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