vector that needs to be cleanedup is
619 		 * registered at the cpu's IRR. If so, then this is not
620 		 * the best time to clean it up. Lets clean it up in the
621 		 * next attempt by sending another IRQ_MOVE_CLEANUP_VECTOR
622 		 * to myself.
623 		 */
624 		if (irr  & (1 << (vector % 32))) {
625 			apic->send_IPI_self(IRQ_MOVE_CLEANUP_VECTOR);
626 			goto unlock;
627 		}
628 		__this_cpu_write(vector_irq[vector], VECTOR_UNUSED);
629 		cpumask_clear_cpu(me, data->old_domain);
630 unlock:
631 		raw_spin_unlock(&desc->lock);
632 	}
633 
634 	raw_spin_unlock(&vector_lock);
635 
636 	exiting_irq();
637 }
638 
639 static void __irq_complete_move(struct irq_cfg *cfg, unsigned vector)
640 {
641 	unsigned me;
642 	struct apic_chip_data *data;
643 
644 	data = container_of(cfg, struct apic_chip_data, cfg);
645 	if (likely(!data->move_in_progress))
646 		return;
647 
648 	me = smp_processor_id();
649 	if (vector == data->cfg.vector && cpumask_test_cpu(me, data->domain))
650 		__send_cleanup_vector(data);
651 }
652 
653 void irq_complete_move(struct irq_cfg *cfg)
654 {
655 	__irq_complete_move(cfg, ~get_irq_regs()->orig_ax);
656 }
657 
658 /*
659  * Called from fixup_irqs() with @desc->lock held and interrupts disabled.
660  */
661 void irq_force_complete_move(struct irq_desc *desc)
662 {
663 	struct irq_data *irqdata = irq_desc_get_irq_data(desc);
664 	struct apic_chip_data *data = apic_chip_data(irqdata);
665 	struct irq_cfg *cfg = data ? &data->cfg : NULL;
666 	unsigned int cpu;
667 
668 	if (!cfg)
669 		return;
670 
671 	/*
672 	 * This is tricky. If the cleanup of @data->old_domain has not been
673 	 * done yet, then the following setaffinity call will fail with
674 	 * -EBUSY. This can leave the interrupt in a stale state.
675 	 *
676 	 * All CPUs are stuck in stop machine with interrupts disabled so
677 	 * calling __irq_complete_move() would be completely pointless.
678 	 */
679 	raw_spin_lock(&vector_lock);
680 	/*
681 	 * Clean out all offline cpus (including the outgoing one) from the
682 	 * old_domain mask.
683 	 */
684 	cpumask_and(data->old_domain, data->old_domain, cpu_online_mask);
685 
686 	/*
687 	 * If move_in_progress is cleared and the old_domain mask is empty,
688 	 * then there is nothing to cleanup. fixup_irqs() will take care of
689 	 * the stale vectors on the outgoing cpu.
690 	 */
691 	if (!data->move_in_progress && cpumask_empty(data->old_domain)) {
692 		raw_spin_unlock(&vector_lock);
693 		return;
694 	}
695 
696 	/*
697 	 * 1) The interrupt is in move_in_progress state. That means that we
698 	 *    have not seen an interrupt since the io_apic was reprogrammed to
699 	 *    the new vector.
700 	 *
701 	 * 2) The interrupt has fired on the new vector, but the cleanup IPIs
702 	 *    have not been processed yet.
703 	 */
704 	if (data->move_in_progress) {
705 		/*
706 		 * In theory there is a race:
707 		 *
708 		 * set_ioapic(new_vector) <-- Interrupt is raised before update
709 		 *			      is effective, i.e. it's raised on
710 		 *			      the old vector.
711 		 *
712 		 * So if the target cpu cannot handle that interrupt before
713 		 * the old vector is cleaned up, we get a spurious interrupt
714 		 * and in the worst case the ioapic irq line becomes stale.
715 		 *
716 		 * But in case of cpu hotplug this should be a non issue
717 		 * because if the affinity update happens right before all
718 		 * cpus rendevouz in stop machine, there is no way that the
719 		 * interrupt can be blocked on the target cpu because all cpus
720 		 * loops first with interrupts enabled in stop machine, so the
721 		 * old vector is not yet cleaned up when the interrupt fires.
722 		 *
723 		 * So the only way to run into this issue is if the delivery
724 		 * of the interrupt on the apic/system bus would be delayed
725 		 * beyond the point where the target cpu disables interrupts
726 		 * in stop machine. I doubt that it can happen, but at least
727 		 * there is a theroretical chance. Virtualization might be
728 		 * able to expose this, but AFAICT the IOAPIC emulation is not
729 		 * as stupid as the real hardware.
730 		 *
731 		 * Anyway, there is nothing we can do about that at this point
732 		 * w/o refactoring the whole fixup_irq() business completely.
733 		 * We print at least the irq number and the old vector number,
734 		 * so we have the necessary information when a problem in that
735 		 * area arises.
736 		 */
737 		pr_warn("IRQ fixup: irq %d move in progress, old vector %d\n",
738 			irqdata->irq, cfg->old_vector);
739 	}
740 	/*
741 	 * If old_domain is not empty, then other cpus still have the irq
742 	 * descriptor set in their vector array. Clean it up.
743 	 */
744 	for_each_cpu(cpu, data->old_domain)
745 		per_cpu(vector_irq, cpu)[cfg->old_vector] = VECTOR_UNUSED;
746 
747 	/* Cleanup the left overs of the (half finished) move */
748 	cpumask_clear(data->old_domain);
749 	data->move_in_progress = 0;
750 	raw_spin_unlock(&vector_lock);
751 }
752 #endif
753 
754 static void __init print_APIC_field(int base)
755 {
756 	int i;
757 
758 	printk(KERN_DEBUG);
759 
760 	for (i = 0; i < 8; i++)
761 		pr_cont("%08x", apic_read(base + i*0x10));
762 
763 	pr_cont("\n");
764 }
765 
766 static void __init print_local_APIC(void *dummy)
767 {
768 	unsigned int i, v, ver, maxlvt;
769 	u64 icr;
770 
771 	pr_debug("printing local APIC contents on CPU#%d/%d:\n",
772 		 smp_processor_id(), hard_smp_processor_id());
773 	v = apic_read(APIC_ID);
774 	pr_info("... APIC ID:      %08x (%01x)\n", v, read_apic_id());
775 	v = apic_read(APIC_LVR);
776 	pr_info("... APIC VERSION: %08x\n", v);
777 	ver = GET_APIC_VERSION(v);
778 	maxlvt = lapic_get_maxlvt();
779 
780 	v = apic_read(APIC_TASKPRI);
781 	pr_debug("... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
782 
783 	/* !82489DX */
784 	if (APIC_INTEGRATED(ver)) {
785 		if (!APIC_XAPIC(ver)) {
786 			v = apic_read(APIC_ARBPRI);
787 			pr_debug("... APIC ARBPRI: %08x (%02x)\n",
788 				 v, v & APIC_ARBPRI_MASK);
789 		}
790 		v = apic_read(APIC_PROCPRI);
791 		pr_debug("... APIC PROCPRI: %08x\n", v);
792 	}
793 
794 	/*
795 	 * Remote read supported only in the 82489DX and local APIC for
796 	 * Pentium processors.
797 	 */
798 	if (!APIC_INTEGRATED(ver) || maxlvt == 3) {
799 		v = apic_read(APIC_RRR);
800 		pr_debug("... APIC RRR: %08x\n", v);
801 	}
802 
803 	v = apic_read(APIC_LDR);
804 	pr_debug("... APIC LDR: %08x\n", v);
805 	if (!x2apic_enabled()) {
806 		v = apic_read(APIC_DFR);
807 		pr_debug("... APIC DFR: %08x\n", v);
808 	}
809 	v = apic_read(APIC_SPIV);
810 	pr_debug("... APIC SPIV: %08x\n", v);
811 
812 	pr_debug("... APIC ISR field:\n");
813 	print_APIC_field(APIC_ISR);
814 	pr_debug("... APIC TMR field:\n");
815 	print_APIC_field(APIC_TMR);
816 	pr_debug("... APIC IRR field:\n");
817 	print_APIC_field(APIC_IRR);
818 
819 	/* !82489DX */
820 	if (APIC_INTEGRATED(ver)) {
821 		/* Due to the Pentium erratum 3AP. */
822 		if (maxlvt > 3)
823 			apic_write(APIC_ESR, 0);
824 
825 		v = apic_read(APIC_ESR);
826 		pr_debug("... APIC ESR: %08x\n", v);
827 	}
828 
829 	icr = apic_icr_read();
830 	pr_debug("... APIC ICR: %08x\n", (u32)icr);
831 	pr_debug("... APIC ICR2: %08x\n", (u32)(icr >> 32));
832 
833 	v = apic_read(APIC_LVTT);
834 	pr_debug("... APIC LVTT: %08x\n", v);
835 
836 	if (maxlvt > 3) {
837 		/* PC is LVT#4. */
838 		v = apic_read(APIC_LVTPC);
839 		pr_debug("... APIC LVTPC: %08x\n", v);
840 	}
841 	v = apic_read(APIC_LVT0);
842 	pr_debug("... APIC LVT0: %08x\n", v);
843 	v = apic_read(APIC_LVT1);
844 	pr_debug("... APIC LVT1: %08x\n", v);
845 
846 	if (maxlvt > 2) {
847 		/* ERR is LVT#3. */
848 		v = apic_read(APIC_LVTERR);
849 		pr_debug("... APIC LVTERR: %08x\n", v);
850 	}
851 
852 	v = apic_read(APIC_TMICT);
853 	pr_debug("... APIC TMICT: %08x\n", v);
854 	v = apic_read(APIC_TMCCT);
855 	pr_debug("... APIC TMCCT: %08x\n", v);
856 	v = apic_read(APIC_TDCR);
857 	pr_debug("... APIC TDCR: %08x\n", v);
858 
859 	if (boot_cpu_has(X86_FEATURE_EXTAPIC)) {
860 		v = apic_read(APIC_EFEAT);
861 		maxlvt = (v >> 16) & 0xff;
862 		pr_debug("... APIC EFEAT: %08x\n", v);
863 		v = apic_read(APIC_ECTRL);
864 		pr_debug("... APIC ECTRL: %08x\n", v);
865 		for (i = 0; i < maxlvt; i++) {
866 			v = apic_read(APIC_EILVTn(i));
867 			pr_debug("... APIC EILVT%d: %08x\n", i, v);
868 		}
869 	}
870 	pr_cont("\n");
871 }
872 
873 static void __init print_local_APICs(int maxcpu)
874 {
875 	int cpu;
876 
877 	if (!maxcpu)
878 		return;
879 
880 	preempt_disable();
881 	for_each_online_cpu(cpu) {
882 		if (cpu >= maxcpu)
883 			break;
884 		smp_call_function_single(cpu, print_local_APIC, NULL, 1);
885 	}
886 	preempt_enable();
887 }
888 
889 static void __init print_PIC(void)
890 {
891 	unsigned int v;
892 	unsigned long flags;
893 
894 	if (!nr_legacy_irqs())
895 		return;
896 
897 	pr_debug("\nprinting PIC contents\n");
898 
899 	raw_spin_lock_irqsave(&i8259A_lock, flags);
900 
901 	v = inb(0xa1) << 8 | inb(0x21);
902 	pr_debug("... PIC  IMR: %04x\n", v);
903 
904 	v = inb(0xa0) << 8 | inb(0x20);
905 	pr_debug("... PIC  IRR: %04x\n", v);
906 
907 	outb(0x0b, 0xa0);
908 	outb(0x0b, 0x20);
909 	v = inb(0xa0) << 8 | inb(0x20);
910 	outb(0x0a, 0xa0);
911 	outb(0x0a, 0x20);
912 
913 	raw_spin_unlock_irqrestore(&i8259A_lock, flags);
914 
915 	pr_debug("... PIC  ISR: %04x\n", v);
916 
917 	v = inb(0x4d1) << 8 | inb(0x4d0);
918 	pr_debug("... PIC ELCR: %04x\n", v);
919 }
920 
921 static int show_lapic __initdata = 1;
922 static __init int setup_show_lapic(char *arg)
923 {
924 	int num = -1;
925 
926 	if (strcmp(arg, "all") == 0) {
927 		show_lapic = CONFIG_NR_CPUS;
928 	} else {
929 		get_option(&arg, &num);
930 		if (num >= 0)
931 			show_lapic = num;
932 	}
933 
934 	return 1;
935 }
936 __setup("show_lapic=", setup_show_lapic);
937 
938 static int __init print_ICs(void)
939 {
940 	if (apic_verbosity == APIC_QUIET)
941 		return 0;
942 
943 	print_PIC();
944 
945 	/* don't print out if apic is not there */
946 	if (!cpu_has_apic && !apic_from_smp_config())
947 		return 0;
948 
949 	print_local_APICs(show_lapic);
950 	print_IO_APICs();
951 
952 	return 0;
953 }
954 
955 late_initcall(print_ICs);
956