xref: /linux/arch/x86/kernel/apic/io_apic.c (revision f2ee442115c9b6219083c019939a9cc0c9abb2f8)
1 /*
2  *	Intel IO-APIC support for multi-Pentium hosts.
3  *
4  *	Copyright (C) 1997, 1998, 1999, 2000, 2009 Ingo Molnar, Hajnalka Szabo
5  *
6  *	Many thanks to Stig Venaas for trying out countless experimental
7  *	patches and reporting/debugging problems patiently!
8  *
9  *	(c) 1999, Multiple IO-APIC support, developed by
10  *	Ken-ichi Yaku <yaku@css1.kbnes.nec.co.jp> and
11  *      Hidemi Kishimoto <kisimoto@css1.kbnes.nec.co.jp>,
12  *	further tested and cleaned up by Zach Brown <zab@redhat.com>
13  *	and Ingo Molnar <mingo@redhat.com>
14  *
15  *	Fixes
16  *	Maciej W. Rozycki	:	Bits for genuine 82489DX APICs;
17  *					thanks to Eric Gilmore
18  *					and Rolf G. Tews
19  *					for testing these extensively
20  *	Paul Diefenbaugh	:	Added full ACPI support
21  */
22 
23 #include <linux/mm.h>
24 #include <linux/interrupt.h>
25 #include <linux/init.h>
26 #include <linux/delay.h>
27 #include <linux/sched.h>
28 #include <linux/pci.h>
29 #include <linux/mc146818rtc.h>
30 #include <linux/compiler.h>
31 #include <linux/acpi.h>
32 #include <linux/module.h>
33 #include <linux/syscore_ops.h>
34 #include <linux/msi.h>
35 #include <linux/htirq.h>
36 #include <linux/freezer.h>
37 #include <linux/kthread.h>
38 #include <linux/jiffies.h>	/* time_after() */
39 #include <linux/slab.h>
40 #ifdef CONFIG_ACPI
41 #include <acpi/acpi_bus.h>
42 #endif
43 #include <linux/bootmem.h>
44 #include <linux/dmar.h>
45 #include <linux/hpet.h>
46 
47 #include <asm/idle.h>
48 #include <asm/io.h>
49 #include <asm/smp.h>
50 #include <asm/cpu.h>
51 #include <asm/desc.h>
52 #include <asm/proto.h>
53 #include <asm/acpi.h>
54 #include <asm/dma.h>
55 #include <asm/timer.h>
56 #include <asm/i8259.h>
57 #include <asm/msidef.h>
58 #include <asm/hypertransport.h>
59 #include <asm/setup.h>
60 #include <asm/irq_remapping.h>
61 #include <asm/hpet.h>
62 #include <asm/hw_irq.h>
63 
64 #include <asm/apic.h>
65 
66 #define __apicdebuginit(type) static type __init
67 #define for_each_irq_pin(entry, head) \
68 	for (entry = head; entry; entry = entry->next)
69 
70 /*
71  *      Is the SiS APIC rmw bug present ?
72  *      -1 = don't know, 0 = no, 1 = yes
73  */
74 int sis_apic_bug = -1;
75 
76 static DEFINE_RAW_SPINLOCK(ioapic_lock);
77 static DEFINE_RAW_SPINLOCK(vector_lock);
78 
79 static struct ioapic {
80 	/*
81 	 * # of IRQ routing registers
82 	 */
83 	int nr_registers;
84 	/*
85 	 * Saved state during suspend/resume, or while enabling intr-remap.
86 	 */
87 	struct IO_APIC_route_entry *saved_registers;
88 	/* I/O APIC config */
89 	struct mpc_ioapic mp_config;
90 	/* IO APIC gsi routing info */
91 	struct mp_ioapic_gsi  gsi_config;
92 	DECLARE_BITMAP(pin_programmed, MP_MAX_IOAPIC_PIN + 1);
93 } ioapics[MAX_IO_APICS];
94 
95 #define mpc_ioapic_ver(ioapic_idx)	ioapics[ioapic_idx].mp_config.apicver
96 
97 int mpc_ioapic_id(int ioapic_idx)
98 {
99 	return ioapics[ioapic_idx].mp_config.apicid;
100 }
101 
102 unsigned int mpc_ioapic_addr(int ioapic_idx)
103 {
104 	return ioapics[ioapic_idx].mp_config.apicaddr;
105 }
106 
107 struct mp_ioapic_gsi *mp_ioapic_gsi_routing(int ioapic_idx)
108 {
109 	return &ioapics[ioapic_idx].gsi_config;
110 }
111 
112 int nr_ioapics;
113 
114 /* The one past the highest gsi number used */
115 u32 gsi_top;
116 
117 /* MP IRQ source entries */
118 struct mpc_intsrc mp_irqs[MAX_IRQ_SOURCES];
119 
120 /* # of MP IRQ source entries */
121 int mp_irq_entries;
122 
123 /* GSI interrupts */
124 static int nr_irqs_gsi = NR_IRQS_LEGACY;
125 
126 #if defined (CONFIG_MCA) || defined (CONFIG_EISA)
127 int mp_bus_id_to_type[MAX_MP_BUSSES];
128 #endif
129 
130 DECLARE_BITMAP(mp_bus_not_pci, MAX_MP_BUSSES);
131 
132 int skip_ioapic_setup;
133 
134 /**
135  * disable_ioapic_support() - disables ioapic support at runtime
136  */
137 void disable_ioapic_support(void)
138 {
139 #ifdef CONFIG_PCI
140 	noioapicquirk = 1;
141 	noioapicreroute = -1;
142 #endif
143 	skip_ioapic_setup = 1;
144 }
145 
146 static int __init parse_noapic(char *str)
147 {
148 	/* disable IO-APIC */
149 	disable_ioapic_support();
150 	return 0;
151 }
152 early_param("noapic", parse_noapic);
153 
154 static int io_apic_setup_irq_pin(unsigned int irq, int node,
155 				 struct io_apic_irq_attr *attr);
156 
157 /* Will be called in mpparse/acpi/sfi codes for saving IRQ info */
158 void mp_save_irq(struct mpc_intsrc *m)
159 {
160 	int i;
161 
162 	apic_printk(APIC_VERBOSE, "Int: type %d, pol %d, trig %d, bus %02x,"
163 		" IRQ %02x, APIC ID %x, APIC INT %02x\n",
164 		m->irqtype, m->irqflag & 3, (m->irqflag >> 2) & 3, m->srcbus,
165 		m->srcbusirq, m->dstapic, m->dstirq);
166 
167 	for (i = 0; i < mp_irq_entries; i++) {
168 		if (!memcmp(&mp_irqs[i], m, sizeof(*m)))
169 			return;
170 	}
171 
172 	memcpy(&mp_irqs[mp_irq_entries], m, sizeof(*m));
173 	if (++mp_irq_entries == MAX_IRQ_SOURCES)
174 		panic("Max # of irq sources exceeded!!\n");
175 }
176 
177 struct irq_pin_list {
178 	int apic, pin;
179 	struct irq_pin_list *next;
180 };
181 
182 static struct irq_pin_list *alloc_irq_pin_list(int node)
183 {
184 	return kzalloc_node(sizeof(struct irq_pin_list), GFP_KERNEL, node);
185 }
186 
187 
188 /* irq_cfg is indexed by the sum of all RTEs in all I/O APICs. */
189 static struct irq_cfg irq_cfgx[NR_IRQS_LEGACY];
190 
191 int __init arch_early_irq_init(void)
192 {
193 	struct irq_cfg *cfg;
194 	int count, node, i;
195 
196 	if (!legacy_pic->nr_legacy_irqs) {
197 		nr_irqs_gsi = 0;
198 		io_apic_irqs = ~0UL;
199 	}
200 
201 	for (i = 0; i < nr_ioapics; i++) {
202 		ioapics[i].saved_registers =
203 			kzalloc(sizeof(struct IO_APIC_route_entry) *
204 				ioapics[i].nr_registers, GFP_KERNEL);
205 		if (!ioapics[i].saved_registers)
206 			pr_err("IOAPIC %d: suspend/resume impossible!\n", i);
207 	}
208 
209 	cfg = irq_cfgx;
210 	count = ARRAY_SIZE(irq_cfgx);
211 	node = cpu_to_node(0);
212 
213 	/* Make sure the legacy interrupts are marked in the bitmap */
214 	irq_reserve_irqs(0, legacy_pic->nr_legacy_irqs);
215 
216 	for (i = 0; i < count; i++) {
217 		irq_set_chip_data(i, &cfg[i]);
218 		zalloc_cpumask_var_node(&cfg[i].domain, GFP_KERNEL, node);
219 		zalloc_cpumask_var_node(&cfg[i].old_domain, GFP_KERNEL, node);
220 		/*
221 		 * For legacy IRQ's, start with assigning irq0 to irq15 to
222 		 * IRQ0_VECTOR to IRQ15_VECTOR on cpu 0.
223 		 */
224 		if (i < legacy_pic->nr_legacy_irqs) {
225 			cfg[i].vector = IRQ0_VECTOR + i;
226 			cpumask_set_cpu(0, cfg[i].domain);
227 		}
228 	}
229 
230 	return 0;
231 }
232 
233 static struct irq_cfg *irq_cfg(unsigned int irq)
234 {
235 	return irq_get_chip_data(irq);
236 }
237 
238 static struct irq_cfg *alloc_irq_cfg(unsigned int irq, int node)
239 {
240 	struct irq_cfg *cfg;
241 
242 	cfg = kzalloc_node(sizeof(*cfg), GFP_KERNEL, node);
243 	if (!cfg)
244 		return NULL;
245 	if (!zalloc_cpumask_var_node(&cfg->domain, GFP_KERNEL, node))
246 		goto out_cfg;
247 	if (!zalloc_cpumask_var_node(&cfg->old_domain, GFP_KERNEL, node))
248 		goto out_domain;
249 	return cfg;
250 out_domain:
251 	free_cpumask_var(cfg->domain);
252 out_cfg:
253 	kfree(cfg);
254 	return NULL;
255 }
256 
257 static void free_irq_cfg(unsigned int at, struct irq_cfg *cfg)
258 {
259 	if (!cfg)
260 		return;
261 	irq_set_chip_data(at, NULL);
262 	free_cpumask_var(cfg->domain);
263 	free_cpumask_var(cfg->old_domain);
264 	kfree(cfg);
265 }
266 
267 static struct irq_cfg *alloc_irq_and_cfg_at(unsigned int at, int node)
268 {
269 	int res = irq_alloc_desc_at(at, node);
270 	struct irq_cfg *cfg;
271 
272 	if (res < 0) {
273 		if (res != -EEXIST)
274 			return NULL;
275 		cfg = irq_get_chip_data(at);
276 		if (cfg)
277 			return cfg;
278 	}
279 
280 	cfg = alloc_irq_cfg(at, node);
281 	if (cfg)
282 		irq_set_chip_data(at, cfg);
283 	else
284 		irq_free_desc(at);
285 	return cfg;
286 }
287 
288 static int alloc_irq_from(unsigned int from, int node)
289 {
290 	return irq_alloc_desc_from(from, node);
291 }
292 
293 static void free_irq_at(unsigned int at, struct irq_cfg *cfg)
294 {
295 	free_irq_cfg(at, cfg);
296 	irq_free_desc(at);
297 }
298 
299 struct io_apic {
300 	unsigned int index;
301 	unsigned int unused[3];
302 	unsigned int data;
303 	unsigned int unused2[11];
304 	unsigned int eoi;
305 };
306 
307 static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
308 {
309 	return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
310 		+ (mpc_ioapic_addr(idx) & ~PAGE_MASK);
311 }
312 
313 static inline void io_apic_eoi(unsigned int apic, unsigned int vector)
314 {
315 	struct io_apic __iomem *io_apic = io_apic_base(apic);
316 	writel(vector, &io_apic->eoi);
317 }
318 
319 static inline unsigned int io_apic_read(unsigned int apic, unsigned int reg)
320 {
321 	struct io_apic __iomem *io_apic = io_apic_base(apic);
322 	writel(reg, &io_apic->index);
323 	return readl(&io_apic->data);
324 }
325 
326 static inline void io_apic_write(unsigned int apic, unsigned int reg, unsigned int value)
327 {
328 	struct io_apic __iomem *io_apic = io_apic_base(apic);
329 	writel(reg, &io_apic->index);
330 	writel(value, &io_apic->data);
331 }
332 
333 /*
334  * Re-write a value: to be used for read-modify-write
335  * cycles where the read already set up the index register.
336  *
337  * Older SiS APIC requires we rewrite the index register
338  */
339 static inline void io_apic_modify(unsigned int apic, unsigned int reg, unsigned int value)
340 {
341 	struct io_apic __iomem *io_apic = io_apic_base(apic);
342 
343 	if (sis_apic_bug)
344 		writel(reg, &io_apic->index);
345 	writel(value, &io_apic->data);
346 }
347 
348 static bool io_apic_level_ack_pending(struct irq_cfg *cfg)
349 {
350 	struct irq_pin_list *entry;
351 	unsigned long flags;
352 
353 	raw_spin_lock_irqsave(&ioapic_lock, flags);
354 	for_each_irq_pin(entry, cfg->irq_2_pin) {
355 		unsigned int reg;
356 		int pin;
357 
358 		pin = entry->pin;
359 		reg = io_apic_read(entry->apic, 0x10 + pin*2);
360 		/* Is the remote IRR bit set? */
361 		if (reg & IO_APIC_REDIR_REMOTE_IRR) {
362 			raw_spin_unlock_irqrestore(&ioapic_lock, flags);
363 			return true;
364 		}
365 	}
366 	raw_spin_unlock_irqrestore(&ioapic_lock, flags);
367 
368 	return false;
369 }
370 
371 union entry_union {
372 	struct { u32 w1, w2; };
373 	struct IO_APIC_route_entry entry;
374 };
375 
376 static struct IO_APIC_route_entry __ioapic_read_entry(int apic, int pin)
377 {
378 	union entry_union eu;
379 
380 	eu.w1 = io_apic_read(apic, 0x10 + 2 * pin);
381 	eu.w2 = io_apic_read(apic, 0x11 + 2 * pin);
382 	return eu.entry;
383 }
384 
385 static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
386 {
387 	union entry_union eu;
388 	unsigned long flags;
389 	raw_spin_lock_irqsave(&ioapic_lock, flags);
390 	eu.entry = __ioapic_read_entry(apic, pin);
391 	raw_spin_unlock_irqrestore(&ioapic_lock, flags);
392 	return eu.entry;
393 }
394 
395 /*
396  * When we write a new IO APIC routing entry, we need to write the high
397  * word first! If the mask bit in the low word is clear, we will enable
398  * the interrupt, and we need to make sure the entry is fully populated
399  * before that happens.
400  */
401 static void
402 __ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
403 {
404 	union entry_union eu = {{0, 0}};
405 
406 	eu.entry = e;
407 	io_apic_write(apic, 0x11 + 2*pin, eu.w2);
408 	io_apic_write(apic, 0x10 + 2*pin, eu.w1);
409 }
410 
411 static void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
412 {
413 	unsigned long flags;
414 	raw_spin_lock_irqsave(&ioapic_lock, flags);
415 	__ioapic_write_entry(apic, pin, e);
416 	raw_spin_unlock_irqrestore(&ioapic_lock, flags);
417 }
418 
419 /*
420  * When we mask an IO APIC routing entry, we need to write the low
421  * word first, in order to set the mask bit before we change the
422  * high bits!
423  */
424 static void ioapic_mask_entry(int apic, int pin)
425 {
426 	unsigned long flags;
427 	union entry_union eu = { .entry.mask = 1 };
428 
429 	raw_spin_lock_irqsave(&ioapic_lock, flags);
430 	io_apic_write(apic, 0x10 + 2*pin, eu.w1);
431 	io_apic_write(apic, 0x11 + 2*pin, eu.w2);
432 	raw_spin_unlock_irqrestore(&ioapic_lock, flags);
433 }
434 
435 /*
436  * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
437  * shared ISA-space IRQs, so we have to support them. We are super
438  * fast in the common case, and fast for shared ISA-space IRQs.
439  */
440 static int
441 __add_pin_to_irq_node(struct irq_cfg *cfg, int node, int apic, int pin)
442 {
443 	struct irq_pin_list **last, *entry;
444 
445 	/* don't allow duplicates */
446 	last = &cfg->irq_2_pin;
447 	for_each_irq_pin(entry, cfg->irq_2_pin) {
448 		if (entry->apic == apic && entry->pin == pin)
449 			return 0;
450 		last = &entry->next;
451 	}
452 
453 	entry = alloc_irq_pin_list(node);
454 	if (!entry) {
455 		printk(KERN_ERR "can not alloc irq_pin_list (%d,%d,%d)\n",
456 				node, apic, pin);
457 		return -ENOMEM;
458 	}
459 	entry->apic = apic;
460 	entry->pin = pin;
461 
462 	*last = entry;
463 	return 0;
464 }
465 
466 static void add_pin_to_irq_node(struct irq_cfg *cfg, int node, int apic, int pin)
467 {
468 	if (__add_pin_to_irq_node(cfg, node, apic, pin))
469 		panic("IO-APIC: failed to add irq-pin. Can not proceed\n");
470 }
471 
472 /*
473  * Reroute an IRQ to a different pin.
474  */
475 static void __init replace_pin_at_irq_node(struct irq_cfg *cfg, int node,
476 					   int oldapic, int oldpin,
477 					   int newapic, int newpin)
478 {
479 	struct irq_pin_list *entry;
480 
481 	for_each_irq_pin(entry, cfg->irq_2_pin) {
482 		if (entry->apic == oldapic && entry->pin == oldpin) {
483 			entry->apic = newapic;
484 			entry->pin = newpin;
485 			/* every one is different, right? */
486 			return;
487 		}
488 	}
489 
490 	/* old apic/pin didn't exist, so just add new ones */
491 	add_pin_to_irq_node(cfg, node, newapic, newpin);
492 }
493 
494 static void __io_apic_modify_irq(struct irq_pin_list *entry,
495 				 int mask_and, int mask_or,
496 				 void (*final)(struct irq_pin_list *entry))
497 {
498 	unsigned int reg, pin;
499 
500 	pin = entry->pin;
501 	reg = io_apic_read(entry->apic, 0x10 + pin * 2);
502 	reg &= mask_and;
503 	reg |= mask_or;
504 	io_apic_modify(entry->apic, 0x10 + pin * 2, reg);
505 	if (final)
506 		final(entry);
507 }
508 
509 static void io_apic_modify_irq(struct irq_cfg *cfg,
510 			       int mask_and, int mask_or,
511 			       void (*final)(struct irq_pin_list *entry))
512 {
513 	struct irq_pin_list *entry;
514 
515 	for_each_irq_pin(entry, cfg->irq_2_pin)
516 		__io_apic_modify_irq(entry, mask_and, mask_or, final);
517 }
518 
519 static void io_apic_sync(struct irq_pin_list *entry)
520 {
521 	/*
522 	 * Synchronize the IO-APIC and the CPU by doing
523 	 * a dummy read from the IO-APIC
524 	 */
525 	struct io_apic __iomem *io_apic;
526 	io_apic = io_apic_base(entry->apic);
527 	readl(&io_apic->data);
528 }
529 
530 static void mask_ioapic(struct irq_cfg *cfg)
531 {
532 	unsigned long flags;
533 
534 	raw_spin_lock_irqsave(&ioapic_lock, flags);
535 	io_apic_modify_irq(cfg, ~0, IO_APIC_REDIR_MASKED, &io_apic_sync);
536 	raw_spin_unlock_irqrestore(&ioapic_lock, flags);
537 }
538 
539 static void mask_ioapic_irq(struct irq_data *data)
540 {
541 	mask_ioapic(data->chip_data);
542 }
543 
544 static void __unmask_ioapic(struct irq_cfg *cfg)
545 {
546 	io_apic_modify_irq(cfg, ~IO_APIC_REDIR_MASKED, 0, NULL);
547 }
548 
549 static void unmask_ioapic(struct irq_cfg *cfg)
550 {
551 	unsigned long flags;
552 
553 	raw_spin_lock_irqsave(&ioapic_lock, flags);
554 	__unmask_ioapic(cfg);
555 	raw_spin_unlock_irqrestore(&ioapic_lock, flags);
556 }
557 
558 static void unmask_ioapic_irq(struct irq_data *data)
559 {
560 	unmask_ioapic(data->chip_data);
561 }
562 
563 /*
564  * IO-APIC versions below 0x20 don't support EOI register.
565  * For the record, here is the information about various versions:
566  *     0Xh     82489DX
567  *     1Xh     I/OAPIC or I/O(x)APIC which are not PCI 2.2 Compliant
568  *     2Xh     I/O(x)APIC which is PCI 2.2 Compliant
569  *     30h-FFh Reserved
570  *
571  * Some of the Intel ICH Specs (ICH2 to ICH5) documents the io-apic
572  * version as 0x2. This is an error with documentation and these ICH chips
573  * use io-apic's of version 0x20.
574  *
575  * For IO-APIC's with EOI register, we use that to do an explicit EOI.
576  * Otherwise, we simulate the EOI message manually by changing the trigger
577  * mode to edge and then back to level, with RTE being masked during this.
578  */
579 static void __eoi_ioapic_pin(int apic, int pin, int vector, struct irq_cfg *cfg)
580 {
581 	if (mpc_ioapic_ver(apic) >= 0x20) {
582 		/*
583 		 * Intr-remapping uses pin number as the virtual vector
584 		 * in the RTE. Actual vector is programmed in
585 		 * intr-remapping table entry. Hence for the io-apic
586 		 * EOI we use the pin number.
587 		 */
588 		if (cfg && irq_remapped(cfg))
589 			io_apic_eoi(apic, pin);
590 		else
591 			io_apic_eoi(apic, vector);
592 	} else {
593 		struct IO_APIC_route_entry entry, entry1;
594 
595 		entry = entry1 = __ioapic_read_entry(apic, pin);
596 
597 		/*
598 		 * Mask the entry and change the trigger mode to edge.
599 		 */
600 		entry1.mask = 1;
601 		entry1.trigger = IOAPIC_EDGE;
602 
603 		__ioapic_write_entry(apic, pin, entry1);
604 
605 		/*
606 		 * Restore the previous level triggered entry.
607 		 */
608 		__ioapic_write_entry(apic, pin, entry);
609 	}
610 }
611 
612 static void eoi_ioapic_irq(unsigned int irq, struct irq_cfg *cfg)
613 {
614 	struct irq_pin_list *entry;
615 	unsigned long flags;
616 
617 	raw_spin_lock_irqsave(&ioapic_lock, flags);
618 	for_each_irq_pin(entry, cfg->irq_2_pin)
619 		__eoi_ioapic_pin(entry->apic, entry->pin, cfg->vector, cfg);
620 	raw_spin_unlock_irqrestore(&ioapic_lock, flags);
621 }
622 
623 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
624 {
625 	struct IO_APIC_route_entry entry;
626 
627 	/* Check delivery_mode to be sure we're not clearing an SMI pin */
628 	entry = ioapic_read_entry(apic, pin);
629 	if (entry.delivery_mode == dest_SMI)
630 		return;
631 
632 	/*
633 	 * Make sure the entry is masked and re-read the contents to check
634 	 * if it is a level triggered pin and if the remote-IRR is set.
635 	 */
636 	if (!entry.mask) {
637 		entry.mask = 1;
638 		ioapic_write_entry(apic, pin, entry);
639 		entry = ioapic_read_entry(apic, pin);
640 	}
641 
642 	if (entry.irr) {
643 		unsigned long flags;
644 
645 		/*
646 		 * Make sure the trigger mode is set to level. Explicit EOI
647 		 * doesn't clear the remote-IRR if the trigger mode is not
648 		 * set to level.
649 		 */
650 		if (!entry.trigger) {
651 			entry.trigger = IOAPIC_LEVEL;
652 			ioapic_write_entry(apic, pin, entry);
653 		}
654 
655 		raw_spin_lock_irqsave(&ioapic_lock, flags);
656 		__eoi_ioapic_pin(apic, pin, entry.vector, NULL);
657 		raw_spin_unlock_irqrestore(&ioapic_lock, flags);
658 	}
659 
660 	/*
661 	 * Clear the rest of the bits in the IO-APIC RTE except for the mask
662 	 * bit.
663 	 */
664 	ioapic_mask_entry(apic, pin);
665 	entry = ioapic_read_entry(apic, pin);
666 	if (entry.irr)
667 		printk(KERN_ERR "Unable to reset IRR for apic: %d, pin :%d\n",
668 		       mpc_ioapic_id(apic), pin);
669 }
670 
671 static void clear_IO_APIC (void)
672 {
673 	int apic, pin;
674 
675 	for (apic = 0; apic < nr_ioapics; apic++)
676 		for (pin = 0; pin < ioapics[apic].nr_registers; pin++)
677 			clear_IO_APIC_pin(apic, pin);
678 }
679 
680 #ifdef CONFIG_X86_32
681 /*
682  * support for broken MP BIOSs, enables hand-redirection of PIRQ0-7 to
683  * specific CPU-side IRQs.
684  */
685 
686 #define MAX_PIRQS 8
687 static int pirq_entries[MAX_PIRQS] = {
688 	[0 ... MAX_PIRQS - 1] = -1
689 };
690 
691 static int __init ioapic_pirq_setup(char *str)
692 {
693 	int i, max;
694 	int ints[MAX_PIRQS+1];
695 
696 	get_options(str, ARRAY_SIZE(ints), ints);
697 
698 	apic_printk(APIC_VERBOSE, KERN_INFO
699 			"PIRQ redirection, working around broken MP-BIOS.\n");
700 	max = MAX_PIRQS;
701 	if (ints[0] < MAX_PIRQS)
702 		max = ints[0];
703 
704 	for (i = 0; i < max; i++) {
705 		apic_printk(APIC_VERBOSE, KERN_DEBUG
706 				"... PIRQ%d -> IRQ %d\n", i, ints[i+1]);
707 		/*
708 		 * PIRQs are mapped upside down, usually.
709 		 */
710 		pirq_entries[MAX_PIRQS-i-1] = ints[i+1];
711 	}
712 	return 1;
713 }
714 
715 __setup("pirq=", ioapic_pirq_setup);
716 #endif /* CONFIG_X86_32 */
717 
718 /*
719  * Saves all the IO-APIC RTE's
720  */
721 int save_ioapic_entries(void)
722 {
723 	int apic, pin;
724 	int err = 0;
725 
726 	for (apic = 0; apic < nr_ioapics; apic++) {
727 		if (!ioapics[apic].saved_registers) {
728 			err = -ENOMEM;
729 			continue;
730 		}
731 
732 		for (pin = 0; pin < ioapics[apic].nr_registers; pin++)
733 			ioapics[apic].saved_registers[pin] =
734 				ioapic_read_entry(apic, pin);
735 	}
736 
737 	return err;
738 }
739 
740 /*
741  * Mask all IO APIC entries.
742  */
743 void mask_ioapic_entries(void)
744 {
745 	int apic, pin;
746 
747 	for (apic = 0; apic < nr_ioapics; apic++) {
748 		if (!ioapics[apic].saved_registers)
749 			continue;
750 
751 		for (pin = 0; pin < ioapics[apic].nr_registers; pin++) {
752 			struct IO_APIC_route_entry entry;
753 
754 			entry = ioapics[apic].saved_registers[pin];
755 			if (!entry.mask) {
756 				entry.mask = 1;
757 				ioapic_write_entry(apic, pin, entry);
758 			}
759 		}
760 	}
761 }
762 
763 /*
764  * Restore IO APIC entries which was saved in the ioapic structure.
765  */
766 int restore_ioapic_entries(void)
767 {
768 	int apic, pin;
769 
770 	for (apic = 0; apic < nr_ioapics; apic++) {
771 		if (!ioapics[apic].saved_registers)
772 			continue;
773 
774 		for (pin = 0; pin < ioapics[apic].nr_registers; pin++)
775 			ioapic_write_entry(apic, pin,
776 					   ioapics[apic].saved_registers[pin]);
777 	}
778 	return 0;
779 }
780 
781 /*
782  * Find the IRQ entry number of a certain pin.
783  */
784 static int find_irq_entry(int ioapic_idx, int pin, int type)
785 {
786 	int i;
787 
788 	for (i = 0; i < mp_irq_entries; i++)
789 		if (mp_irqs[i].irqtype == type &&
790 		    (mp_irqs[i].dstapic == mpc_ioapic_id(ioapic_idx) ||
791 		     mp_irqs[i].dstapic == MP_APIC_ALL) &&
792 		    mp_irqs[i].dstirq == pin)
793 			return i;
794 
795 	return -1;
796 }
797 
798 /*
799  * Find the pin to which IRQ[irq] (ISA) is connected
800  */
801 static int __init find_isa_irq_pin(int irq, int type)
802 {
803 	int i;
804 
805 	for (i = 0; i < mp_irq_entries; i++) {
806 		int lbus = mp_irqs[i].srcbus;
807 
808 		if (test_bit(lbus, mp_bus_not_pci) &&
809 		    (mp_irqs[i].irqtype == type) &&
810 		    (mp_irqs[i].srcbusirq == irq))
811 
812 			return mp_irqs[i].dstirq;
813 	}
814 	return -1;
815 }
816 
817 static int __init find_isa_irq_apic(int irq, int type)
818 {
819 	int i;
820 
821 	for (i = 0; i < mp_irq_entries; i++) {
822 		int lbus = mp_irqs[i].srcbus;
823 
824 		if (test_bit(lbus, mp_bus_not_pci) &&
825 		    (mp_irqs[i].irqtype == type) &&
826 		    (mp_irqs[i].srcbusirq == irq))
827 			break;
828 	}
829 
830 	if (i < mp_irq_entries) {
831 		int ioapic_idx;
832 
833 		for (ioapic_idx = 0; ioapic_idx < nr_ioapics; ioapic_idx++)
834 			if (mpc_ioapic_id(ioapic_idx) == mp_irqs[i].dstapic)
835 				return ioapic_idx;
836 	}
837 
838 	return -1;
839 }
840 
841 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
842 /*
843  * EISA Edge/Level control register, ELCR
844  */
845 static int EISA_ELCR(unsigned int irq)
846 {
847 	if (irq < legacy_pic->nr_legacy_irqs) {
848 		unsigned int port = 0x4d0 + (irq >> 3);
849 		return (inb(port) >> (irq & 7)) & 1;
850 	}
851 	apic_printk(APIC_VERBOSE, KERN_INFO
852 			"Broken MPtable reports ISA irq %d\n", irq);
853 	return 0;
854 }
855 
856 #endif
857 
858 /* ISA interrupts are always polarity zero edge triggered,
859  * when listed as conforming in the MP table. */
860 
861 #define default_ISA_trigger(idx)	(0)
862 #define default_ISA_polarity(idx)	(0)
863 
864 /* EISA interrupts are always polarity zero and can be edge or level
865  * trigger depending on the ELCR value.  If an interrupt is listed as
866  * EISA conforming in the MP table, that means its trigger type must
867  * be read in from the ELCR */
868 
869 #define default_EISA_trigger(idx)	(EISA_ELCR(mp_irqs[idx].srcbusirq))
870 #define default_EISA_polarity(idx)	default_ISA_polarity(idx)
871 
872 /* PCI interrupts are always polarity one level triggered,
873  * when listed as conforming in the MP table. */
874 
875 #define default_PCI_trigger(idx)	(1)
876 #define default_PCI_polarity(idx)	(1)
877 
878 /* MCA interrupts are always polarity zero level triggered,
879  * when listed as conforming in the MP table. */
880 
881 #define default_MCA_trigger(idx)	(1)
882 #define default_MCA_polarity(idx)	default_ISA_polarity(idx)
883 
884 static int irq_polarity(int idx)
885 {
886 	int bus = mp_irqs[idx].srcbus;
887 	int polarity;
888 
889 	/*
890 	 * Determine IRQ line polarity (high active or low active):
891 	 */
892 	switch (mp_irqs[idx].irqflag & 3)
893 	{
894 		case 0: /* conforms, ie. bus-type dependent polarity */
895 			if (test_bit(bus, mp_bus_not_pci))
896 				polarity = default_ISA_polarity(idx);
897 			else
898 				polarity = default_PCI_polarity(idx);
899 			break;
900 		case 1: /* high active */
901 		{
902 			polarity = 0;
903 			break;
904 		}
905 		case 2: /* reserved */
906 		{
907 			printk(KERN_WARNING "broken BIOS!!\n");
908 			polarity = 1;
909 			break;
910 		}
911 		case 3: /* low active */
912 		{
913 			polarity = 1;
914 			break;
915 		}
916 		default: /* invalid */
917 		{
918 			printk(KERN_WARNING "broken BIOS!!\n");
919 			polarity = 1;
920 			break;
921 		}
922 	}
923 	return polarity;
924 }
925 
926 static int irq_trigger(int idx)
927 {
928 	int bus = mp_irqs[idx].srcbus;
929 	int trigger;
930 
931 	/*
932 	 * Determine IRQ trigger mode (edge or level sensitive):
933 	 */
934 	switch ((mp_irqs[idx].irqflag>>2) & 3)
935 	{
936 		case 0: /* conforms, ie. bus-type dependent */
937 			if (test_bit(bus, mp_bus_not_pci))
938 				trigger = default_ISA_trigger(idx);
939 			else
940 				trigger = default_PCI_trigger(idx);
941 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
942 			switch (mp_bus_id_to_type[bus]) {
943 				case MP_BUS_ISA: /* ISA pin */
944 				{
945 					/* set before the switch */
946 					break;
947 				}
948 				case MP_BUS_EISA: /* EISA pin */
949 				{
950 					trigger = default_EISA_trigger(idx);
951 					break;
952 				}
953 				case MP_BUS_PCI: /* PCI pin */
954 				{
955 					/* set before the switch */
956 					break;
957 				}
958 				case MP_BUS_MCA: /* MCA pin */
959 				{
960 					trigger = default_MCA_trigger(idx);
961 					break;
962 				}
963 				default:
964 				{
965 					printk(KERN_WARNING "broken BIOS!!\n");
966 					trigger = 1;
967 					break;
968 				}
969 			}
970 #endif
971 			break;
972 		case 1: /* edge */
973 		{
974 			trigger = 0;
975 			break;
976 		}
977 		case 2: /* reserved */
978 		{
979 			printk(KERN_WARNING "broken BIOS!!\n");
980 			trigger = 1;
981 			break;
982 		}
983 		case 3: /* level */
984 		{
985 			trigger = 1;
986 			break;
987 		}
988 		default: /* invalid */
989 		{
990 			printk(KERN_WARNING "broken BIOS!!\n");
991 			trigger = 0;
992 			break;
993 		}
994 	}
995 	return trigger;
996 }
997 
998 static int pin_2_irq(int idx, int apic, int pin)
999 {
1000 	int irq;
1001 	int bus = mp_irqs[idx].srcbus;
1002 	struct mp_ioapic_gsi *gsi_cfg = mp_ioapic_gsi_routing(apic);
1003 
1004 	/*
1005 	 * Debugging check, we are in big trouble if this message pops up!
1006 	 */
1007 	if (mp_irqs[idx].dstirq != pin)
1008 		printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
1009 
1010 	if (test_bit(bus, mp_bus_not_pci)) {
1011 		irq = mp_irqs[idx].srcbusirq;
1012 	} else {
1013 		u32 gsi = gsi_cfg->gsi_base + pin;
1014 
1015 		if (gsi >= NR_IRQS_LEGACY)
1016 			irq = gsi;
1017 		else
1018 			irq = gsi_top + gsi;
1019 	}
1020 
1021 #ifdef CONFIG_X86_32
1022 	/*
1023 	 * PCI IRQ command line redirection. Yes, limits are hardcoded.
1024 	 */
1025 	if ((pin >= 16) && (pin <= 23)) {
1026 		if (pirq_entries[pin-16] != -1) {
1027 			if (!pirq_entries[pin-16]) {
1028 				apic_printk(APIC_VERBOSE, KERN_DEBUG
1029 						"disabling PIRQ%d\n", pin-16);
1030 			} else {
1031 				irq = pirq_entries[pin-16];
1032 				apic_printk(APIC_VERBOSE, KERN_DEBUG
1033 						"using PIRQ%d -> IRQ %d\n",
1034 						pin-16, irq);
1035 			}
1036 		}
1037 	}
1038 #endif
1039 
1040 	return irq;
1041 }
1042 
1043 /*
1044  * Find a specific PCI IRQ entry.
1045  * Not an __init, possibly needed by modules
1046  */
1047 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin,
1048 				struct io_apic_irq_attr *irq_attr)
1049 {
1050 	int ioapic_idx, i, best_guess = -1;
1051 
1052 	apic_printk(APIC_DEBUG,
1053 		    "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
1054 		    bus, slot, pin);
1055 	if (test_bit(bus, mp_bus_not_pci)) {
1056 		apic_printk(APIC_VERBOSE,
1057 			    "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
1058 		return -1;
1059 	}
1060 	for (i = 0; i < mp_irq_entries; i++) {
1061 		int lbus = mp_irqs[i].srcbus;
1062 
1063 		for (ioapic_idx = 0; ioapic_idx < nr_ioapics; ioapic_idx++)
1064 			if (mpc_ioapic_id(ioapic_idx) == mp_irqs[i].dstapic ||
1065 			    mp_irqs[i].dstapic == MP_APIC_ALL)
1066 				break;
1067 
1068 		if (!test_bit(lbus, mp_bus_not_pci) &&
1069 		    !mp_irqs[i].irqtype &&
1070 		    (bus == lbus) &&
1071 		    (slot == ((mp_irqs[i].srcbusirq >> 2) & 0x1f))) {
1072 			int irq = pin_2_irq(i, ioapic_idx, mp_irqs[i].dstirq);
1073 
1074 			if (!(ioapic_idx || IO_APIC_IRQ(irq)))
1075 				continue;
1076 
1077 			if (pin == (mp_irqs[i].srcbusirq & 3)) {
1078 				set_io_apic_irq_attr(irq_attr, ioapic_idx,
1079 						     mp_irqs[i].dstirq,
1080 						     irq_trigger(i),
1081 						     irq_polarity(i));
1082 				return irq;
1083 			}
1084 			/*
1085 			 * Use the first all-but-pin matching entry as a
1086 			 * best-guess fuzzy result for broken mptables.
1087 			 */
1088 			if (best_guess < 0) {
1089 				set_io_apic_irq_attr(irq_attr, ioapic_idx,
1090 						     mp_irqs[i].dstirq,
1091 						     irq_trigger(i),
1092 						     irq_polarity(i));
1093 				best_guess = irq;
1094 			}
1095 		}
1096 	}
1097 	return best_guess;
1098 }
1099 EXPORT_SYMBOL(IO_APIC_get_PCI_irq_vector);
1100 
1101 void lock_vector_lock(void)
1102 {
1103 	/* Used to the online set of cpus does not change
1104 	 * during assign_irq_vector.
1105 	 */
1106 	raw_spin_lock(&vector_lock);
1107 }
1108 
1109 void unlock_vector_lock(void)
1110 {
1111 	raw_spin_unlock(&vector_lock);
1112 }
1113 
1114 static int
1115 __assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask)
1116 {
1117 	/*
1118 	 * NOTE! The local APIC isn't very good at handling
1119 	 * multiple interrupts at the same interrupt level.
1120 	 * As the interrupt level is determined by taking the
1121 	 * vector number and shifting that right by 4, we
1122 	 * want to spread these out a bit so that they don't
1123 	 * all fall in the same interrupt level.
1124 	 *
1125 	 * Also, we've got to be careful not to trash gate
1126 	 * 0x80, because int 0x80 is hm, kind of importantish. ;)
1127 	 */
1128 	static int current_vector = FIRST_EXTERNAL_VECTOR + VECTOR_OFFSET_START;
1129 	static int current_offset = VECTOR_OFFSET_START % 8;
1130 	unsigned int old_vector;
1131 	int cpu, err;
1132 	cpumask_var_t tmp_mask;
1133 
1134 	if (cfg->move_in_progress)
1135 		return -EBUSY;
1136 
1137 	if (!alloc_cpumask_var(&tmp_mask, GFP_ATOMIC))
1138 		return -ENOMEM;
1139 
1140 	old_vector = cfg->vector;
1141 	if (old_vector) {
1142 		cpumask_and(tmp_mask, mask, cpu_online_mask);
1143 		cpumask_and(tmp_mask, cfg->domain, tmp_mask);
1144 		if (!cpumask_empty(tmp_mask)) {
1145 			free_cpumask_var(tmp_mask);
1146 			return 0;
1147 		}
1148 	}
1149 
1150 	/* Only try and allocate irqs on cpus that are present */
1151 	err = -ENOSPC;
1152 	for_each_cpu_and(cpu, mask, cpu_online_mask) {
1153 		int new_cpu;
1154 		int vector, offset;
1155 
1156 		apic->vector_allocation_domain(cpu, tmp_mask);
1157 
1158 		vector = current_vector;
1159 		offset = current_offset;
1160 next:
1161 		vector += 8;
1162 		if (vector >= first_system_vector) {
1163 			/* If out of vectors on large boxen, must share them. */
1164 			offset = (offset + 1) % 8;
1165 			vector = FIRST_EXTERNAL_VECTOR + offset;
1166 		}
1167 		if (unlikely(current_vector == vector))
1168 			continue;
1169 
1170 		if (test_bit(vector, used_vectors))
1171 			goto next;
1172 
1173 		for_each_cpu_and(new_cpu, tmp_mask, cpu_online_mask)
1174 			if (per_cpu(vector_irq, new_cpu)[vector] != -1)
1175 				goto next;
1176 		/* Found one! */
1177 		current_vector = vector;
1178 		current_offset = offset;
1179 		if (old_vector) {
1180 			cfg->move_in_progress = 1;
1181 			cpumask_copy(cfg->old_domain, cfg->domain);
1182 		}
1183 		for_each_cpu_and(new_cpu, tmp_mask, cpu_online_mask)
1184 			per_cpu(vector_irq, new_cpu)[vector] = irq;
1185 		cfg->vector = vector;
1186 		cpumask_copy(cfg->domain, tmp_mask);
1187 		err = 0;
1188 		break;
1189 	}
1190 	free_cpumask_var(tmp_mask);
1191 	return err;
1192 }
1193 
1194 int assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask)
1195 {
1196 	int err;
1197 	unsigned long flags;
1198 
1199 	raw_spin_lock_irqsave(&vector_lock, flags);
1200 	err = __assign_irq_vector(irq, cfg, mask);
1201 	raw_spin_unlock_irqrestore(&vector_lock, flags);
1202 	return err;
1203 }
1204 
1205 static void __clear_irq_vector(int irq, struct irq_cfg *cfg)
1206 {
1207 	int cpu, vector;
1208 
1209 	BUG_ON(!cfg->vector);
1210 
1211 	vector = cfg->vector;
1212 	for_each_cpu_and(cpu, cfg->domain, cpu_online_mask)
1213 		per_cpu(vector_irq, cpu)[vector] = -1;
1214 
1215 	cfg->vector = 0;
1216 	cpumask_clear(cfg->domain);
1217 
1218 	if (likely(!cfg->move_in_progress))
1219 		return;
1220 	for_each_cpu_and(cpu, cfg->old_domain, cpu_online_mask) {
1221 		for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS;
1222 								vector++) {
1223 			if (per_cpu(vector_irq, cpu)[vector] != irq)
1224 				continue;
1225 			per_cpu(vector_irq, cpu)[vector] = -1;
1226 			break;
1227 		}
1228 	}
1229 	cfg->move_in_progress = 0;
1230 }
1231 
1232 void __setup_vector_irq(int cpu)
1233 {
1234 	/* Initialize vector_irq on a new cpu */
1235 	int irq, vector;
1236 	struct irq_cfg *cfg;
1237 
1238 	/*
1239 	 * vector_lock will make sure that we don't run into irq vector
1240 	 * assignments that might be happening on another cpu in parallel,
1241 	 * while we setup our initial vector to irq mappings.
1242 	 */
1243 	raw_spin_lock(&vector_lock);
1244 	/* Mark the inuse vectors */
1245 	for_each_active_irq(irq) {
1246 		cfg = irq_get_chip_data(irq);
1247 		if (!cfg)
1248 			continue;
1249 		/*
1250 		 * If it is a legacy IRQ handled by the legacy PIC, this cpu
1251 		 * will be part of the irq_cfg's domain.
1252 		 */
1253 		if (irq < legacy_pic->nr_legacy_irqs && !IO_APIC_IRQ(irq))
1254 			cpumask_set_cpu(cpu, cfg->domain);
1255 
1256 		if (!cpumask_test_cpu(cpu, cfg->domain))
1257 			continue;
1258 		vector = cfg->vector;
1259 		per_cpu(vector_irq, cpu)[vector] = irq;
1260 	}
1261 	/* Mark the free vectors */
1262 	for (vector = 0; vector < NR_VECTORS; ++vector) {
1263 		irq = per_cpu(vector_irq, cpu)[vector];
1264 		if (irq < 0)
1265 			continue;
1266 
1267 		cfg = irq_cfg(irq);
1268 		if (!cpumask_test_cpu(cpu, cfg->domain))
1269 			per_cpu(vector_irq, cpu)[vector] = -1;
1270 	}
1271 	raw_spin_unlock(&vector_lock);
1272 }
1273 
1274 static struct irq_chip ioapic_chip;
1275 
1276 #ifdef CONFIG_X86_32
1277 static inline int IO_APIC_irq_trigger(int irq)
1278 {
1279 	int apic, idx, pin;
1280 
1281 	for (apic = 0; apic < nr_ioapics; apic++) {
1282 		for (pin = 0; pin < ioapics[apic].nr_registers; pin++) {
1283 			idx = find_irq_entry(apic, pin, mp_INT);
1284 			if ((idx != -1) && (irq == pin_2_irq(idx, apic, pin)))
1285 				return irq_trigger(idx);
1286 		}
1287 	}
1288 	/*
1289          * nonexistent IRQs are edge default
1290          */
1291 	return 0;
1292 }
1293 #else
1294 static inline int IO_APIC_irq_trigger(int irq)
1295 {
1296 	return 1;
1297 }
1298 #endif
1299 
1300 static void ioapic_register_intr(unsigned int irq, struct irq_cfg *cfg,
1301 				 unsigned long trigger)
1302 {
1303 	struct irq_chip *chip = &ioapic_chip;
1304 	irq_flow_handler_t hdl;
1305 	bool fasteoi;
1306 
1307 	if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1308 	    trigger == IOAPIC_LEVEL) {
1309 		irq_set_status_flags(irq, IRQ_LEVEL);
1310 		fasteoi = true;
1311 	} else {
1312 		irq_clear_status_flags(irq, IRQ_LEVEL);
1313 		fasteoi = false;
1314 	}
1315 
1316 	if (irq_remapped(cfg)) {
1317 		irq_set_status_flags(irq, IRQ_MOVE_PCNTXT);
1318 		irq_remap_modify_chip_defaults(chip);
1319 		fasteoi = trigger != 0;
1320 	}
1321 
1322 	hdl = fasteoi ? handle_fasteoi_irq : handle_edge_irq;
1323 	irq_set_chip_and_handler_name(irq, chip, hdl,
1324 				      fasteoi ? "fasteoi" : "edge");
1325 }
1326 
1327 
1328 static int setup_ir_ioapic_entry(int irq,
1329 			      struct IR_IO_APIC_route_entry *entry,
1330 			      unsigned int destination, int vector,
1331 			      struct io_apic_irq_attr *attr)
1332 {
1333 	int index;
1334 	struct irte irte;
1335 	int ioapic_id = mpc_ioapic_id(attr->ioapic);
1336 	struct intel_iommu *iommu = map_ioapic_to_ir(ioapic_id);
1337 
1338 	if (!iommu) {
1339 		pr_warn("No mapping iommu for ioapic %d\n", ioapic_id);
1340 		return -ENODEV;
1341 	}
1342 
1343 	index = alloc_irte(iommu, irq, 1);
1344 	if (index < 0) {
1345 		pr_warn("Failed to allocate IRTE for ioapic %d\n", ioapic_id);
1346 		return -ENOMEM;
1347 	}
1348 
1349 	prepare_irte(&irte, vector, destination);
1350 
1351 	/* Set source-id of interrupt request */
1352 	set_ioapic_sid(&irte, ioapic_id);
1353 
1354 	modify_irte(irq, &irte);
1355 
1356 	apic_printk(APIC_VERBOSE, KERN_DEBUG "IOAPIC[%d]: "
1357 		"Set IRTE entry (P:%d FPD:%d Dst_Mode:%d "
1358 		"Redir_hint:%d Trig_Mode:%d Dlvry_Mode:%X "
1359 		"Avail:%X Vector:%02X Dest:%08X "
1360 		"SID:%04X SQ:%X SVT:%X)\n",
1361 		attr->ioapic, irte.present, irte.fpd, irte.dst_mode,
1362 		irte.redir_hint, irte.trigger_mode, irte.dlvry_mode,
1363 		irte.avail, irte.vector, irte.dest_id,
1364 		irte.sid, irte.sq, irte.svt);
1365 
1366 	memset(entry, 0, sizeof(*entry));
1367 
1368 	entry->index2	= (index >> 15) & 0x1;
1369 	entry->zero	= 0;
1370 	entry->format	= 1;
1371 	entry->index	= (index & 0x7fff);
1372 	/*
1373 	 * IO-APIC RTE will be configured with virtual vector.
1374 	 * irq handler will do the explicit EOI to the io-apic.
1375 	 */
1376 	entry->vector	= attr->ioapic_pin;
1377 	entry->mask	= 0;			/* enable IRQ */
1378 	entry->trigger	= attr->trigger;
1379 	entry->polarity	= attr->polarity;
1380 
1381 	/* Mask level triggered irqs.
1382 	 * Use IRQ_DELAYED_DISABLE for edge triggered irqs.
1383 	 */
1384 	if (attr->trigger)
1385 		entry->mask = 1;
1386 
1387 	return 0;
1388 }
1389 
1390 static int setup_ioapic_entry(int irq, struct IO_APIC_route_entry *entry,
1391 			       unsigned int destination, int vector,
1392 			       struct io_apic_irq_attr *attr)
1393 {
1394 	if (intr_remapping_enabled)
1395 		return setup_ir_ioapic_entry(irq,
1396 			 (struct IR_IO_APIC_route_entry *)entry,
1397 			 destination, vector, attr);
1398 
1399 	memset(entry, 0, sizeof(*entry));
1400 
1401 	entry->delivery_mode = apic->irq_delivery_mode;
1402 	entry->dest_mode     = apic->irq_dest_mode;
1403 	entry->dest	     = destination;
1404 	entry->vector	     = vector;
1405 	entry->mask	     = 0;			/* enable IRQ */
1406 	entry->trigger	     = attr->trigger;
1407 	entry->polarity	     = attr->polarity;
1408 
1409 	/*
1410 	 * Mask level triggered irqs.
1411 	 * Use IRQ_DELAYED_DISABLE for edge triggered irqs.
1412 	 */
1413 	if (attr->trigger)
1414 		entry->mask = 1;
1415 
1416 	return 0;
1417 }
1418 
1419 static void setup_ioapic_irq(unsigned int irq, struct irq_cfg *cfg,
1420 				struct io_apic_irq_attr *attr)
1421 {
1422 	struct IO_APIC_route_entry entry;
1423 	unsigned int dest;
1424 
1425 	if (!IO_APIC_IRQ(irq))
1426 		return;
1427 	/*
1428 	 * For legacy irqs, cfg->domain starts with cpu 0 for legacy
1429 	 * controllers like 8259. Now that IO-APIC can handle this irq, update
1430 	 * the cfg->domain.
1431 	 */
1432 	if (irq < legacy_pic->nr_legacy_irqs && cpumask_test_cpu(0, cfg->domain))
1433 		apic->vector_allocation_domain(0, cfg->domain);
1434 
1435 	if (assign_irq_vector(irq, cfg, apic->target_cpus()))
1436 		return;
1437 
1438 	dest = apic->cpu_mask_to_apicid_and(cfg->domain, apic->target_cpus());
1439 
1440 	apic_printk(APIC_VERBOSE,KERN_DEBUG
1441 		    "IOAPIC[%d]: Set routing entry (%d-%d -> 0x%x -> "
1442 		    "IRQ %d Mode:%i Active:%i Dest:%d)\n",
1443 		    attr->ioapic, mpc_ioapic_id(attr->ioapic), attr->ioapic_pin,
1444 		    cfg->vector, irq, attr->trigger, attr->polarity, dest);
1445 
1446 	if (setup_ioapic_entry(irq, &entry, dest, cfg->vector, attr)) {
1447 		pr_warn("Failed to setup ioapic entry for ioapic  %d, pin %d\n",
1448 			mpc_ioapic_id(attr->ioapic), attr->ioapic_pin);
1449 		__clear_irq_vector(irq, cfg);
1450 
1451 		return;
1452 	}
1453 
1454 	ioapic_register_intr(irq, cfg, attr->trigger);
1455 	if (irq < legacy_pic->nr_legacy_irqs)
1456 		legacy_pic->mask(irq);
1457 
1458 	ioapic_write_entry(attr->ioapic, attr->ioapic_pin, entry);
1459 }
1460 
1461 static bool __init io_apic_pin_not_connected(int idx, int ioapic_idx, int pin)
1462 {
1463 	if (idx != -1)
1464 		return false;
1465 
1466 	apic_printk(APIC_VERBOSE, KERN_DEBUG " apic %d pin %d not connected\n",
1467 		    mpc_ioapic_id(ioapic_idx), pin);
1468 	return true;
1469 }
1470 
1471 static void __init __io_apic_setup_irqs(unsigned int ioapic_idx)
1472 {
1473 	int idx, node = cpu_to_node(0);
1474 	struct io_apic_irq_attr attr;
1475 	unsigned int pin, irq;
1476 
1477 	for (pin = 0; pin < ioapics[ioapic_idx].nr_registers; pin++) {
1478 		idx = find_irq_entry(ioapic_idx, pin, mp_INT);
1479 		if (io_apic_pin_not_connected(idx, ioapic_idx, pin))
1480 			continue;
1481 
1482 		irq = pin_2_irq(idx, ioapic_idx, pin);
1483 
1484 		if ((ioapic_idx > 0) && (irq > 16))
1485 			continue;
1486 
1487 		/*
1488 		 * Skip the timer IRQ if there's a quirk handler
1489 		 * installed and if it returns 1:
1490 		 */
1491 		if (apic->multi_timer_check &&
1492 		    apic->multi_timer_check(ioapic_idx, irq))
1493 			continue;
1494 
1495 		set_io_apic_irq_attr(&attr, ioapic_idx, pin, irq_trigger(idx),
1496 				     irq_polarity(idx));
1497 
1498 		io_apic_setup_irq_pin(irq, node, &attr);
1499 	}
1500 }
1501 
1502 static void __init setup_IO_APIC_irqs(void)
1503 {
1504 	unsigned int ioapic_idx;
1505 
1506 	apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
1507 
1508 	for (ioapic_idx = 0; ioapic_idx < nr_ioapics; ioapic_idx++)
1509 		__io_apic_setup_irqs(ioapic_idx);
1510 }
1511 
1512 /*
1513  * for the gsit that is not in first ioapic
1514  * but could not use acpi_register_gsi()
1515  * like some special sci in IBM x3330
1516  */
1517 void setup_IO_APIC_irq_extra(u32 gsi)
1518 {
1519 	int ioapic_idx = 0, pin, idx, irq, node = cpu_to_node(0);
1520 	struct io_apic_irq_attr attr;
1521 
1522 	/*
1523 	 * Convert 'gsi' to 'ioapic.pin'.
1524 	 */
1525 	ioapic_idx = mp_find_ioapic(gsi);
1526 	if (ioapic_idx < 0)
1527 		return;
1528 
1529 	pin = mp_find_ioapic_pin(ioapic_idx, gsi);
1530 	idx = find_irq_entry(ioapic_idx, pin, mp_INT);
1531 	if (idx == -1)
1532 		return;
1533 
1534 	irq = pin_2_irq(idx, ioapic_idx, pin);
1535 
1536 	/* Only handle the non legacy irqs on secondary ioapics */
1537 	if (ioapic_idx == 0 || irq < NR_IRQS_LEGACY)
1538 		return;
1539 
1540 	set_io_apic_irq_attr(&attr, ioapic_idx, pin, irq_trigger(idx),
1541 			     irq_polarity(idx));
1542 
1543 	io_apic_setup_irq_pin_once(irq, node, &attr);
1544 }
1545 
1546 /*
1547  * Set up the timer pin, possibly with the 8259A-master behind.
1548  */
1549 static void __init setup_timer_IRQ0_pin(unsigned int ioapic_idx,
1550 					 unsigned int pin, int vector)
1551 {
1552 	struct IO_APIC_route_entry entry;
1553 
1554 	if (intr_remapping_enabled)
1555 		return;
1556 
1557 	memset(&entry, 0, sizeof(entry));
1558 
1559 	/*
1560 	 * We use logical delivery to get the timer IRQ
1561 	 * to the first CPU.
1562 	 */
1563 	entry.dest_mode = apic->irq_dest_mode;
1564 	entry.mask = 0;			/* don't mask IRQ for edge */
1565 	entry.dest = apic->cpu_mask_to_apicid(apic->target_cpus());
1566 	entry.delivery_mode = apic->irq_delivery_mode;
1567 	entry.polarity = 0;
1568 	entry.trigger = 0;
1569 	entry.vector = vector;
1570 
1571 	/*
1572 	 * The timer IRQ doesn't have to know that behind the
1573 	 * scene we may have a 8259A-master in AEOI mode ...
1574 	 */
1575 	irq_set_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq,
1576 				      "edge");
1577 
1578 	/*
1579 	 * Add it to the IO-APIC irq-routing table:
1580 	 */
1581 	ioapic_write_entry(ioapic_idx, pin, entry);
1582 }
1583 
1584 __apicdebuginit(void) print_IO_APIC(int ioapic_idx)
1585 {
1586 	int i;
1587 	union IO_APIC_reg_00 reg_00;
1588 	union IO_APIC_reg_01 reg_01;
1589 	union IO_APIC_reg_02 reg_02;
1590 	union IO_APIC_reg_03 reg_03;
1591 	unsigned long flags;
1592 
1593 	raw_spin_lock_irqsave(&ioapic_lock, flags);
1594 	reg_00.raw = io_apic_read(ioapic_idx, 0);
1595 	reg_01.raw = io_apic_read(ioapic_idx, 1);
1596 	if (reg_01.bits.version >= 0x10)
1597 		reg_02.raw = io_apic_read(ioapic_idx, 2);
1598 	if (reg_01.bits.version >= 0x20)
1599 		reg_03.raw = io_apic_read(ioapic_idx, 3);
1600 	raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1601 
1602 	printk("\n");
1603 	printk(KERN_DEBUG "IO APIC #%d......\n", mpc_ioapic_id(ioapic_idx));
1604 	printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
1605 	printk(KERN_DEBUG ".......    : physical APIC id: %02X\n", reg_00.bits.ID);
1606 	printk(KERN_DEBUG ".......    : Delivery Type: %X\n", reg_00.bits.delivery_type);
1607 	printk(KERN_DEBUG ".......    : LTS          : %X\n", reg_00.bits.LTS);
1608 
1609 	printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)&reg_01);
1610 	printk(KERN_DEBUG ".......     : max redirection entries: %02X\n",
1611 		reg_01.bits.entries);
1612 
1613 	printk(KERN_DEBUG ".......     : PRQ implemented: %X\n", reg_01.bits.PRQ);
1614 	printk(KERN_DEBUG ".......     : IO APIC version: %02X\n",
1615 		reg_01.bits.version);
1616 
1617 	/*
1618 	 * Some Intel chipsets with IO APIC VERSION of 0x1? don't have reg_02,
1619 	 * but the value of reg_02 is read as the previous read register
1620 	 * value, so ignore it if reg_02 == reg_01.
1621 	 */
1622 	if (reg_01.bits.version >= 0x10 && reg_02.raw != reg_01.raw) {
1623 		printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
1624 		printk(KERN_DEBUG ".......     : arbitration: %02X\n", reg_02.bits.arbitration);
1625 	}
1626 
1627 	/*
1628 	 * Some Intel chipsets with IO APIC VERSION of 0x2? don't have reg_02
1629 	 * or reg_03, but the value of reg_0[23] is read as the previous read
1630 	 * register value, so ignore it if reg_03 == reg_0[12].
1631 	 */
1632 	if (reg_01.bits.version >= 0x20 && reg_03.raw != reg_02.raw &&
1633 	    reg_03.raw != reg_01.raw) {
1634 		printk(KERN_DEBUG ".... register #03: %08X\n", reg_03.raw);
1635 		printk(KERN_DEBUG ".......     : Boot DT    : %X\n", reg_03.bits.boot_DT);
1636 	}
1637 
1638 	printk(KERN_DEBUG ".... IRQ redirection table:\n");
1639 
1640 	if (intr_remapping_enabled) {
1641 		printk(KERN_DEBUG " NR Indx Fmt Mask Trig IRR"
1642 			" Pol Stat Indx2 Zero Vect:\n");
1643 	} else {
1644 		printk(KERN_DEBUG " NR Dst Mask Trig IRR Pol"
1645 			" Stat Dmod Deli Vect:\n");
1646 	}
1647 
1648 	for (i = 0; i <= reg_01.bits.entries; i++) {
1649 		if (intr_remapping_enabled) {
1650 			struct IO_APIC_route_entry entry;
1651 			struct IR_IO_APIC_route_entry *ir_entry;
1652 
1653 			entry = ioapic_read_entry(ioapic_idx, i);
1654 			ir_entry = (struct IR_IO_APIC_route_entry *) &entry;
1655 			printk(KERN_DEBUG " %02x %04X ",
1656 				i,
1657 				ir_entry->index
1658 			);
1659 			printk("%1d   %1d    %1d    %1d   %1d   "
1660 				"%1d    %1d     %X    %02X\n",
1661 				ir_entry->format,
1662 				ir_entry->mask,
1663 				ir_entry->trigger,
1664 				ir_entry->irr,
1665 				ir_entry->polarity,
1666 				ir_entry->delivery_status,
1667 				ir_entry->index2,
1668 				ir_entry->zero,
1669 				ir_entry->vector
1670 			);
1671 		} else {
1672 			struct IO_APIC_route_entry entry;
1673 
1674 			entry = ioapic_read_entry(ioapic_idx, i);
1675 			printk(KERN_DEBUG " %02x %02X  ",
1676 				i,
1677 				entry.dest
1678 			);
1679 			printk("%1d    %1d    %1d   %1d   %1d    "
1680 				"%1d    %1d    %02X\n",
1681 				entry.mask,
1682 				entry.trigger,
1683 				entry.irr,
1684 				entry.polarity,
1685 				entry.delivery_status,
1686 				entry.dest_mode,
1687 				entry.delivery_mode,
1688 				entry.vector
1689 			);
1690 		}
1691 	}
1692 }
1693 
1694 __apicdebuginit(void) print_IO_APICs(void)
1695 {
1696 	int ioapic_idx;
1697 	struct irq_cfg *cfg;
1698 	unsigned int irq;
1699 
1700 	printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
1701 	for (ioapic_idx = 0; ioapic_idx < nr_ioapics; ioapic_idx++)
1702 		printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
1703 		       mpc_ioapic_id(ioapic_idx),
1704 		       ioapics[ioapic_idx].nr_registers);
1705 
1706 	/*
1707 	 * We are a bit conservative about what we expect.  We have to
1708 	 * know about every hardware change ASAP.
1709 	 */
1710 	printk(KERN_INFO "testing the IO APIC.......................\n");
1711 
1712 	for (ioapic_idx = 0; ioapic_idx < nr_ioapics; ioapic_idx++)
1713 		print_IO_APIC(ioapic_idx);
1714 
1715 	printk(KERN_DEBUG "IRQ to pin mappings:\n");
1716 	for_each_active_irq(irq) {
1717 		struct irq_pin_list *entry;
1718 
1719 		cfg = irq_get_chip_data(irq);
1720 		if (!cfg)
1721 			continue;
1722 		entry = cfg->irq_2_pin;
1723 		if (!entry)
1724 			continue;
1725 		printk(KERN_DEBUG "IRQ%d ", irq);
1726 		for_each_irq_pin(entry, cfg->irq_2_pin)
1727 			printk("-> %d:%d", entry->apic, entry->pin);
1728 		printk("\n");
1729 	}
1730 
1731 	printk(KERN_INFO ".................................... done.\n");
1732 }
1733 
1734 __apicdebuginit(void) print_APIC_field(int base)
1735 {
1736 	int i;
1737 
1738 	printk(KERN_DEBUG);
1739 
1740 	for (i = 0; i < 8; i++)
1741 		printk(KERN_CONT "%08x", apic_read(base + i*0x10));
1742 
1743 	printk(KERN_CONT "\n");
1744 }
1745 
1746 __apicdebuginit(void) print_local_APIC(void *dummy)
1747 {
1748 	unsigned int i, v, ver, maxlvt;
1749 	u64 icr;
1750 
1751 	printk(KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1752 		smp_processor_id(), hard_smp_processor_id());
1753 	v = apic_read(APIC_ID);
1754 	printk(KERN_INFO "... APIC ID:      %08x (%01x)\n", v, read_apic_id());
1755 	v = apic_read(APIC_LVR);
1756 	printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1757 	ver = GET_APIC_VERSION(v);
1758 	maxlvt = lapic_get_maxlvt();
1759 
1760 	v = apic_read(APIC_TASKPRI);
1761 	printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1762 
1763 	if (APIC_INTEGRATED(ver)) {                     /* !82489DX */
1764 		if (!APIC_XAPIC(ver)) {
1765 			v = apic_read(APIC_ARBPRI);
1766 			printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
1767 			       v & APIC_ARBPRI_MASK);
1768 		}
1769 		v = apic_read(APIC_PROCPRI);
1770 		printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1771 	}
1772 
1773 	/*
1774 	 * Remote read supported only in the 82489DX and local APIC for
1775 	 * Pentium processors.
1776 	 */
1777 	if (!APIC_INTEGRATED(ver) || maxlvt == 3) {
1778 		v = apic_read(APIC_RRR);
1779 		printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1780 	}
1781 
1782 	v = apic_read(APIC_LDR);
1783 	printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1784 	if (!x2apic_enabled()) {
1785 		v = apic_read(APIC_DFR);
1786 		printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1787 	}
1788 	v = apic_read(APIC_SPIV);
1789 	printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1790 
1791 	printk(KERN_DEBUG "... APIC ISR field:\n");
1792 	print_APIC_field(APIC_ISR);
1793 	printk(KERN_DEBUG "... APIC TMR field:\n");
1794 	print_APIC_field(APIC_TMR);
1795 	printk(KERN_DEBUG "... APIC IRR field:\n");
1796 	print_APIC_field(APIC_IRR);
1797 
1798 	if (APIC_INTEGRATED(ver)) {             /* !82489DX */
1799 		if (maxlvt > 3)         /* Due to the Pentium erratum 3AP. */
1800 			apic_write(APIC_ESR, 0);
1801 
1802 		v = apic_read(APIC_ESR);
1803 		printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1804 	}
1805 
1806 	icr = apic_icr_read();
1807 	printk(KERN_DEBUG "... APIC ICR: %08x\n", (u32)icr);
1808 	printk(KERN_DEBUG "... APIC ICR2: %08x\n", (u32)(icr >> 32));
1809 
1810 	v = apic_read(APIC_LVTT);
1811 	printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1812 
1813 	if (maxlvt > 3) {                       /* PC is LVT#4. */
1814 		v = apic_read(APIC_LVTPC);
1815 		printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1816 	}
1817 	v = apic_read(APIC_LVT0);
1818 	printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1819 	v = apic_read(APIC_LVT1);
1820 	printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1821 
1822 	if (maxlvt > 2) {			/* ERR is LVT#3. */
1823 		v = apic_read(APIC_LVTERR);
1824 		printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1825 	}
1826 
1827 	v = apic_read(APIC_TMICT);
1828 	printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1829 	v = apic_read(APIC_TMCCT);
1830 	printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1831 	v = apic_read(APIC_TDCR);
1832 	printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1833 
1834 	if (boot_cpu_has(X86_FEATURE_EXTAPIC)) {
1835 		v = apic_read(APIC_EFEAT);
1836 		maxlvt = (v >> 16) & 0xff;
1837 		printk(KERN_DEBUG "... APIC EFEAT: %08x\n", v);
1838 		v = apic_read(APIC_ECTRL);
1839 		printk(KERN_DEBUG "... APIC ECTRL: %08x\n", v);
1840 		for (i = 0; i < maxlvt; i++) {
1841 			v = apic_read(APIC_EILVTn(i));
1842 			printk(KERN_DEBUG "... APIC EILVT%d: %08x\n", i, v);
1843 		}
1844 	}
1845 	printk("\n");
1846 }
1847 
1848 __apicdebuginit(void) print_local_APICs(int maxcpu)
1849 {
1850 	int cpu;
1851 
1852 	if (!maxcpu)
1853 		return;
1854 
1855 	preempt_disable();
1856 	for_each_online_cpu(cpu) {
1857 		if (cpu >= maxcpu)
1858 			break;
1859 		smp_call_function_single(cpu, print_local_APIC, NULL, 1);
1860 	}
1861 	preempt_enable();
1862 }
1863 
1864 __apicdebuginit(void) print_PIC(void)
1865 {
1866 	unsigned int v;
1867 	unsigned long flags;
1868 
1869 	if (!legacy_pic->nr_legacy_irqs)
1870 		return;
1871 
1872 	printk(KERN_DEBUG "\nprinting PIC contents\n");
1873 
1874 	raw_spin_lock_irqsave(&i8259A_lock, flags);
1875 
1876 	v = inb(0xa1) << 8 | inb(0x21);
1877 	printk(KERN_DEBUG "... PIC  IMR: %04x\n", v);
1878 
1879 	v = inb(0xa0) << 8 | inb(0x20);
1880 	printk(KERN_DEBUG "... PIC  IRR: %04x\n", v);
1881 
1882 	outb(0x0b,0xa0);
1883 	outb(0x0b,0x20);
1884 	v = inb(0xa0) << 8 | inb(0x20);
1885 	outb(0x0a,0xa0);
1886 	outb(0x0a,0x20);
1887 
1888 	raw_spin_unlock_irqrestore(&i8259A_lock, flags);
1889 
1890 	printk(KERN_DEBUG "... PIC  ISR: %04x\n", v);
1891 
1892 	v = inb(0x4d1) << 8 | inb(0x4d0);
1893 	printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1894 }
1895 
1896 static int __initdata show_lapic = 1;
1897 static __init int setup_show_lapic(char *arg)
1898 {
1899 	int num = -1;
1900 
1901 	if (strcmp(arg, "all") == 0) {
1902 		show_lapic = CONFIG_NR_CPUS;
1903 	} else {
1904 		get_option(&arg, &num);
1905 		if (num >= 0)
1906 			show_lapic = num;
1907 	}
1908 
1909 	return 1;
1910 }
1911 __setup("show_lapic=", setup_show_lapic);
1912 
1913 __apicdebuginit(int) print_ICs(void)
1914 {
1915 	if (apic_verbosity == APIC_QUIET)
1916 		return 0;
1917 
1918 	print_PIC();
1919 
1920 	/* don't print out if apic is not there */
1921 	if (!cpu_has_apic && !apic_from_smp_config())
1922 		return 0;
1923 
1924 	print_local_APICs(show_lapic);
1925 	print_IO_APICs();
1926 
1927 	return 0;
1928 }
1929 
1930 late_initcall(print_ICs);
1931 
1932 
1933 /* Where if anywhere is the i8259 connect in external int mode */
1934 static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
1935 
1936 void __init enable_IO_APIC(void)
1937 {
1938 	int i8259_apic, i8259_pin;
1939 	int apic;
1940 
1941 	if (!legacy_pic->nr_legacy_irqs)
1942 		return;
1943 
1944 	for(apic = 0; apic < nr_ioapics; apic++) {
1945 		int pin;
1946 		/* See if any of the pins is in ExtINT mode */
1947 		for (pin = 0; pin < ioapics[apic].nr_registers; pin++) {
1948 			struct IO_APIC_route_entry entry;
1949 			entry = ioapic_read_entry(apic, pin);
1950 
1951 			/* If the interrupt line is enabled and in ExtInt mode
1952 			 * I have found the pin where the i8259 is connected.
1953 			 */
1954 			if ((entry.mask == 0) && (entry.delivery_mode == dest_ExtINT)) {
1955 				ioapic_i8259.apic = apic;
1956 				ioapic_i8259.pin  = pin;
1957 				goto found_i8259;
1958 			}
1959 		}
1960 	}
1961  found_i8259:
1962 	/* Look to see what if the MP table has reported the ExtINT */
1963 	/* If we could not find the appropriate pin by looking at the ioapic
1964 	 * the i8259 probably is not connected the ioapic but give the
1965 	 * mptable a chance anyway.
1966 	 */
1967 	i8259_pin  = find_isa_irq_pin(0, mp_ExtINT);
1968 	i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
1969 	/* Trust the MP table if nothing is setup in the hardware */
1970 	if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
1971 		printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
1972 		ioapic_i8259.pin  = i8259_pin;
1973 		ioapic_i8259.apic = i8259_apic;
1974 	}
1975 	/* Complain if the MP table and the hardware disagree */
1976 	if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
1977 		(i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
1978 	{
1979 		printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
1980 	}
1981 
1982 	/*
1983 	 * Do not trust the IO-APIC being empty at bootup
1984 	 */
1985 	clear_IO_APIC();
1986 }
1987 
1988 /*
1989  * Not an __init, needed by the reboot code
1990  */
1991 void disable_IO_APIC(void)
1992 {
1993 	/*
1994 	 * Clear the IO-APIC before rebooting:
1995 	 */
1996 	clear_IO_APIC();
1997 
1998 	if (!legacy_pic->nr_legacy_irqs)
1999 		return;
2000 
2001 	/*
2002 	 * If the i8259 is routed through an IOAPIC
2003 	 * Put that IOAPIC in virtual wire mode
2004 	 * so legacy interrupts can be delivered.
2005 	 *
2006 	 * With interrupt-remapping, for now we will use virtual wire A mode,
2007 	 * as virtual wire B is little complex (need to configure both
2008 	 * IOAPIC RTE as well as interrupt-remapping table entry).
2009 	 * As this gets called during crash dump, keep this simple for now.
2010 	 */
2011 	if (ioapic_i8259.pin != -1 && !intr_remapping_enabled) {
2012 		struct IO_APIC_route_entry entry;
2013 
2014 		memset(&entry, 0, sizeof(entry));
2015 		entry.mask            = 0; /* Enabled */
2016 		entry.trigger         = 0; /* Edge */
2017 		entry.irr             = 0;
2018 		entry.polarity        = 0; /* High */
2019 		entry.delivery_status = 0;
2020 		entry.dest_mode       = 0; /* Physical */
2021 		entry.delivery_mode   = dest_ExtINT; /* ExtInt */
2022 		entry.vector          = 0;
2023 		entry.dest            = read_apic_id();
2024 
2025 		/*
2026 		 * Add it to the IO-APIC irq-routing table:
2027 		 */
2028 		ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
2029 	}
2030 
2031 	/*
2032 	 * Use virtual wire A mode when interrupt remapping is enabled.
2033 	 */
2034 	if (cpu_has_apic || apic_from_smp_config())
2035 		disconnect_bsp_APIC(!intr_remapping_enabled &&
2036 				ioapic_i8259.pin != -1);
2037 }
2038 
2039 #ifdef CONFIG_X86_32
2040 /*
2041  * function to set the IO-APIC physical IDs based on the
2042  * values stored in the MPC table.
2043  *
2044  * by Matt Domsch <Matt_Domsch@dell.com>  Tue Dec 21 12:25:05 CST 1999
2045  */
2046 void __init setup_ioapic_ids_from_mpc_nocheck(void)
2047 {
2048 	union IO_APIC_reg_00 reg_00;
2049 	physid_mask_t phys_id_present_map;
2050 	int ioapic_idx;
2051 	int i;
2052 	unsigned char old_id;
2053 	unsigned long flags;
2054 
2055 	/*
2056 	 * This is broken; anything with a real cpu count has to
2057 	 * circumvent this idiocy regardless.
2058 	 */
2059 	apic->ioapic_phys_id_map(&phys_cpu_present_map, &phys_id_present_map);
2060 
2061 	/*
2062 	 * Set the IOAPIC ID to the value stored in the MPC table.
2063 	 */
2064 	for (ioapic_idx = 0; ioapic_idx < nr_ioapics; ioapic_idx++) {
2065 		/* Read the register 0 value */
2066 		raw_spin_lock_irqsave(&ioapic_lock, flags);
2067 		reg_00.raw = io_apic_read(ioapic_idx, 0);
2068 		raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2069 
2070 		old_id = mpc_ioapic_id(ioapic_idx);
2071 
2072 		if (mpc_ioapic_id(ioapic_idx) >= get_physical_broadcast()) {
2073 			printk(KERN_ERR "BIOS bug, IO-APIC#%d ID is %d in the MPC table!...\n",
2074 				ioapic_idx, mpc_ioapic_id(ioapic_idx));
2075 			printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
2076 				reg_00.bits.ID);
2077 			ioapics[ioapic_idx].mp_config.apicid = reg_00.bits.ID;
2078 		}
2079 
2080 		/*
2081 		 * Sanity check, is the ID really free? Every APIC in a
2082 		 * system must have a unique ID or we get lots of nice
2083 		 * 'stuck on smp_invalidate_needed IPI wait' messages.
2084 		 */
2085 		if (apic->check_apicid_used(&phys_id_present_map,
2086 					    mpc_ioapic_id(ioapic_idx))) {
2087 			printk(KERN_ERR "BIOS bug, IO-APIC#%d ID %d is already used!...\n",
2088 				ioapic_idx, mpc_ioapic_id(ioapic_idx));
2089 			for (i = 0; i < get_physical_broadcast(); i++)
2090 				if (!physid_isset(i, phys_id_present_map))
2091 					break;
2092 			if (i >= get_physical_broadcast())
2093 				panic("Max APIC ID exceeded!\n");
2094 			printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
2095 				i);
2096 			physid_set(i, phys_id_present_map);
2097 			ioapics[ioapic_idx].mp_config.apicid = i;
2098 		} else {
2099 			physid_mask_t tmp;
2100 			apic->apicid_to_cpu_present(mpc_ioapic_id(ioapic_idx),
2101 						    &tmp);
2102 			apic_printk(APIC_VERBOSE, "Setting %d in the "
2103 					"phys_id_present_map\n",
2104 					mpc_ioapic_id(ioapic_idx));
2105 			physids_or(phys_id_present_map, phys_id_present_map, tmp);
2106 		}
2107 
2108 		/*
2109 		 * We need to adjust the IRQ routing table
2110 		 * if the ID changed.
2111 		 */
2112 		if (old_id != mpc_ioapic_id(ioapic_idx))
2113 			for (i = 0; i < mp_irq_entries; i++)
2114 				if (mp_irqs[i].dstapic == old_id)
2115 					mp_irqs[i].dstapic
2116 						= mpc_ioapic_id(ioapic_idx);
2117 
2118 		/*
2119 		 * Update the ID register according to the right value
2120 		 * from the MPC table if they are different.
2121 		 */
2122 		if (mpc_ioapic_id(ioapic_idx) == reg_00.bits.ID)
2123 			continue;
2124 
2125 		apic_printk(APIC_VERBOSE, KERN_INFO
2126 			"...changing IO-APIC physical APIC ID to %d ...",
2127 			mpc_ioapic_id(ioapic_idx));
2128 
2129 		reg_00.bits.ID = mpc_ioapic_id(ioapic_idx);
2130 		raw_spin_lock_irqsave(&ioapic_lock, flags);
2131 		io_apic_write(ioapic_idx, 0, reg_00.raw);
2132 		raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2133 
2134 		/*
2135 		 * Sanity check
2136 		 */
2137 		raw_spin_lock_irqsave(&ioapic_lock, flags);
2138 		reg_00.raw = io_apic_read(ioapic_idx, 0);
2139 		raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2140 		if (reg_00.bits.ID != mpc_ioapic_id(ioapic_idx))
2141 			printk("could not set ID!\n");
2142 		else
2143 			apic_printk(APIC_VERBOSE, " ok.\n");
2144 	}
2145 }
2146 
2147 void __init setup_ioapic_ids_from_mpc(void)
2148 {
2149 
2150 	if (acpi_ioapic)
2151 		return;
2152 	/*
2153 	 * Don't check I/O APIC IDs for xAPIC systems.  They have
2154 	 * no meaning without the serial APIC bus.
2155 	 */
2156 	if (!(boot_cpu_data.x86_vendor == X86_VENDOR_INTEL)
2157 		|| APIC_XAPIC(apic_version[boot_cpu_physical_apicid]))
2158 		return;
2159 	setup_ioapic_ids_from_mpc_nocheck();
2160 }
2161 #endif
2162 
2163 int no_timer_check __initdata;
2164 
2165 static int __init notimercheck(char *s)
2166 {
2167 	no_timer_check = 1;
2168 	return 1;
2169 }
2170 __setup("no_timer_check", notimercheck);
2171 
2172 /*
2173  * There is a nasty bug in some older SMP boards, their mptable lies
2174  * about the timer IRQ. We do the following to work around the situation:
2175  *
2176  *	- timer IRQ defaults to IO-APIC IRQ
2177  *	- if this function detects that timer IRQs are defunct, then we fall
2178  *	  back to ISA timer IRQs
2179  */
2180 static int __init timer_irq_works(void)
2181 {
2182 	unsigned long t1 = jiffies;
2183 	unsigned long flags;
2184 
2185 	if (no_timer_check)
2186 		return 1;
2187 
2188 	local_save_flags(flags);
2189 	local_irq_enable();
2190 	/* Let ten ticks pass... */
2191 	mdelay((10 * 1000) / HZ);
2192 	local_irq_restore(flags);
2193 
2194 	/*
2195 	 * Expect a few ticks at least, to be sure some possible
2196 	 * glue logic does not lock up after one or two first
2197 	 * ticks in a non-ExtINT mode.  Also the local APIC
2198 	 * might have cached one ExtINT interrupt.  Finally, at
2199 	 * least one tick may be lost due to delays.
2200 	 */
2201 
2202 	/* jiffies wrap? */
2203 	if (time_after(jiffies, t1 + 4))
2204 		return 1;
2205 	return 0;
2206 }
2207 
2208 /*
2209  * In the SMP+IOAPIC case it might happen that there are an unspecified
2210  * number of pending IRQ events unhandled. These cases are very rare,
2211  * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
2212  * better to do it this way as thus we do not have to be aware of
2213  * 'pending' interrupts in the IRQ path, except at this point.
2214  */
2215 /*
2216  * Edge triggered needs to resend any interrupt
2217  * that was delayed but this is now handled in the device
2218  * independent code.
2219  */
2220 
2221 /*
2222  * Starting up a edge-triggered IO-APIC interrupt is
2223  * nasty - we need to make sure that we get the edge.
2224  * If it is already asserted for some reason, we need
2225  * return 1 to indicate that is was pending.
2226  *
2227  * This is not complete - we should be able to fake
2228  * an edge even if it isn't on the 8259A...
2229  */
2230 
2231 static unsigned int startup_ioapic_irq(struct irq_data *data)
2232 {
2233 	int was_pending = 0, irq = data->irq;
2234 	unsigned long flags;
2235 
2236 	raw_spin_lock_irqsave(&ioapic_lock, flags);
2237 	if (irq < legacy_pic->nr_legacy_irqs) {
2238 		legacy_pic->mask(irq);
2239 		if (legacy_pic->irq_pending(irq))
2240 			was_pending = 1;
2241 	}
2242 	__unmask_ioapic(data->chip_data);
2243 	raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2244 
2245 	return was_pending;
2246 }
2247 
2248 static int ioapic_retrigger_irq(struct irq_data *data)
2249 {
2250 	struct irq_cfg *cfg = data->chip_data;
2251 	unsigned long flags;
2252 
2253 	raw_spin_lock_irqsave(&vector_lock, flags);
2254 	apic->send_IPI_mask(cpumask_of(cpumask_first(cfg->domain)), cfg->vector);
2255 	raw_spin_unlock_irqrestore(&vector_lock, flags);
2256 
2257 	return 1;
2258 }
2259 
2260 /*
2261  * Level and edge triggered IO-APIC interrupts need different handling,
2262  * so we use two separate IRQ descriptors. Edge triggered IRQs can be
2263  * handled with the level-triggered descriptor, but that one has slightly
2264  * more overhead. Level-triggered interrupts cannot be handled with the
2265  * edge-triggered handler, without risking IRQ storms and other ugly
2266  * races.
2267  */
2268 
2269 #ifdef CONFIG_SMP
2270 void send_cleanup_vector(struct irq_cfg *cfg)
2271 {
2272 	cpumask_var_t cleanup_mask;
2273 
2274 	if (unlikely(!alloc_cpumask_var(&cleanup_mask, GFP_ATOMIC))) {
2275 		unsigned int i;
2276 		for_each_cpu_and(i, cfg->old_domain, cpu_online_mask)
2277 			apic->send_IPI_mask(cpumask_of(i), IRQ_MOVE_CLEANUP_VECTOR);
2278 	} else {
2279 		cpumask_and(cleanup_mask, cfg->old_domain, cpu_online_mask);
2280 		apic->send_IPI_mask(cleanup_mask, IRQ_MOVE_CLEANUP_VECTOR);
2281 		free_cpumask_var(cleanup_mask);
2282 	}
2283 	cfg->move_in_progress = 0;
2284 }
2285 
2286 static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, struct irq_cfg *cfg)
2287 {
2288 	int apic, pin;
2289 	struct irq_pin_list *entry;
2290 	u8 vector = cfg->vector;
2291 
2292 	for_each_irq_pin(entry, cfg->irq_2_pin) {
2293 		unsigned int reg;
2294 
2295 		apic = entry->apic;
2296 		pin = entry->pin;
2297 		/*
2298 		 * With interrupt-remapping, destination information comes
2299 		 * from interrupt-remapping table entry.
2300 		 */
2301 		if (!irq_remapped(cfg))
2302 			io_apic_write(apic, 0x11 + pin*2, dest);
2303 		reg = io_apic_read(apic, 0x10 + pin*2);
2304 		reg &= ~IO_APIC_REDIR_VECTOR_MASK;
2305 		reg |= vector;
2306 		io_apic_modify(apic, 0x10 + pin*2, reg);
2307 	}
2308 }
2309 
2310 /*
2311  * Either sets data->affinity to a valid value, and returns
2312  * ->cpu_mask_to_apicid of that in dest_id, or returns -1 and
2313  * leaves data->affinity untouched.
2314  */
2315 int __ioapic_set_affinity(struct irq_data *data, const struct cpumask *mask,
2316 			  unsigned int *dest_id)
2317 {
2318 	struct irq_cfg *cfg = data->chip_data;
2319 
2320 	if (!cpumask_intersects(mask, cpu_online_mask))
2321 		return -1;
2322 
2323 	if (assign_irq_vector(data->irq, data->chip_data, mask))
2324 		return -1;
2325 
2326 	cpumask_copy(data->affinity, mask);
2327 
2328 	*dest_id = apic->cpu_mask_to_apicid_and(mask, cfg->domain);
2329 	return 0;
2330 }
2331 
2332 static int
2333 ioapic_set_affinity(struct irq_data *data, const struct cpumask *mask,
2334 		    bool force)
2335 {
2336 	unsigned int dest, irq = data->irq;
2337 	unsigned long flags;
2338 	int ret;
2339 
2340 	raw_spin_lock_irqsave(&ioapic_lock, flags);
2341 	ret = __ioapic_set_affinity(data, mask, &dest);
2342 	if (!ret) {
2343 		/* Only the high 8 bits are valid. */
2344 		dest = SET_APIC_LOGICAL_ID(dest);
2345 		__target_IO_APIC_irq(irq, dest, data->chip_data);
2346 	}
2347 	raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2348 	return ret;
2349 }
2350 
2351 #ifdef CONFIG_IRQ_REMAP
2352 
2353 /*
2354  * Migrate the IO-APIC irq in the presence of intr-remapping.
2355  *
2356  * For both level and edge triggered, irq migration is a simple atomic
2357  * update(of vector and cpu destination) of IRTE and flush the hardware cache.
2358  *
2359  * For level triggered, we eliminate the io-apic RTE modification (with the
2360  * updated vector information), by using a virtual vector (io-apic pin number).
2361  * Real vector that is used for interrupting cpu will be coming from
2362  * the interrupt-remapping table entry.
2363  *
2364  * As the migration is a simple atomic update of IRTE, the same mechanism
2365  * is used to migrate MSI irq's in the presence of interrupt-remapping.
2366  */
2367 static int
2368 ir_ioapic_set_affinity(struct irq_data *data, const struct cpumask *mask,
2369 		       bool force)
2370 {
2371 	struct irq_cfg *cfg = data->chip_data;
2372 	unsigned int dest, irq = data->irq;
2373 	struct irte irte;
2374 
2375 	if (!cpumask_intersects(mask, cpu_online_mask))
2376 		return -EINVAL;
2377 
2378 	if (get_irte(irq, &irte))
2379 		return -EBUSY;
2380 
2381 	if (assign_irq_vector(irq, cfg, mask))
2382 		return -EBUSY;
2383 
2384 	dest = apic->cpu_mask_to_apicid_and(cfg->domain, mask);
2385 
2386 	irte.vector = cfg->vector;
2387 	irte.dest_id = IRTE_DEST(dest);
2388 
2389 	/*
2390 	 * Atomically updates the IRTE with the new destination, vector
2391 	 * and flushes the interrupt entry cache.
2392 	 */
2393 	modify_irte(irq, &irte);
2394 
2395 	/*
2396 	 * After this point, all the interrupts will start arriving
2397 	 * at the new destination. So, time to cleanup the previous
2398 	 * vector allocation.
2399 	 */
2400 	if (cfg->move_in_progress)
2401 		send_cleanup_vector(cfg);
2402 
2403 	cpumask_copy(data->affinity, mask);
2404 	return 0;
2405 }
2406 
2407 #else
2408 static inline int
2409 ir_ioapic_set_affinity(struct irq_data *data, const struct cpumask *mask,
2410 		       bool force)
2411 {
2412 	return 0;
2413 }
2414 #endif
2415 
2416 asmlinkage void smp_irq_move_cleanup_interrupt(void)
2417 {
2418 	unsigned vector, me;
2419 
2420 	ack_APIC_irq();
2421 	exit_idle();
2422 	irq_enter();
2423 
2424 	me = smp_processor_id();
2425 	for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
2426 		unsigned int irq;
2427 		unsigned int irr;
2428 		struct irq_desc *desc;
2429 		struct irq_cfg *cfg;
2430 		irq = __this_cpu_read(vector_irq[vector]);
2431 
2432 		if (irq == -1)
2433 			continue;
2434 
2435 		desc = irq_to_desc(irq);
2436 		if (!desc)
2437 			continue;
2438 
2439 		cfg = irq_cfg(irq);
2440 		raw_spin_lock(&desc->lock);
2441 
2442 		/*
2443 		 * Check if the irq migration is in progress. If so, we
2444 		 * haven't received the cleanup request yet for this irq.
2445 		 */
2446 		if (cfg->move_in_progress)
2447 			goto unlock;
2448 
2449 		if (vector == cfg->vector && cpumask_test_cpu(me, cfg->domain))
2450 			goto unlock;
2451 
2452 		irr = apic_read(APIC_IRR + (vector / 32 * 0x10));
2453 		/*
2454 		 * Check if the vector that needs to be cleanedup is
2455 		 * registered at the cpu's IRR. If so, then this is not
2456 		 * the best time to clean it up. Lets clean it up in the
2457 		 * next attempt by sending another IRQ_MOVE_CLEANUP_VECTOR
2458 		 * to myself.
2459 		 */
2460 		if (irr  & (1 << (vector % 32))) {
2461 			apic->send_IPI_self(IRQ_MOVE_CLEANUP_VECTOR);
2462 			goto unlock;
2463 		}
2464 		__this_cpu_write(vector_irq[vector], -1);
2465 unlock:
2466 		raw_spin_unlock(&desc->lock);
2467 	}
2468 
2469 	irq_exit();
2470 }
2471 
2472 static void __irq_complete_move(struct irq_cfg *cfg, unsigned vector)
2473 {
2474 	unsigned me;
2475 
2476 	if (likely(!cfg->move_in_progress))
2477 		return;
2478 
2479 	me = smp_processor_id();
2480 
2481 	if (vector == cfg->vector && cpumask_test_cpu(me, cfg->domain))
2482 		send_cleanup_vector(cfg);
2483 }
2484 
2485 static void irq_complete_move(struct irq_cfg *cfg)
2486 {
2487 	__irq_complete_move(cfg, ~get_irq_regs()->orig_ax);
2488 }
2489 
2490 void irq_force_complete_move(int irq)
2491 {
2492 	struct irq_cfg *cfg = irq_get_chip_data(irq);
2493 
2494 	if (!cfg)
2495 		return;
2496 
2497 	__irq_complete_move(cfg, cfg->vector);
2498 }
2499 #else
2500 static inline void irq_complete_move(struct irq_cfg *cfg) { }
2501 #endif
2502 
2503 static void ack_apic_edge(struct irq_data *data)
2504 {
2505 	irq_complete_move(data->chip_data);
2506 	irq_move_irq(data);
2507 	ack_APIC_irq();
2508 }
2509 
2510 atomic_t irq_mis_count;
2511 
2512 static void ack_apic_level(struct irq_data *data)
2513 {
2514 	struct irq_cfg *cfg = data->chip_data;
2515 	int i, do_unmask_irq = 0, irq = data->irq;
2516 	unsigned long v;
2517 
2518 	irq_complete_move(cfg);
2519 #ifdef CONFIG_GENERIC_PENDING_IRQ
2520 	/* If we are moving the irq we need to mask it */
2521 	if (unlikely(irqd_is_setaffinity_pending(data))) {
2522 		do_unmask_irq = 1;
2523 		mask_ioapic(cfg);
2524 	}
2525 #endif
2526 
2527 	/*
2528 	 * It appears there is an erratum which affects at least version 0x11
2529 	 * of I/O APIC (that's the 82093AA and cores integrated into various
2530 	 * chipsets).  Under certain conditions a level-triggered interrupt is
2531 	 * erroneously delivered as edge-triggered one but the respective IRR
2532 	 * bit gets set nevertheless.  As a result the I/O unit expects an EOI
2533 	 * message but it will never arrive and further interrupts are blocked
2534 	 * from the source.  The exact reason is so far unknown, but the
2535 	 * phenomenon was observed when two consecutive interrupt requests
2536 	 * from a given source get delivered to the same CPU and the source is
2537 	 * temporarily disabled in between.
2538 	 *
2539 	 * A workaround is to simulate an EOI message manually.  We achieve it
2540 	 * by setting the trigger mode to edge and then to level when the edge
2541 	 * trigger mode gets detected in the TMR of a local APIC for a
2542 	 * level-triggered interrupt.  We mask the source for the time of the
2543 	 * operation to prevent an edge-triggered interrupt escaping meanwhile.
2544 	 * The idea is from Manfred Spraul.  --macro
2545 	 *
2546 	 * Also in the case when cpu goes offline, fixup_irqs() will forward
2547 	 * any unhandled interrupt on the offlined cpu to the new cpu
2548 	 * destination that is handling the corresponding interrupt. This
2549 	 * interrupt forwarding is done via IPI's. Hence, in this case also
2550 	 * level-triggered io-apic interrupt will be seen as an edge
2551 	 * interrupt in the IRR. And we can't rely on the cpu's EOI
2552 	 * to be broadcasted to the IO-APIC's which will clear the remoteIRR
2553 	 * corresponding to the level-triggered interrupt. Hence on IO-APIC's
2554 	 * supporting EOI register, we do an explicit EOI to clear the
2555 	 * remote IRR and on IO-APIC's which don't have an EOI register,
2556 	 * we use the above logic (mask+edge followed by unmask+level) from
2557 	 * Manfred Spraul to clear the remote IRR.
2558 	 */
2559 	i = cfg->vector;
2560 	v = apic_read(APIC_TMR + ((i & ~0x1f) >> 1));
2561 
2562 	/*
2563 	 * We must acknowledge the irq before we move it or the acknowledge will
2564 	 * not propagate properly.
2565 	 */
2566 	ack_APIC_irq();
2567 
2568 	/*
2569 	 * Tail end of clearing remote IRR bit (either by delivering the EOI
2570 	 * message via io-apic EOI register write or simulating it using
2571 	 * mask+edge followed by unnask+level logic) manually when the
2572 	 * level triggered interrupt is seen as the edge triggered interrupt
2573 	 * at the cpu.
2574 	 */
2575 	if (!(v & (1 << (i & 0x1f)))) {
2576 		atomic_inc(&irq_mis_count);
2577 
2578 		eoi_ioapic_irq(irq, cfg);
2579 	}
2580 
2581 	/* Now we can move and renable the irq */
2582 	if (unlikely(do_unmask_irq)) {
2583 		/* Only migrate the irq if the ack has been received.
2584 		 *
2585 		 * On rare occasions the broadcast level triggered ack gets
2586 		 * delayed going to ioapics, and if we reprogram the
2587 		 * vector while Remote IRR is still set the irq will never
2588 		 * fire again.
2589 		 *
2590 		 * To prevent this scenario we read the Remote IRR bit
2591 		 * of the ioapic.  This has two effects.
2592 		 * - On any sane system the read of the ioapic will
2593 		 *   flush writes (and acks) going to the ioapic from
2594 		 *   this cpu.
2595 		 * - We get to see if the ACK has actually been delivered.
2596 		 *
2597 		 * Based on failed experiments of reprogramming the
2598 		 * ioapic entry from outside of irq context starting
2599 		 * with masking the ioapic entry and then polling until
2600 		 * Remote IRR was clear before reprogramming the
2601 		 * ioapic I don't trust the Remote IRR bit to be
2602 		 * completey accurate.
2603 		 *
2604 		 * However there appears to be no other way to plug
2605 		 * this race, so if the Remote IRR bit is not
2606 		 * accurate and is causing problems then it is a hardware bug
2607 		 * and you can go talk to the chipset vendor about it.
2608 		 */
2609 		if (!io_apic_level_ack_pending(cfg))
2610 			irq_move_masked_irq(data);
2611 		unmask_ioapic(cfg);
2612 	}
2613 }
2614 
2615 #ifdef CONFIG_IRQ_REMAP
2616 static void ir_ack_apic_edge(struct irq_data *data)
2617 {
2618 	ack_APIC_irq();
2619 }
2620 
2621 static void ir_ack_apic_level(struct irq_data *data)
2622 {
2623 	ack_APIC_irq();
2624 	eoi_ioapic_irq(data->irq, data->chip_data);
2625 }
2626 
2627 static void ir_print_prefix(struct irq_data *data, struct seq_file *p)
2628 {
2629 	seq_printf(p, " IR-%s", data->chip->name);
2630 }
2631 
2632 static void irq_remap_modify_chip_defaults(struct irq_chip *chip)
2633 {
2634 	chip->irq_print_chip = ir_print_prefix;
2635 	chip->irq_ack = ir_ack_apic_edge;
2636 	chip->irq_eoi = ir_ack_apic_level;
2637 
2638 #ifdef CONFIG_SMP
2639 	chip->irq_set_affinity = ir_ioapic_set_affinity;
2640 #endif
2641 }
2642 #endif /* CONFIG_IRQ_REMAP */
2643 
2644 static struct irq_chip ioapic_chip __read_mostly = {
2645 	.name			= "IO-APIC",
2646 	.irq_startup		= startup_ioapic_irq,
2647 	.irq_mask		= mask_ioapic_irq,
2648 	.irq_unmask		= unmask_ioapic_irq,
2649 	.irq_ack		= ack_apic_edge,
2650 	.irq_eoi		= ack_apic_level,
2651 #ifdef CONFIG_SMP
2652 	.irq_set_affinity	= ioapic_set_affinity,
2653 #endif
2654 	.irq_retrigger		= ioapic_retrigger_irq,
2655 };
2656 
2657 static inline void init_IO_APIC_traps(void)
2658 {
2659 	struct irq_cfg *cfg;
2660 	unsigned int irq;
2661 
2662 	/*
2663 	 * NOTE! The local APIC isn't very good at handling
2664 	 * multiple interrupts at the same interrupt level.
2665 	 * As the interrupt level is determined by taking the
2666 	 * vector number and shifting that right by 4, we
2667 	 * want to spread these out a bit so that they don't
2668 	 * all fall in the same interrupt level.
2669 	 *
2670 	 * Also, we've got to be careful not to trash gate
2671 	 * 0x80, because int 0x80 is hm, kind of importantish. ;)
2672 	 */
2673 	for_each_active_irq(irq) {
2674 		cfg = irq_get_chip_data(irq);
2675 		if (IO_APIC_IRQ(irq) && cfg && !cfg->vector) {
2676 			/*
2677 			 * Hmm.. We don't have an entry for this,
2678 			 * so default to an old-fashioned 8259
2679 			 * interrupt if we can..
2680 			 */
2681 			if (irq < legacy_pic->nr_legacy_irqs)
2682 				legacy_pic->make_irq(irq);
2683 			else
2684 				/* Strange. Oh, well.. */
2685 				irq_set_chip(irq, &no_irq_chip);
2686 		}
2687 	}
2688 }
2689 
2690 /*
2691  * The local APIC irq-chip implementation:
2692  */
2693 
2694 static void mask_lapic_irq(struct irq_data *data)
2695 {
2696 	unsigned long v;
2697 
2698 	v = apic_read(APIC_LVT0);
2699 	apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
2700 }
2701 
2702 static void unmask_lapic_irq(struct irq_data *data)
2703 {
2704 	unsigned long v;
2705 
2706 	v = apic_read(APIC_LVT0);
2707 	apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
2708 }
2709 
2710 static void ack_lapic_irq(struct irq_data *data)
2711 {
2712 	ack_APIC_irq();
2713 }
2714 
2715 static struct irq_chip lapic_chip __read_mostly = {
2716 	.name		= "local-APIC",
2717 	.irq_mask	= mask_lapic_irq,
2718 	.irq_unmask	= unmask_lapic_irq,
2719 	.irq_ack	= ack_lapic_irq,
2720 };
2721 
2722 static void lapic_register_intr(int irq)
2723 {
2724 	irq_clear_status_flags(irq, IRQ_LEVEL);
2725 	irq_set_chip_and_handler_name(irq, &lapic_chip, handle_edge_irq,
2726 				      "edge");
2727 }
2728 
2729 /*
2730  * This looks a bit hackish but it's about the only one way of sending
2731  * a few INTA cycles to 8259As and any associated glue logic.  ICR does
2732  * not support the ExtINT mode, unfortunately.  We need to send these
2733  * cycles as some i82489DX-based boards have glue logic that keeps the
2734  * 8259A interrupt line asserted until INTA.  --macro
2735  */
2736 static inline void __init unlock_ExtINT_logic(void)
2737 {
2738 	int apic, pin, i;
2739 	struct IO_APIC_route_entry entry0, entry1;
2740 	unsigned char save_control, save_freq_select;
2741 
2742 	pin  = find_isa_irq_pin(8, mp_INT);
2743 	if (pin == -1) {
2744 		WARN_ON_ONCE(1);
2745 		return;
2746 	}
2747 	apic = find_isa_irq_apic(8, mp_INT);
2748 	if (apic == -1) {
2749 		WARN_ON_ONCE(1);
2750 		return;
2751 	}
2752 
2753 	entry0 = ioapic_read_entry(apic, pin);
2754 	clear_IO_APIC_pin(apic, pin);
2755 
2756 	memset(&entry1, 0, sizeof(entry1));
2757 
2758 	entry1.dest_mode = 0;			/* physical delivery */
2759 	entry1.mask = 0;			/* unmask IRQ now */
2760 	entry1.dest = hard_smp_processor_id();
2761 	entry1.delivery_mode = dest_ExtINT;
2762 	entry1.polarity = entry0.polarity;
2763 	entry1.trigger = 0;
2764 	entry1.vector = 0;
2765 
2766 	ioapic_write_entry(apic, pin, entry1);
2767 
2768 	save_control = CMOS_READ(RTC_CONTROL);
2769 	save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
2770 	CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
2771 		   RTC_FREQ_SELECT);
2772 	CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
2773 
2774 	i = 100;
2775 	while (i-- > 0) {
2776 		mdelay(10);
2777 		if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
2778 			i -= 10;
2779 	}
2780 
2781 	CMOS_WRITE(save_control, RTC_CONTROL);
2782 	CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
2783 	clear_IO_APIC_pin(apic, pin);
2784 
2785 	ioapic_write_entry(apic, pin, entry0);
2786 }
2787 
2788 static int disable_timer_pin_1 __initdata;
2789 /* Actually the next is obsolete, but keep it for paranoid reasons -AK */
2790 static int __init disable_timer_pin_setup(char *arg)
2791 {
2792 	disable_timer_pin_1 = 1;
2793 	return 0;
2794 }
2795 early_param("disable_timer_pin_1", disable_timer_pin_setup);
2796 
2797 int timer_through_8259 __initdata;
2798 
2799 /*
2800  * This code may look a bit paranoid, but it's supposed to cooperate with
2801  * a wide range of boards and BIOS bugs.  Fortunately only the timer IRQ
2802  * is so screwy.  Thanks to Brian Perkins for testing/hacking this beast
2803  * fanatically on his truly buggy board.
2804  *
2805  * FIXME: really need to revamp this for all platforms.
2806  */
2807 static inline void __init check_timer(void)
2808 {
2809 	struct irq_cfg *cfg = irq_get_chip_data(0);
2810 	int node = cpu_to_node(0);
2811 	int apic1, pin1, apic2, pin2;
2812 	unsigned long flags;
2813 	int no_pin1 = 0;
2814 
2815 	local_irq_save(flags);
2816 
2817 	/*
2818 	 * get/set the timer IRQ vector:
2819 	 */
2820 	legacy_pic->mask(0);
2821 	assign_irq_vector(0, cfg, apic->target_cpus());
2822 
2823 	/*
2824 	 * As IRQ0 is to be enabled in the 8259A, the virtual
2825 	 * wire has to be disabled in the local APIC.  Also
2826 	 * timer interrupts need to be acknowledged manually in
2827 	 * the 8259A for the i82489DX when using the NMI
2828 	 * watchdog as that APIC treats NMIs as level-triggered.
2829 	 * The AEOI mode will finish them in the 8259A
2830 	 * automatically.
2831 	 */
2832 	apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
2833 	legacy_pic->init(1);
2834 
2835 	pin1  = find_isa_irq_pin(0, mp_INT);
2836 	apic1 = find_isa_irq_apic(0, mp_INT);
2837 	pin2  = ioapic_i8259.pin;
2838 	apic2 = ioapic_i8259.apic;
2839 
2840 	apic_printk(APIC_QUIET, KERN_INFO "..TIMER: vector=0x%02X "
2841 		    "apic1=%d pin1=%d apic2=%d pin2=%d\n",
2842 		    cfg->vector, apic1, pin1, apic2, pin2);
2843 
2844 	/*
2845 	 * Some BIOS writers are clueless and report the ExtINTA
2846 	 * I/O APIC input from the cascaded 8259A as the timer
2847 	 * interrupt input.  So just in case, if only one pin
2848 	 * was found above, try it both directly and through the
2849 	 * 8259A.
2850 	 */
2851 	if (pin1 == -1) {
2852 		if (intr_remapping_enabled)
2853 			panic("BIOS bug: timer not connected to IO-APIC");
2854 		pin1 = pin2;
2855 		apic1 = apic2;
2856 		no_pin1 = 1;
2857 	} else if (pin2 == -1) {
2858 		pin2 = pin1;
2859 		apic2 = apic1;
2860 	}
2861 
2862 	if (pin1 != -1) {
2863 		/*
2864 		 * Ok, does IRQ0 through the IOAPIC work?
2865 		 */
2866 		if (no_pin1) {
2867 			add_pin_to_irq_node(cfg, node, apic1, pin1);
2868 			setup_timer_IRQ0_pin(apic1, pin1, cfg->vector);
2869 		} else {
2870 			/* for edge trigger, setup_ioapic_irq already
2871 			 * leave it unmasked.
2872 			 * so only need to unmask if it is level-trigger
2873 			 * do we really have level trigger timer?
2874 			 */
2875 			int idx;
2876 			idx = find_irq_entry(apic1, pin1, mp_INT);
2877 			if (idx != -1 && irq_trigger(idx))
2878 				unmask_ioapic(cfg);
2879 		}
2880 		if (timer_irq_works()) {
2881 			if (disable_timer_pin_1 > 0)
2882 				clear_IO_APIC_pin(0, pin1);
2883 			goto out;
2884 		}
2885 		if (intr_remapping_enabled)
2886 			panic("timer doesn't work through Interrupt-remapped IO-APIC");
2887 		local_irq_disable();
2888 		clear_IO_APIC_pin(apic1, pin1);
2889 		if (!no_pin1)
2890 			apic_printk(APIC_QUIET, KERN_ERR "..MP-BIOS bug: "
2891 				    "8254 timer not connected to IO-APIC\n");
2892 
2893 		apic_printk(APIC_QUIET, KERN_INFO "...trying to set up timer "
2894 			    "(IRQ0) through the 8259A ...\n");
2895 		apic_printk(APIC_QUIET, KERN_INFO
2896 			    "..... (found apic %d pin %d) ...\n", apic2, pin2);
2897 		/*
2898 		 * legacy devices should be connected to IO APIC #0
2899 		 */
2900 		replace_pin_at_irq_node(cfg, node, apic1, pin1, apic2, pin2);
2901 		setup_timer_IRQ0_pin(apic2, pin2, cfg->vector);
2902 		legacy_pic->unmask(0);
2903 		if (timer_irq_works()) {
2904 			apic_printk(APIC_QUIET, KERN_INFO "....... works.\n");
2905 			timer_through_8259 = 1;
2906 			goto out;
2907 		}
2908 		/*
2909 		 * Cleanup, just in case ...
2910 		 */
2911 		local_irq_disable();
2912 		legacy_pic->mask(0);
2913 		clear_IO_APIC_pin(apic2, pin2);
2914 		apic_printk(APIC_QUIET, KERN_INFO "....... failed.\n");
2915 	}
2916 
2917 	apic_printk(APIC_QUIET, KERN_INFO
2918 		    "...trying to set up timer as Virtual Wire IRQ...\n");
2919 
2920 	lapic_register_intr(0);
2921 	apic_write(APIC_LVT0, APIC_DM_FIXED | cfg->vector);	/* Fixed mode */
2922 	legacy_pic->unmask(0);
2923 
2924 	if (timer_irq_works()) {
2925 		apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
2926 		goto out;
2927 	}
2928 	local_irq_disable();
2929 	legacy_pic->mask(0);
2930 	apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | cfg->vector);
2931 	apic_printk(APIC_QUIET, KERN_INFO "..... failed.\n");
2932 
2933 	apic_printk(APIC_QUIET, KERN_INFO
2934 		    "...trying to set up timer as ExtINT IRQ...\n");
2935 
2936 	legacy_pic->init(0);
2937 	legacy_pic->make_irq(0);
2938 	apic_write(APIC_LVT0, APIC_DM_EXTINT);
2939 
2940 	unlock_ExtINT_logic();
2941 
2942 	if (timer_irq_works()) {
2943 		apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
2944 		goto out;
2945 	}
2946 	local_irq_disable();
2947 	apic_printk(APIC_QUIET, KERN_INFO "..... failed :(.\n");
2948 	panic("IO-APIC + timer doesn't work!  Boot with apic=debug and send a "
2949 		"report.  Then try booting with the 'noapic' option.\n");
2950 out:
2951 	local_irq_restore(flags);
2952 }
2953 
2954 /*
2955  * Traditionally ISA IRQ2 is the cascade IRQ, and is not available
2956  * to devices.  However there may be an I/O APIC pin available for
2957  * this interrupt regardless.  The pin may be left unconnected, but
2958  * typically it will be reused as an ExtINT cascade interrupt for
2959  * the master 8259A.  In the MPS case such a pin will normally be
2960  * reported as an ExtINT interrupt in the MP table.  With ACPI
2961  * there is no provision for ExtINT interrupts, and in the absence
2962  * of an override it would be treated as an ordinary ISA I/O APIC
2963  * interrupt, that is edge-triggered and unmasked by default.  We
2964  * used to do this, but it caused problems on some systems because
2965  * of the NMI watchdog and sometimes IRQ0 of the 8254 timer using
2966  * the same ExtINT cascade interrupt to drive the local APIC of the
2967  * bootstrap processor.  Therefore we refrain from routing IRQ2 to
2968  * the I/O APIC in all cases now.  No actual device should request
2969  * it anyway.  --macro
2970  */
2971 #define PIC_IRQS	(1UL << PIC_CASCADE_IR)
2972 
2973 void __init setup_IO_APIC(void)
2974 {
2975 
2976 	/*
2977 	 * calling enable_IO_APIC() is moved to setup_local_APIC for BP
2978 	 */
2979 	io_apic_irqs = legacy_pic->nr_legacy_irqs ? ~PIC_IRQS : ~0UL;
2980 
2981 	apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
2982 	/*
2983          * Set up IO-APIC IRQ routing.
2984          */
2985 	x86_init.mpparse.setup_ioapic_ids();
2986 
2987 	sync_Arb_IDs();
2988 	setup_IO_APIC_irqs();
2989 	init_IO_APIC_traps();
2990 	if (legacy_pic->nr_legacy_irqs)
2991 		check_timer();
2992 }
2993 
2994 /*
2995  *      Called after all the initialization is done. If we didn't find any
2996  *      APIC bugs then we can allow the modify fast path
2997  */
2998 
2999 static int __init io_apic_bug_finalize(void)
3000 {
3001 	if (sis_apic_bug == -1)
3002 		sis_apic_bug = 0;
3003 	return 0;
3004 }
3005 
3006 late_initcall(io_apic_bug_finalize);
3007 
3008 static void resume_ioapic_id(int ioapic_idx)
3009 {
3010 	unsigned long flags;
3011 	union IO_APIC_reg_00 reg_00;
3012 
3013 	raw_spin_lock_irqsave(&ioapic_lock, flags);
3014 	reg_00.raw = io_apic_read(ioapic_idx, 0);
3015 	if (reg_00.bits.ID != mpc_ioapic_id(ioapic_idx)) {
3016 		reg_00.bits.ID = mpc_ioapic_id(ioapic_idx);
3017 		io_apic_write(ioapic_idx, 0, reg_00.raw);
3018 	}
3019 	raw_spin_unlock_irqrestore(&ioapic_lock, flags);
3020 }
3021 
3022 static void ioapic_resume(void)
3023 {
3024 	int ioapic_idx;
3025 
3026 	for (ioapic_idx = nr_ioapics - 1; ioapic_idx >= 0; ioapic_idx--)
3027 		resume_ioapic_id(ioapic_idx);
3028 
3029 	restore_ioapic_entries();
3030 }
3031 
3032 static struct syscore_ops ioapic_syscore_ops = {
3033 	.suspend = save_ioapic_entries,
3034 	.resume = ioapic_resume,
3035 };
3036 
3037 static int __init ioapic_init_ops(void)
3038 {
3039 	register_syscore_ops(&ioapic_syscore_ops);
3040 
3041 	return 0;
3042 }
3043 
3044 device_initcall(ioapic_init_ops);
3045 
3046 /*
3047  * Dynamic irq allocate and deallocation
3048  */
3049 unsigned int create_irq_nr(unsigned int from, int node)
3050 {
3051 	struct irq_cfg *cfg;
3052 	unsigned long flags;
3053 	unsigned int ret = 0;
3054 	int irq;
3055 
3056 	if (from < nr_irqs_gsi)
3057 		from = nr_irqs_gsi;
3058 
3059 	irq = alloc_irq_from(from, node);
3060 	if (irq < 0)
3061 		return 0;
3062 	cfg = alloc_irq_cfg(irq, node);
3063 	if (!cfg) {
3064 		free_irq_at(irq, NULL);
3065 		return 0;
3066 	}
3067 
3068 	raw_spin_lock_irqsave(&vector_lock, flags);
3069 	if (!__assign_irq_vector(irq, cfg, apic->target_cpus()))
3070 		ret = irq;
3071 	raw_spin_unlock_irqrestore(&vector_lock, flags);
3072 
3073 	if (ret) {
3074 		irq_set_chip_data(irq, cfg);
3075 		irq_clear_status_flags(irq, IRQ_NOREQUEST);
3076 	} else {
3077 		free_irq_at(irq, cfg);
3078 	}
3079 	return ret;
3080 }
3081 
3082 int create_irq(void)
3083 {
3084 	int node = cpu_to_node(0);
3085 	unsigned int irq_want;
3086 	int irq;
3087 
3088 	irq_want = nr_irqs_gsi;
3089 	irq = create_irq_nr(irq_want, node);
3090 
3091 	if (irq == 0)
3092 		irq = -1;
3093 
3094 	return irq;
3095 }
3096 
3097 void destroy_irq(unsigned int irq)
3098 {
3099 	struct irq_cfg *cfg = irq_get_chip_data(irq);
3100 	unsigned long flags;
3101 
3102 	irq_set_status_flags(irq, IRQ_NOREQUEST|IRQ_NOPROBE);
3103 
3104 	if (irq_remapped(cfg))
3105 		free_irte(irq);
3106 	raw_spin_lock_irqsave(&vector_lock, flags);
3107 	__clear_irq_vector(irq, cfg);
3108 	raw_spin_unlock_irqrestore(&vector_lock, flags);
3109 	free_irq_at(irq, cfg);
3110 }
3111 
3112 /*
3113  * MSI message composition
3114  */
3115 #ifdef CONFIG_PCI_MSI
3116 static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq,
3117 			   struct msi_msg *msg, u8 hpet_id)
3118 {
3119 	struct irq_cfg *cfg;
3120 	int err;
3121 	unsigned dest;
3122 
3123 	if (disable_apic)
3124 		return -ENXIO;
3125 
3126 	cfg = irq_cfg(irq);
3127 	err = assign_irq_vector(irq, cfg, apic->target_cpus());
3128 	if (err)
3129 		return err;
3130 
3131 	dest = apic->cpu_mask_to_apicid_and(cfg->domain, apic->target_cpus());
3132 
3133 	if (irq_remapped(cfg)) {
3134 		struct irte irte;
3135 		int ir_index;
3136 		u16 sub_handle;
3137 
3138 		ir_index = map_irq_to_irte_handle(irq, &sub_handle);
3139 		BUG_ON(ir_index == -1);
3140 
3141 		prepare_irte(&irte, cfg->vector, dest);
3142 
3143 		/* Set source-id of interrupt request */
3144 		if (pdev)
3145 			set_msi_sid(&irte, pdev);
3146 		else
3147 			set_hpet_sid(&irte, hpet_id);
3148 
3149 		modify_irte(irq, &irte);
3150 
3151 		msg->address_hi = MSI_ADDR_BASE_HI;
3152 		msg->data = sub_handle;
3153 		msg->address_lo = MSI_ADDR_BASE_LO | MSI_ADDR_IR_EXT_INT |
3154 				  MSI_ADDR_IR_SHV |
3155 				  MSI_ADDR_IR_INDEX1(ir_index) |
3156 				  MSI_ADDR_IR_INDEX2(ir_index);
3157 	} else {
3158 		if (x2apic_enabled())
3159 			msg->address_hi = MSI_ADDR_BASE_HI |
3160 					  MSI_ADDR_EXT_DEST_ID(dest);
3161 		else
3162 			msg->address_hi = MSI_ADDR_BASE_HI;
3163 
3164 		msg->address_lo =
3165 			MSI_ADDR_BASE_LO |
3166 			((apic->irq_dest_mode == 0) ?
3167 				MSI_ADDR_DEST_MODE_PHYSICAL:
3168 				MSI_ADDR_DEST_MODE_LOGICAL) |
3169 			((apic->irq_delivery_mode != dest_LowestPrio) ?
3170 				MSI_ADDR_REDIRECTION_CPU:
3171 				MSI_ADDR_REDIRECTION_LOWPRI) |
3172 			MSI_ADDR_DEST_ID(dest);
3173 
3174 		msg->data =
3175 			MSI_DATA_TRIGGER_EDGE |
3176 			MSI_DATA_LEVEL_ASSERT |
3177 			((apic->irq_delivery_mode != dest_LowestPrio) ?
3178 				MSI_DATA_DELIVERY_FIXED:
3179 				MSI_DATA_DELIVERY_LOWPRI) |
3180 			MSI_DATA_VECTOR(cfg->vector);
3181 	}
3182 	return err;
3183 }
3184 
3185 #ifdef CONFIG_SMP
3186 static int
3187 msi_set_affinity(struct irq_data *data, const struct cpumask *mask, bool force)
3188 {
3189 	struct irq_cfg *cfg = data->chip_data;
3190 	struct msi_msg msg;
3191 	unsigned int dest;
3192 
3193 	if (__ioapic_set_affinity(data, mask, &dest))
3194 		return -1;
3195 
3196 	__get_cached_msi_msg(data->msi_desc, &msg);
3197 
3198 	msg.data &= ~MSI_DATA_VECTOR_MASK;
3199 	msg.data |= MSI_DATA_VECTOR(cfg->vector);
3200 	msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3201 	msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3202 
3203 	__write_msi_msg(data->msi_desc, &msg);
3204 
3205 	return 0;
3206 }
3207 #endif /* CONFIG_SMP */
3208 
3209 /*
3210  * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
3211  * which implement the MSI or MSI-X Capability Structure.
3212  */
3213 static struct irq_chip msi_chip = {
3214 	.name			= "PCI-MSI",
3215 	.irq_unmask		= unmask_msi_irq,
3216 	.irq_mask		= mask_msi_irq,
3217 	.irq_ack		= ack_apic_edge,
3218 #ifdef CONFIG_SMP
3219 	.irq_set_affinity	= msi_set_affinity,
3220 #endif
3221 	.irq_retrigger		= ioapic_retrigger_irq,
3222 };
3223 
3224 /*
3225  * Map the PCI dev to the corresponding remapping hardware unit
3226  * and allocate 'nvec' consecutive interrupt-remapping table entries
3227  * in it.
3228  */
3229 static int msi_alloc_irte(struct pci_dev *dev, int irq, int nvec)
3230 {
3231 	struct intel_iommu *iommu;
3232 	int index;
3233 
3234 	iommu = map_dev_to_ir(dev);
3235 	if (!iommu) {
3236 		printk(KERN_ERR
3237 		       "Unable to map PCI %s to iommu\n", pci_name(dev));
3238 		return -ENOENT;
3239 	}
3240 
3241 	index = alloc_irte(iommu, irq, nvec);
3242 	if (index < 0) {
3243 		printk(KERN_ERR
3244 		       "Unable to allocate %d IRTE for PCI %s\n", nvec,
3245 		       pci_name(dev));
3246 		return -ENOSPC;
3247 	}
3248 	return index;
3249 }
3250 
3251 static int setup_msi_irq(struct pci_dev *dev, struct msi_desc *msidesc, int irq)
3252 {
3253 	struct irq_chip *chip = &msi_chip;
3254 	struct msi_msg msg;
3255 	int ret;
3256 
3257 	ret = msi_compose_msg(dev, irq, &msg, -1);
3258 	if (ret < 0)
3259 		return ret;
3260 
3261 	irq_set_msi_desc(irq, msidesc);
3262 	write_msi_msg(irq, &msg);
3263 
3264 	if (irq_remapped(irq_get_chip_data(irq))) {
3265 		irq_set_status_flags(irq, IRQ_MOVE_PCNTXT);
3266 		irq_remap_modify_chip_defaults(chip);
3267 	}
3268 
3269 	irq_set_chip_and_handler_name(irq, chip, handle_edge_irq, "edge");
3270 
3271 	dev_printk(KERN_DEBUG, &dev->dev, "irq %d for MSI/MSI-X\n", irq);
3272 
3273 	return 0;
3274 }
3275 
3276 int native_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
3277 {
3278 	int node, ret, sub_handle, index = 0;
3279 	unsigned int irq, irq_want;
3280 	struct msi_desc *msidesc;
3281 	struct intel_iommu *iommu = NULL;
3282 
3283 	/* x86 doesn't support multiple MSI yet */
3284 	if (type == PCI_CAP_ID_MSI && nvec > 1)
3285 		return 1;
3286 
3287 	node = dev_to_node(&dev->dev);
3288 	irq_want = nr_irqs_gsi;
3289 	sub_handle = 0;
3290 	list_for_each_entry(msidesc, &dev->msi_list, list) {
3291 		irq = create_irq_nr(irq_want, node);
3292 		if (irq == 0)
3293 			return -1;
3294 		irq_want = irq + 1;
3295 		if (!intr_remapping_enabled)
3296 			goto no_ir;
3297 
3298 		if (!sub_handle) {
3299 			/*
3300 			 * allocate the consecutive block of IRTE's
3301 			 * for 'nvec'
3302 			 */
3303 			index = msi_alloc_irte(dev, irq, nvec);
3304 			if (index < 0) {
3305 				ret = index;
3306 				goto error;
3307 			}
3308 		} else {
3309 			iommu = map_dev_to_ir(dev);
3310 			if (!iommu) {
3311 				ret = -ENOENT;
3312 				goto error;
3313 			}
3314 			/*
3315 			 * setup the mapping between the irq and the IRTE
3316 			 * base index, the sub_handle pointing to the
3317 			 * appropriate interrupt remap table entry.
3318 			 */
3319 			set_irte_irq(irq, iommu, index, sub_handle);
3320 		}
3321 no_ir:
3322 		ret = setup_msi_irq(dev, msidesc, irq);
3323 		if (ret < 0)
3324 			goto error;
3325 		sub_handle++;
3326 	}
3327 	return 0;
3328 
3329 error:
3330 	destroy_irq(irq);
3331 	return ret;
3332 }
3333 
3334 void native_teardown_msi_irq(unsigned int irq)
3335 {
3336 	destroy_irq(irq);
3337 }
3338 
3339 #ifdef CONFIG_DMAR_TABLE
3340 #ifdef CONFIG_SMP
3341 static int
3342 dmar_msi_set_affinity(struct irq_data *data, const struct cpumask *mask,
3343 		      bool force)
3344 {
3345 	struct irq_cfg *cfg = data->chip_data;
3346 	unsigned int dest, irq = data->irq;
3347 	struct msi_msg msg;
3348 
3349 	if (__ioapic_set_affinity(data, mask, &dest))
3350 		return -1;
3351 
3352 	dmar_msi_read(irq, &msg);
3353 
3354 	msg.data &= ~MSI_DATA_VECTOR_MASK;
3355 	msg.data |= MSI_DATA_VECTOR(cfg->vector);
3356 	msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3357 	msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3358 	msg.address_hi = MSI_ADDR_BASE_HI | MSI_ADDR_EXT_DEST_ID(dest);
3359 
3360 	dmar_msi_write(irq, &msg);
3361 
3362 	return 0;
3363 }
3364 
3365 #endif /* CONFIG_SMP */
3366 
3367 static struct irq_chip dmar_msi_type = {
3368 	.name			= "DMAR_MSI",
3369 	.irq_unmask		= dmar_msi_unmask,
3370 	.irq_mask		= dmar_msi_mask,
3371 	.irq_ack		= ack_apic_edge,
3372 #ifdef CONFIG_SMP
3373 	.irq_set_affinity	= dmar_msi_set_affinity,
3374 #endif
3375 	.irq_retrigger		= ioapic_retrigger_irq,
3376 };
3377 
3378 int arch_setup_dmar_msi(unsigned int irq)
3379 {
3380 	int ret;
3381 	struct msi_msg msg;
3382 
3383 	ret = msi_compose_msg(NULL, irq, &msg, -1);
3384 	if (ret < 0)
3385 		return ret;
3386 	dmar_msi_write(irq, &msg);
3387 	irq_set_chip_and_handler_name(irq, &dmar_msi_type, handle_edge_irq,
3388 				      "edge");
3389 	return 0;
3390 }
3391 #endif
3392 
3393 #ifdef CONFIG_HPET_TIMER
3394 
3395 #ifdef CONFIG_SMP
3396 static int hpet_msi_set_affinity(struct irq_data *data,
3397 				 const struct cpumask *mask, bool force)
3398 {
3399 	struct irq_cfg *cfg = data->chip_data;
3400 	struct msi_msg msg;
3401 	unsigned int dest;
3402 
3403 	if (__ioapic_set_affinity(data, mask, &dest))
3404 		return -1;
3405 
3406 	hpet_msi_read(data->handler_data, &msg);
3407 
3408 	msg.data &= ~MSI_DATA_VECTOR_MASK;
3409 	msg.data |= MSI_DATA_VECTOR(cfg->vector);
3410 	msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3411 	msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3412 
3413 	hpet_msi_write(data->handler_data, &msg);
3414 
3415 	return 0;
3416 }
3417 
3418 #endif /* CONFIG_SMP */
3419 
3420 static struct irq_chip hpet_msi_type = {
3421 	.name = "HPET_MSI",
3422 	.irq_unmask = hpet_msi_unmask,
3423 	.irq_mask = hpet_msi_mask,
3424 	.irq_ack = ack_apic_edge,
3425 #ifdef CONFIG_SMP
3426 	.irq_set_affinity = hpet_msi_set_affinity,
3427 #endif
3428 	.irq_retrigger = ioapic_retrigger_irq,
3429 };
3430 
3431 int arch_setup_hpet_msi(unsigned int irq, unsigned int id)
3432 {
3433 	struct irq_chip *chip = &hpet_msi_type;
3434 	struct msi_msg msg;
3435 	int ret;
3436 
3437 	if (intr_remapping_enabled) {
3438 		struct intel_iommu *iommu = map_hpet_to_ir(id);
3439 		int index;
3440 
3441 		if (!iommu)
3442 			return -1;
3443 
3444 		index = alloc_irte(iommu, irq, 1);
3445 		if (index < 0)
3446 			return -1;
3447 	}
3448 
3449 	ret = msi_compose_msg(NULL, irq, &msg, id);
3450 	if (ret < 0)
3451 		return ret;
3452 
3453 	hpet_msi_write(irq_get_handler_data(irq), &msg);
3454 	irq_set_status_flags(irq, IRQ_MOVE_PCNTXT);
3455 	if (irq_remapped(irq_get_chip_data(irq)))
3456 		irq_remap_modify_chip_defaults(chip);
3457 
3458 	irq_set_chip_and_handler_name(irq, chip, handle_edge_irq, "edge");
3459 	return 0;
3460 }
3461 #endif
3462 
3463 #endif /* CONFIG_PCI_MSI */
3464 /*
3465  * Hypertransport interrupt support
3466  */
3467 #ifdef CONFIG_HT_IRQ
3468 
3469 #ifdef CONFIG_SMP
3470 
3471 static void target_ht_irq(unsigned int irq, unsigned int dest, u8 vector)
3472 {
3473 	struct ht_irq_msg msg;
3474 	fetch_ht_irq_msg(irq, &msg);
3475 
3476 	msg.address_lo &= ~(HT_IRQ_LOW_VECTOR_MASK | HT_IRQ_LOW_DEST_ID_MASK);
3477 	msg.address_hi &= ~(HT_IRQ_HIGH_DEST_ID_MASK);
3478 
3479 	msg.address_lo |= HT_IRQ_LOW_VECTOR(vector) | HT_IRQ_LOW_DEST_ID(dest);
3480 	msg.address_hi |= HT_IRQ_HIGH_DEST_ID(dest);
3481 
3482 	write_ht_irq_msg(irq, &msg);
3483 }
3484 
3485 static int
3486 ht_set_affinity(struct irq_data *data, const struct cpumask *mask, bool force)
3487 {
3488 	struct irq_cfg *cfg = data->chip_data;
3489 	unsigned int dest;
3490 
3491 	if (__ioapic_set_affinity(data, mask, &dest))
3492 		return -1;
3493 
3494 	target_ht_irq(data->irq, dest, cfg->vector);
3495 	return 0;
3496 }
3497 
3498 #endif
3499 
3500 static struct irq_chip ht_irq_chip = {
3501 	.name			= "PCI-HT",
3502 	.irq_mask		= mask_ht_irq,
3503 	.irq_unmask		= unmask_ht_irq,
3504 	.irq_ack		= ack_apic_edge,
3505 #ifdef CONFIG_SMP
3506 	.irq_set_affinity	= ht_set_affinity,
3507 #endif
3508 	.irq_retrigger		= ioapic_retrigger_irq,
3509 };
3510 
3511 int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev)
3512 {
3513 	struct irq_cfg *cfg;
3514 	int err;
3515 
3516 	if (disable_apic)
3517 		return -ENXIO;
3518 
3519 	cfg = irq_cfg(irq);
3520 	err = assign_irq_vector(irq, cfg, apic->target_cpus());
3521 	if (!err) {
3522 		struct ht_irq_msg msg;
3523 		unsigned dest;
3524 
3525 		dest = apic->cpu_mask_to_apicid_and(cfg->domain,
3526 						    apic->target_cpus());
3527 
3528 		msg.address_hi = HT_IRQ_HIGH_DEST_ID(dest);
3529 
3530 		msg.address_lo =
3531 			HT_IRQ_LOW_BASE |
3532 			HT_IRQ_LOW_DEST_ID(dest) |
3533 			HT_IRQ_LOW_VECTOR(cfg->vector) |
3534 			((apic->irq_dest_mode == 0) ?
3535 				HT_IRQ_LOW_DM_PHYSICAL :
3536 				HT_IRQ_LOW_DM_LOGICAL) |
3537 			HT_IRQ_LOW_RQEOI_EDGE |
3538 			((apic->irq_delivery_mode != dest_LowestPrio) ?
3539 				HT_IRQ_LOW_MT_FIXED :
3540 				HT_IRQ_LOW_MT_ARBITRATED) |
3541 			HT_IRQ_LOW_IRQ_MASKED;
3542 
3543 		write_ht_irq_msg(irq, &msg);
3544 
3545 		irq_set_chip_and_handler_name(irq, &ht_irq_chip,
3546 					      handle_edge_irq, "edge");
3547 
3548 		dev_printk(KERN_DEBUG, &dev->dev, "irq %d for HT\n", irq);
3549 	}
3550 	return err;
3551 }
3552 #endif /* CONFIG_HT_IRQ */
3553 
3554 static int
3555 io_apic_setup_irq_pin(unsigned int irq, int node, struct io_apic_irq_attr *attr)
3556 {
3557 	struct irq_cfg *cfg = alloc_irq_and_cfg_at(irq, node);
3558 	int ret;
3559 
3560 	if (!cfg)
3561 		return -EINVAL;
3562 	ret = __add_pin_to_irq_node(cfg, node, attr->ioapic, attr->ioapic_pin);
3563 	if (!ret)
3564 		setup_ioapic_irq(irq, cfg, attr);
3565 	return ret;
3566 }
3567 
3568 int io_apic_setup_irq_pin_once(unsigned int irq, int node,
3569 			       struct io_apic_irq_attr *attr)
3570 {
3571 	unsigned int ioapic_idx = attr->ioapic, pin = attr->ioapic_pin;
3572 	int ret;
3573 
3574 	/* Avoid redundant programming */
3575 	if (test_bit(pin, ioapics[ioapic_idx].pin_programmed)) {
3576 		pr_debug("Pin %d-%d already programmed\n",
3577 			 mpc_ioapic_id(ioapic_idx), pin);
3578 		return 0;
3579 	}
3580 	ret = io_apic_setup_irq_pin(irq, node, attr);
3581 	if (!ret)
3582 		set_bit(pin, ioapics[ioapic_idx].pin_programmed);
3583 	return ret;
3584 }
3585 
3586 static int __init io_apic_get_redir_entries(int ioapic)
3587 {
3588 	union IO_APIC_reg_01	reg_01;
3589 	unsigned long flags;
3590 
3591 	raw_spin_lock_irqsave(&ioapic_lock, flags);
3592 	reg_01.raw = io_apic_read(ioapic, 1);
3593 	raw_spin_unlock_irqrestore(&ioapic_lock, flags);
3594 
3595 	/* The register returns the maximum index redir index
3596 	 * supported, which is one less than the total number of redir
3597 	 * entries.
3598 	 */
3599 	return reg_01.bits.entries + 1;
3600 }
3601 
3602 static void __init probe_nr_irqs_gsi(void)
3603 {
3604 	int nr;
3605 
3606 	nr = gsi_top + NR_IRQS_LEGACY;
3607 	if (nr > nr_irqs_gsi)
3608 		nr_irqs_gsi = nr;
3609 
3610 	printk(KERN_DEBUG "nr_irqs_gsi: %d\n", nr_irqs_gsi);
3611 }
3612 
3613 int get_nr_irqs_gsi(void)
3614 {
3615 	return nr_irqs_gsi;
3616 }
3617 
3618 int __init arch_probe_nr_irqs(void)
3619 {
3620 	int nr;
3621 
3622 	if (nr_irqs > (NR_VECTORS * nr_cpu_ids))
3623 		nr_irqs = NR_VECTORS * nr_cpu_ids;
3624 
3625 	nr = nr_irqs_gsi + 8 * nr_cpu_ids;
3626 #if defined(CONFIG_PCI_MSI) || defined(CONFIG_HT_IRQ)
3627 	/*
3628 	 * for MSI and HT dyn irq
3629 	 */
3630 	nr += nr_irqs_gsi * 16;
3631 #endif
3632 	if (nr < nr_irqs)
3633 		nr_irqs = nr;
3634 
3635 	return NR_IRQS_LEGACY;
3636 }
3637 
3638 int io_apic_set_pci_routing(struct device *dev, int irq,
3639 			    struct io_apic_irq_attr *irq_attr)
3640 {
3641 	int node;
3642 
3643 	if (!IO_APIC_IRQ(irq)) {
3644 		apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
3645 			    irq_attr->ioapic);
3646 		return -EINVAL;
3647 	}
3648 
3649 	node = dev ? dev_to_node(dev) : cpu_to_node(0);
3650 
3651 	return io_apic_setup_irq_pin_once(irq, node, irq_attr);
3652 }
3653 
3654 #ifdef CONFIG_X86_32
3655 static int __init io_apic_get_unique_id(int ioapic, int apic_id)
3656 {
3657 	union IO_APIC_reg_00 reg_00;
3658 	static physid_mask_t apic_id_map = PHYSID_MASK_NONE;
3659 	physid_mask_t tmp;
3660 	unsigned long flags;
3661 	int i = 0;
3662 
3663 	/*
3664 	 * The P4 platform supports up to 256 APIC IDs on two separate APIC
3665 	 * buses (one for LAPICs, one for IOAPICs), where predecessors only
3666 	 * supports up to 16 on one shared APIC bus.
3667 	 *
3668 	 * TBD: Expand LAPIC/IOAPIC support on P4-class systems to take full
3669 	 *      advantage of new APIC bus architecture.
3670 	 */
3671 
3672 	if (physids_empty(apic_id_map))
3673 		apic->ioapic_phys_id_map(&phys_cpu_present_map, &apic_id_map);
3674 
3675 	raw_spin_lock_irqsave(&ioapic_lock, flags);
3676 	reg_00.raw = io_apic_read(ioapic, 0);
3677 	raw_spin_unlock_irqrestore(&ioapic_lock, flags);
3678 
3679 	if (apic_id >= get_physical_broadcast()) {
3680 		printk(KERN_WARNING "IOAPIC[%d]: Invalid apic_id %d, trying "
3681 			"%d\n", ioapic, apic_id, reg_00.bits.ID);
3682 		apic_id = reg_00.bits.ID;
3683 	}
3684 
3685 	/*
3686 	 * Every APIC in a system must have a unique ID or we get lots of nice
3687 	 * 'stuck on smp_invalidate_needed IPI wait' messages.
3688 	 */
3689 	if (apic->check_apicid_used(&apic_id_map, apic_id)) {
3690 
3691 		for (i = 0; i < get_physical_broadcast(); i++) {
3692 			if (!apic->check_apicid_used(&apic_id_map, i))
3693 				break;
3694 		}
3695 
3696 		if (i == get_physical_broadcast())
3697 			panic("Max apic_id exceeded!\n");
3698 
3699 		printk(KERN_WARNING "IOAPIC[%d]: apic_id %d already used, "
3700 			"trying %d\n", ioapic, apic_id, i);
3701 
3702 		apic_id = i;
3703 	}
3704 
3705 	apic->apicid_to_cpu_present(apic_id, &tmp);
3706 	physids_or(apic_id_map, apic_id_map, tmp);
3707 
3708 	if (reg_00.bits.ID != apic_id) {
3709 		reg_00.bits.ID = apic_id;
3710 
3711 		raw_spin_lock_irqsave(&ioapic_lock, flags);
3712 		io_apic_write(ioapic, 0, reg_00.raw);
3713 		reg_00.raw = io_apic_read(ioapic, 0);
3714 		raw_spin_unlock_irqrestore(&ioapic_lock, flags);
3715 
3716 		/* Sanity check */
3717 		if (reg_00.bits.ID != apic_id) {
3718 			printk("IOAPIC[%d]: Unable to change apic_id!\n", ioapic);
3719 			return -1;
3720 		}
3721 	}
3722 
3723 	apic_printk(APIC_VERBOSE, KERN_INFO
3724 			"IOAPIC[%d]: Assigned apic_id %d\n", ioapic, apic_id);
3725 
3726 	return apic_id;
3727 }
3728 
3729 static u8 __init io_apic_unique_id(u8 id)
3730 {
3731 	if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
3732 	    !APIC_XAPIC(apic_version[boot_cpu_physical_apicid]))
3733 		return io_apic_get_unique_id(nr_ioapics, id);
3734 	else
3735 		return id;
3736 }
3737 #else
3738 static u8 __init io_apic_unique_id(u8 id)
3739 {
3740 	int i;
3741 	DECLARE_BITMAP(used, 256);
3742 
3743 	bitmap_zero(used, 256);
3744 	for (i = 0; i < nr_ioapics; i++) {
3745 		__set_bit(mpc_ioapic_id(i), used);
3746 	}
3747 	if (!test_bit(id, used))
3748 		return id;
3749 	return find_first_zero_bit(used, 256);
3750 }
3751 #endif
3752 
3753 static int __init io_apic_get_version(int ioapic)
3754 {
3755 	union IO_APIC_reg_01	reg_01;
3756 	unsigned long flags;
3757 
3758 	raw_spin_lock_irqsave(&ioapic_lock, flags);
3759 	reg_01.raw = io_apic_read(ioapic, 1);
3760 	raw_spin_unlock_irqrestore(&ioapic_lock, flags);
3761 
3762 	return reg_01.bits.version;
3763 }
3764 
3765 int acpi_get_override_irq(u32 gsi, int *trigger, int *polarity)
3766 {
3767 	int ioapic, pin, idx;
3768 
3769 	if (skip_ioapic_setup)
3770 		return -1;
3771 
3772 	ioapic = mp_find_ioapic(gsi);
3773 	if (ioapic < 0)
3774 		return -1;
3775 
3776 	pin = mp_find_ioapic_pin(ioapic, gsi);
3777 	if (pin < 0)
3778 		return -1;
3779 
3780 	idx = find_irq_entry(ioapic, pin, mp_INT);
3781 	if (idx < 0)
3782 		return -1;
3783 
3784 	*trigger = irq_trigger(idx);
3785 	*polarity = irq_polarity(idx);
3786 	return 0;
3787 }
3788 
3789 /*
3790  * This function currently is only a helper for the i386 smp boot process where
3791  * we need to reprogram the ioredtbls to cater for the cpus which have come online
3792  * so mask in all cases should simply be apic->target_cpus()
3793  */
3794 #ifdef CONFIG_SMP
3795 void __init setup_ioapic_dest(void)
3796 {
3797 	int pin, ioapic, irq, irq_entry;
3798 	const struct cpumask *mask;
3799 	struct irq_data *idata;
3800 
3801 	if (skip_ioapic_setup == 1)
3802 		return;
3803 
3804 	for (ioapic = 0; ioapic < nr_ioapics; ioapic++)
3805 	for (pin = 0; pin < ioapics[ioapic].nr_registers; pin++) {
3806 		irq_entry = find_irq_entry(ioapic, pin, mp_INT);
3807 		if (irq_entry == -1)
3808 			continue;
3809 		irq = pin_2_irq(irq_entry, ioapic, pin);
3810 
3811 		if ((ioapic > 0) && (irq > 16))
3812 			continue;
3813 
3814 		idata = irq_get_irq_data(irq);
3815 
3816 		/*
3817 		 * Honour affinities which have been set in early boot
3818 		 */
3819 		if (!irqd_can_balance(idata) || irqd_affinity_was_set(idata))
3820 			mask = idata->affinity;
3821 		else
3822 			mask = apic->target_cpus();
3823 
3824 		if (intr_remapping_enabled)
3825 			ir_ioapic_set_affinity(idata, mask, false);
3826 		else
3827 			ioapic_set_affinity(idata, mask, false);
3828 	}
3829 
3830 }
3831 #endif
3832 
3833 #define IOAPIC_RESOURCE_NAME_SIZE 11
3834 
3835 static struct resource *ioapic_resources;
3836 
3837 static struct resource * __init ioapic_setup_resources(int nr_ioapics)
3838 {
3839 	unsigned long n;
3840 	struct resource *res;
3841 	char *mem;
3842 	int i;
3843 
3844 	if (nr_ioapics <= 0)
3845 		return NULL;
3846 
3847 	n = IOAPIC_RESOURCE_NAME_SIZE + sizeof(struct resource);
3848 	n *= nr_ioapics;
3849 
3850 	mem = alloc_bootmem(n);
3851 	res = (void *)mem;
3852 
3853 	mem += sizeof(struct resource) * nr_ioapics;
3854 
3855 	for (i = 0; i < nr_ioapics; i++) {
3856 		res[i].name = mem;
3857 		res[i].flags = IORESOURCE_MEM | IORESOURCE_BUSY;
3858 		snprintf(mem, IOAPIC_RESOURCE_NAME_SIZE, "IOAPIC %u", i);
3859 		mem += IOAPIC_RESOURCE_NAME_SIZE;
3860 	}
3861 
3862 	ioapic_resources = res;
3863 
3864 	return res;
3865 }
3866 
3867 void __init ioapic_and_gsi_init(void)
3868 {
3869 	unsigned long ioapic_phys, idx = FIX_IO_APIC_BASE_0;
3870 	struct resource *ioapic_res;
3871 	int i;
3872 
3873 	ioapic_res = ioapic_setup_resources(nr_ioapics);
3874 	for (i = 0; i < nr_ioapics; i++) {
3875 		if (smp_found_config) {
3876 			ioapic_phys = mpc_ioapic_addr(i);
3877 #ifdef CONFIG_X86_32
3878 			if (!ioapic_phys) {
3879 				printk(KERN_ERR
3880 				       "WARNING: bogus zero IO-APIC "
3881 				       "address found in MPTABLE, "
3882 				       "disabling IO/APIC support!\n");
3883 				smp_found_config = 0;
3884 				skip_ioapic_setup = 1;
3885 				goto fake_ioapic_page;
3886 			}
3887 #endif
3888 		} else {
3889 #ifdef CONFIG_X86_32
3890 fake_ioapic_page:
3891 #endif
3892 			ioapic_phys = (unsigned long)alloc_bootmem_pages(PAGE_SIZE);
3893 			ioapic_phys = __pa(ioapic_phys);
3894 		}
3895 		set_fixmap_nocache(idx, ioapic_phys);
3896 		apic_printk(APIC_VERBOSE, "mapped IOAPIC to %08lx (%08lx)\n",
3897 			__fix_to_virt(idx) + (ioapic_phys & ~PAGE_MASK),
3898 			ioapic_phys);
3899 		idx++;
3900 
3901 		ioapic_res->start = ioapic_phys;
3902 		ioapic_res->end = ioapic_phys + IO_APIC_SLOT_SIZE - 1;
3903 		ioapic_res++;
3904 	}
3905 
3906 	probe_nr_irqs_gsi();
3907 }
3908 
3909 void __init ioapic_insert_resources(void)
3910 {
3911 	int i;
3912 	struct resource *r = ioapic_resources;
3913 
3914 	if (!r) {
3915 		if (nr_ioapics > 0)
3916 			printk(KERN_ERR
3917 				"IO APIC resources couldn't be allocated.\n");
3918 		return;
3919 	}
3920 
3921 	for (i = 0; i < nr_ioapics; i++) {
3922 		insert_resource(&iomem_resource, r);
3923 		r++;
3924 	}
3925 }
3926 
3927 int mp_find_ioapic(u32 gsi)
3928 {
3929 	int i = 0;
3930 
3931 	if (nr_ioapics == 0)
3932 		return -1;
3933 
3934 	/* Find the IOAPIC that manages this GSI. */
3935 	for (i = 0; i < nr_ioapics; i++) {
3936 		struct mp_ioapic_gsi *gsi_cfg = mp_ioapic_gsi_routing(i);
3937 		if ((gsi >= gsi_cfg->gsi_base)
3938 		    && (gsi <= gsi_cfg->gsi_end))
3939 			return i;
3940 	}
3941 
3942 	printk(KERN_ERR "ERROR: Unable to locate IOAPIC for GSI %d\n", gsi);
3943 	return -1;
3944 }
3945 
3946 int mp_find_ioapic_pin(int ioapic, u32 gsi)
3947 {
3948 	struct mp_ioapic_gsi *gsi_cfg;
3949 
3950 	if (WARN_ON(ioapic == -1))
3951 		return -1;
3952 
3953 	gsi_cfg = mp_ioapic_gsi_routing(ioapic);
3954 	if (WARN_ON(gsi > gsi_cfg->gsi_end))
3955 		return -1;
3956 
3957 	return gsi - gsi_cfg->gsi_base;
3958 }
3959 
3960 static __init int bad_ioapic(unsigned long address)
3961 {
3962 	if (nr_ioapics >= MAX_IO_APICS) {
3963 		printk(KERN_WARNING "WARNING: Max # of I/O APICs (%d) exceeded "
3964 		       "(found %d), skipping\n", MAX_IO_APICS, nr_ioapics);
3965 		return 1;
3966 	}
3967 	if (!address) {
3968 		printk(KERN_WARNING "WARNING: Bogus (zero) I/O APIC address"
3969 		       " found in table, skipping!\n");
3970 		return 1;
3971 	}
3972 	return 0;
3973 }
3974 
3975 void __init mp_register_ioapic(int id, u32 address, u32 gsi_base)
3976 {
3977 	int idx = 0;
3978 	int entries;
3979 	struct mp_ioapic_gsi *gsi_cfg;
3980 
3981 	if (bad_ioapic(address))
3982 		return;
3983 
3984 	idx = nr_ioapics;
3985 
3986 	ioapics[idx].mp_config.type = MP_IOAPIC;
3987 	ioapics[idx].mp_config.flags = MPC_APIC_USABLE;
3988 	ioapics[idx].mp_config.apicaddr = address;
3989 
3990 	set_fixmap_nocache(FIX_IO_APIC_BASE_0 + idx, address);
3991 	ioapics[idx].mp_config.apicid = io_apic_unique_id(id);
3992 	ioapics[idx].mp_config.apicver = io_apic_get_version(idx);
3993 
3994 	/*
3995 	 * Build basic GSI lookup table to facilitate gsi->io_apic lookups
3996 	 * and to prevent reprogramming of IOAPIC pins (PCI GSIs).
3997 	 */
3998 	entries = io_apic_get_redir_entries(idx);
3999 	gsi_cfg = mp_ioapic_gsi_routing(idx);
4000 	gsi_cfg->gsi_base = gsi_base;
4001 	gsi_cfg->gsi_end = gsi_base + entries - 1;
4002 
4003 	/*
4004 	 * The number of IO-APIC IRQ registers (== #pins):
4005 	 */
4006 	ioapics[idx].nr_registers = entries;
4007 
4008 	if (gsi_cfg->gsi_end >= gsi_top)
4009 		gsi_top = gsi_cfg->gsi_end + 1;
4010 
4011 	printk(KERN_INFO "IOAPIC[%d]: apic_id %d, version %d, address 0x%x, "
4012 	       "GSI %d-%d\n", idx, mpc_ioapic_id(idx),
4013 	       mpc_ioapic_ver(idx), mpc_ioapic_addr(idx),
4014 	       gsi_cfg->gsi_base, gsi_cfg->gsi_end);
4015 
4016 	nr_ioapics++;
4017 }
4018 
4019 /* Enable IOAPIC early just for system timer */
4020 void __init pre_init_apic_IRQ0(void)
4021 {
4022 	struct io_apic_irq_attr attr = { 0, 0, 0, 0 };
4023 
4024 	printk(KERN_INFO "Early APIC setup for system timer0\n");
4025 #ifndef CONFIG_SMP
4026 	physid_set_mask_of_physid(boot_cpu_physical_apicid,
4027 					 &phys_cpu_present_map);
4028 #endif
4029 	setup_local_APIC();
4030 
4031 	io_apic_setup_irq_pin(0, 0, &attr);
4032 	irq_set_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq,
4033 				      "edge");
4034 }
4035