Merge branch 'master' of git://oss.sgi.com/xfs/xfs into for-linus
[linux-2.6] / arch / x86 / kernel / apic / io_apic.c
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/sysdev.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 #ifdef CONFIG_ACPI
40 #include <acpi/acpi_bus.h>
41 #endif
42 #include <linux/bootmem.h>
43 #include <linux/dmar.h>
44 #include <linux/hpet.h>
45
46 #include <asm/idle.h>
47 #include <asm/io.h>
48 #include <asm/smp.h>
49 #include <asm/cpu.h>
50 #include <asm/desc.h>
51 #include <asm/proto.h>
52 #include <asm/acpi.h>
53 #include <asm/dma.h>
54 #include <asm/timer.h>
55 #include <asm/i8259.h>
56 #include <asm/nmi.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 #include <asm/uv/uv_hub.h>
64 #include <asm/uv/uv_irq.h>
65
66 #include <asm/apic.h>
67
68 #define __apicdebuginit(type) static type __init
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_SPINLOCK(ioapic_lock);
77 static DEFINE_SPINLOCK(vector_lock);
78
79 /*
80  * # of IRQ routing registers
81  */
82 int nr_ioapic_registers[MAX_IO_APICS];
83
84 /* I/O APIC entries */
85 struct mpc_ioapic mp_ioapics[MAX_IO_APICS];
86 int nr_ioapics;
87
88 /* MP IRQ source entries */
89 struct mpc_intsrc mp_irqs[MAX_IRQ_SOURCES];
90
91 /* # of MP IRQ source entries */
92 int mp_irq_entries;
93
94 #if defined (CONFIG_MCA) || defined (CONFIG_EISA)
95 int mp_bus_id_to_type[MAX_MP_BUSSES];
96 #endif
97
98 DECLARE_BITMAP(mp_bus_not_pci, MAX_MP_BUSSES);
99
100 int skip_ioapic_setup;
101
102 void arch_disable_smp_support(void)
103 {
104 #ifdef CONFIG_PCI
105         noioapicquirk = 1;
106         noioapicreroute = -1;
107 #endif
108         skip_ioapic_setup = 1;
109 }
110
111 static int __init parse_noapic(char *str)
112 {
113         /* disable IO-APIC */
114         arch_disable_smp_support();
115         return 0;
116 }
117 early_param("noapic", parse_noapic);
118
119 struct irq_pin_list;
120
121 /*
122  * This is performance-critical, we want to do it O(1)
123  *
124  * the indexing order of this array favors 1:1 mappings
125  * between pins and IRQs.
126  */
127
128 struct irq_pin_list {
129         int apic, pin;
130         struct irq_pin_list *next;
131 };
132
133 static struct irq_pin_list *get_one_free_irq_2_pin(int node)
134 {
135         struct irq_pin_list *pin;
136
137         pin = kzalloc_node(sizeof(*pin), GFP_ATOMIC, node);
138
139         return pin;
140 }
141
142 struct irq_cfg {
143         struct irq_pin_list *irq_2_pin;
144         cpumask_var_t domain;
145         cpumask_var_t old_domain;
146         unsigned move_cleanup_count;
147         u8 vector;
148         u8 move_in_progress : 1;
149 };
150
151 /* irq_cfg is indexed by the sum of all RTEs in all I/O APICs. */
152 #ifdef CONFIG_SPARSE_IRQ
153 static struct irq_cfg irq_cfgx[] = {
154 #else
155 static struct irq_cfg irq_cfgx[NR_IRQS] = {
156 #endif
157         [0]  = { .vector = IRQ0_VECTOR,  },
158         [1]  = { .vector = IRQ1_VECTOR,  },
159         [2]  = { .vector = IRQ2_VECTOR,  },
160         [3]  = { .vector = IRQ3_VECTOR,  },
161         [4]  = { .vector = IRQ4_VECTOR,  },
162         [5]  = { .vector = IRQ5_VECTOR,  },
163         [6]  = { .vector = IRQ6_VECTOR,  },
164         [7]  = { .vector = IRQ7_VECTOR,  },
165         [8]  = { .vector = IRQ8_VECTOR,  },
166         [9]  = { .vector = IRQ9_VECTOR,  },
167         [10] = { .vector = IRQ10_VECTOR, },
168         [11] = { .vector = IRQ11_VECTOR, },
169         [12] = { .vector = IRQ12_VECTOR, },
170         [13] = { .vector = IRQ13_VECTOR, },
171         [14] = { .vector = IRQ14_VECTOR, },
172         [15] = { .vector = IRQ15_VECTOR, },
173 };
174
175 int __init arch_early_irq_init(void)
176 {
177         struct irq_cfg *cfg;
178         struct irq_desc *desc;
179         int count;
180         int node;
181         int i;
182
183         cfg = irq_cfgx;
184         count = ARRAY_SIZE(irq_cfgx);
185         node= cpu_to_node(boot_cpu_id);
186
187         for (i = 0; i < count; i++) {
188                 desc = irq_to_desc(i);
189                 desc->chip_data = &cfg[i];
190                 zalloc_cpumask_var_node(&cfg[i].domain, GFP_NOWAIT, node);
191                 zalloc_cpumask_var_node(&cfg[i].old_domain, GFP_NOWAIT, node);
192                 if (i < NR_IRQS_LEGACY)
193                         cpumask_setall(cfg[i].domain);
194         }
195
196         return 0;
197 }
198
199 #ifdef CONFIG_SPARSE_IRQ
200 static struct irq_cfg *irq_cfg(unsigned int irq)
201 {
202         struct irq_cfg *cfg = NULL;
203         struct irq_desc *desc;
204
205         desc = irq_to_desc(irq);
206         if (desc)
207                 cfg = desc->chip_data;
208
209         return cfg;
210 }
211
212 static struct irq_cfg *get_one_free_irq_cfg(int node)
213 {
214         struct irq_cfg *cfg;
215
216         cfg = kzalloc_node(sizeof(*cfg), GFP_ATOMIC, node);
217         if (cfg) {
218                 if (!alloc_cpumask_var_node(&cfg->domain, GFP_ATOMIC, node)) {
219                         kfree(cfg);
220                         cfg = NULL;
221                 } else if (!alloc_cpumask_var_node(&cfg->old_domain,
222                                                           GFP_ATOMIC, node)) {
223                         free_cpumask_var(cfg->domain);
224                         kfree(cfg);
225                         cfg = NULL;
226                 } else {
227                         cpumask_clear(cfg->domain);
228                         cpumask_clear(cfg->old_domain);
229                 }
230         }
231
232         return cfg;
233 }
234
235 int arch_init_chip_data(struct irq_desc *desc, int node)
236 {
237         struct irq_cfg *cfg;
238
239         cfg = desc->chip_data;
240         if (!cfg) {
241                 desc->chip_data = get_one_free_irq_cfg(node);
242                 if (!desc->chip_data) {
243                         printk(KERN_ERR "can not alloc irq_cfg\n");
244                         BUG_ON(1);
245                 }
246         }
247
248         return 0;
249 }
250
251 /* for move_irq_desc */
252 static void
253 init_copy_irq_2_pin(struct irq_cfg *old_cfg, struct irq_cfg *cfg, int node)
254 {
255         struct irq_pin_list *old_entry, *head, *tail, *entry;
256
257         cfg->irq_2_pin = NULL;
258         old_entry = old_cfg->irq_2_pin;
259         if (!old_entry)
260                 return;
261
262         entry = get_one_free_irq_2_pin(node);
263         if (!entry)
264                 return;
265
266         entry->apic     = old_entry->apic;
267         entry->pin      = old_entry->pin;
268         head            = entry;
269         tail            = entry;
270         old_entry       = old_entry->next;
271         while (old_entry) {
272                 entry = get_one_free_irq_2_pin(node);
273                 if (!entry) {
274                         entry = head;
275                         while (entry) {
276                                 head = entry->next;
277                                 kfree(entry);
278                                 entry = head;
279                         }
280                         /* still use the old one */
281                         return;
282                 }
283                 entry->apic     = old_entry->apic;
284                 entry->pin      = old_entry->pin;
285                 tail->next      = entry;
286                 tail            = entry;
287                 old_entry       = old_entry->next;
288         }
289
290         tail->next = NULL;
291         cfg->irq_2_pin = head;
292 }
293
294 static void free_irq_2_pin(struct irq_cfg *old_cfg, struct irq_cfg *cfg)
295 {
296         struct irq_pin_list *entry, *next;
297
298         if (old_cfg->irq_2_pin == cfg->irq_2_pin)
299                 return;
300
301         entry = old_cfg->irq_2_pin;
302
303         while (entry) {
304                 next = entry->next;
305                 kfree(entry);
306                 entry = next;
307         }
308         old_cfg->irq_2_pin = NULL;
309 }
310
311 void arch_init_copy_chip_data(struct irq_desc *old_desc,
312                                  struct irq_desc *desc, int node)
313 {
314         struct irq_cfg *cfg;
315         struct irq_cfg *old_cfg;
316
317         cfg = get_one_free_irq_cfg(node);
318
319         if (!cfg)
320                 return;
321
322         desc->chip_data = cfg;
323
324         old_cfg = old_desc->chip_data;
325
326         memcpy(cfg, old_cfg, sizeof(struct irq_cfg));
327
328         init_copy_irq_2_pin(old_cfg, cfg, node);
329 }
330
331 static void free_irq_cfg(struct irq_cfg *old_cfg)
332 {
333         kfree(old_cfg);
334 }
335
336 void arch_free_chip_data(struct irq_desc *old_desc, struct irq_desc *desc)
337 {
338         struct irq_cfg *old_cfg, *cfg;
339
340         old_cfg = old_desc->chip_data;
341         cfg = desc->chip_data;
342
343         if (old_cfg == cfg)
344                 return;
345
346         if (old_cfg) {
347                 free_irq_2_pin(old_cfg, cfg);
348                 free_irq_cfg(old_cfg);
349                 old_desc->chip_data = NULL;
350         }
351 }
352 /* end for move_irq_desc */
353
354 #else
355 static struct irq_cfg *irq_cfg(unsigned int irq)
356 {
357         return irq < nr_irqs ? irq_cfgx + irq : NULL;
358 }
359
360 #endif
361
362 struct io_apic {
363         unsigned int index;
364         unsigned int unused[3];
365         unsigned int data;
366         unsigned int unused2[11];
367         unsigned int eoi;
368 };
369
370 static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
371 {
372         return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
373                 + (mp_ioapics[idx].apicaddr & ~PAGE_MASK);
374 }
375
376 static inline void io_apic_eoi(unsigned int apic, unsigned int vector)
377 {
378         struct io_apic __iomem *io_apic = io_apic_base(apic);
379         writel(vector, &io_apic->eoi);
380 }
381
382 static inline unsigned int io_apic_read(unsigned int apic, unsigned int reg)
383 {
384         struct io_apic __iomem *io_apic = io_apic_base(apic);
385         writel(reg, &io_apic->index);
386         return readl(&io_apic->data);
387 }
388
389 static inline void io_apic_write(unsigned int apic, unsigned int reg, unsigned int value)
390 {
391         struct io_apic __iomem *io_apic = io_apic_base(apic);
392         writel(reg, &io_apic->index);
393         writel(value, &io_apic->data);
394 }
395
396 /*
397  * Re-write a value: to be used for read-modify-write
398  * cycles where the read already set up the index register.
399  *
400  * Older SiS APIC requires we rewrite the index register
401  */
402 static inline void io_apic_modify(unsigned int apic, unsigned int reg, unsigned int value)
403 {
404         struct io_apic __iomem *io_apic = io_apic_base(apic);
405
406         if (sis_apic_bug)
407                 writel(reg, &io_apic->index);
408         writel(value, &io_apic->data);
409 }
410
411 static bool io_apic_level_ack_pending(struct irq_cfg *cfg)
412 {
413         struct irq_pin_list *entry;
414         unsigned long flags;
415
416         spin_lock_irqsave(&ioapic_lock, flags);
417         entry = cfg->irq_2_pin;
418         for (;;) {
419                 unsigned int reg;
420                 int pin;
421
422                 if (!entry)
423                         break;
424                 pin = entry->pin;
425                 reg = io_apic_read(entry->apic, 0x10 + pin*2);
426                 /* Is the remote IRR bit set? */
427                 if (reg & IO_APIC_REDIR_REMOTE_IRR) {
428                         spin_unlock_irqrestore(&ioapic_lock, flags);
429                         return true;
430                 }
431                 if (!entry->next)
432                         break;
433                 entry = entry->next;
434         }
435         spin_unlock_irqrestore(&ioapic_lock, flags);
436
437         return false;
438 }
439
440 union entry_union {
441         struct { u32 w1, w2; };
442         struct IO_APIC_route_entry entry;
443 };
444
445 static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
446 {
447         union entry_union eu;
448         unsigned long flags;
449         spin_lock_irqsave(&ioapic_lock, flags);
450         eu.w1 = io_apic_read(apic, 0x10 + 2 * pin);
451         eu.w2 = io_apic_read(apic, 0x11 + 2 * pin);
452         spin_unlock_irqrestore(&ioapic_lock, flags);
453         return eu.entry;
454 }
455
456 /*
457  * When we write a new IO APIC routing entry, we need to write the high
458  * word first! If the mask bit in the low word is clear, we will enable
459  * the interrupt, and we need to make sure the entry is fully populated
460  * before that happens.
461  */
462 static void
463 __ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
464 {
465         union entry_union eu;
466         eu.entry = e;
467         io_apic_write(apic, 0x11 + 2*pin, eu.w2);
468         io_apic_write(apic, 0x10 + 2*pin, eu.w1);
469 }
470
471 void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
472 {
473         unsigned long flags;
474         spin_lock_irqsave(&ioapic_lock, flags);
475         __ioapic_write_entry(apic, pin, e);
476         spin_unlock_irqrestore(&ioapic_lock, flags);
477 }
478
479 /*
480  * When we mask an IO APIC routing entry, we need to write the low
481  * word first, in order to set the mask bit before we change the
482  * high bits!
483  */
484 static void ioapic_mask_entry(int apic, int pin)
485 {
486         unsigned long flags;
487         union entry_union eu = { .entry.mask = 1 };
488
489         spin_lock_irqsave(&ioapic_lock, flags);
490         io_apic_write(apic, 0x10 + 2*pin, eu.w1);
491         io_apic_write(apic, 0x11 + 2*pin, eu.w2);
492         spin_unlock_irqrestore(&ioapic_lock, flags);
493 }
494
495 /*
496  * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
497  * shared ISA-space IRQs, so we have to support them. We are super
498  * fast in the common case, and fast for shared ISA-space IRQs.
499  */
500 static void add_pin_to_irq_node(struct irq_cfg *cfg, int node, int apic, int pin)
501 {
502         struct irq_pin_list *entry;
503
504         entry = cfg->irq_2_pin;
505         if (!entry) {
506                 entry = get_one_free_irq_2_pin(node);
507                 if (!entry) {
508                         printk(KERN_ERR "can not alloc irq_2_pin to add %d - %d\n",
509                                         apic, pin);
510                         return;
511                 }
512                 cfg->irq_2_pin = entry;
513                 entry->apic = apic;
514                 entry->pin = pin;
515                 return;
516         }
517
518         while (entry->next) {
519                 /* not again, please */
520                 if (entry->apic == apic && entry->pin == pin)
521                         return;
522
523                 entry = entry->next;
524         }
525
526         entry->next = get_one_free_irq_2_pin(node);
527         entry = entry->next;
528         entry->apic = apic;
529         entry->pin = pin;
530 }
531
532 /*
533  * Reroute an IRQ to a different pin.
534  */
535 static void __init replace_pin_at_irq_node(struct irq_cfg *cfg, int node,
536                                       int oldapic, int oldpin,
537                                       int newapic, int newpin)
538 {
539         struct irq_pin_list *entry = cfg->irq_2_pin;
540         int replaced = 0;
541
542         while (entry) {
543                 if (entry->apic == oldapic && entry->pin == oldpin) {
544                         entry->apic = newapic;
545                         entry->pin = newpin;
546                         replaced = 1;
547                         /* every one is different, right? */
548                         break;
549                 }
550                 entry = entry->next;
551         }
552
553         /* why? call replace before add? */
554         if (!replaced)
555                 add_pin_to_irq_node(cfg, node, newapic, newpin);
556 }
557
558 static inline void io_apic_modify_irq(struct irq_cfg *cfg,
559                                 int mask_and, int mask_or,
560                                 void (*final)(struct irq_pin_list *entry))
561 {
562         int pin;
563         struct irq_pin_list *entry;
564
565         for (entry = cfg->irq_2_pin; entry != NULL; entry = entry->next) {
566                 unsigned int reg;
567                 pin = entry->pin;
568                 reg = io_apic_read(entry->apic, 0x10 + pin * 2);
569                 reg &= mask_and;
570                 reg |= mask_or;
571                 io_apic_modify(entry->apic, 0x10 + pin * 2, reg);
572                 if (final)
573                         final(entry);
574         }
575 }
576
577 static void __unmask_IO_APIC_irq(struct irq_cfg *cfg)
578 {
579         io_apic_modify_irq(cfg, ~IO_APIC_REDIR_MASKED, 0, NULL);
580 }
581
582 #ifdef CONFIG_X86_64
583 static void io_apic_sync(struct irq_pin_list *entry)
584 {
585         /*
586          * Synchronize the IO-APIC and the CPU by doing
587          * a dummy read from the IO-APIC
588          */
589         struct io_apic __iomem *io_apic;
590         io_apic = io_apic_base(entry->apic);
591         readl(&io_apic->data);
592 }
593
594 static void __mask_IO_APIC_irq(struct irq_cfg *cfg)
595 {
596         io_apic_modify_irq(cfg, ~0, IO_APIC_REDIR_MASKED, &io_apic_sync);
597 }
598 #else /* CONFIG_X86_32 */
599 static void __mask_IO_APIC_irq(struct irq_cfg *cfg)
600 {
601         io_apic_modify_irq(cfg, ~0, IO_APIC_REDIR_MASKED, NULL);
602 }
603
604 static void __mask_and_edge_IO_APIC_irq(struct irq_cfg *cfg)
605 {
606         io_apic_modify_irq(cfg, ~IO_APIC_REDIR_LEVEL_TRIGGER,
607                         IO_APIC_REDIR_MASKED, NULL);
608 }
609
610 static void __unmask_and_level_IO_APIC_irq(struct irq_cfg *cfg)
611 {
612         io_apic_modify_irq(cfg, ~IO_APIC_REDIR_MASKED,
613                         IO_APIC_REDIR_LEVEL_TRIGGER, NULL);
614 }
615 #endif /* CONFIG_X86_32 */
616
617 static void mask_IO_APIC_irq_desc(struct irq_desc *desc)
618 {
619         struct irq_cfg *cfg = desc->chip_data;
620         unsigned long flags;
621
622         BUG_ON(!cfg);
623
624         spin_lock_irqsave(&ioapic_lock, flags);
625         __mask_IO_APIC_irq(cfg);
626         spin_unlock_irqrestore(&ioapic_lock, flags);
627 }
628
629 static void unmask_IO_APIC_irq_desc(struct irq_desc *desc)
630 {
631         struct irq_cfg *cfg = desc->chip_data;
632         unsigned long flags;
633
634         spin_lock_irqsave(&ioapic_lock, flags);
635         __unmask_IO_APIC_irq(cfg);
636         spin_unlock_irqrestore(&ioapic_lock, flags);
637 }
638
639 static void mask_IO_APIC_irq(unsigned int irq)
640 {
641         struct irq_desc *desc = irq_to_desc(irq);
642
643         mask_IO_APIC_irq_desc(desc);
644 }
645 static void unmask_IO_APIC_irq(unsigned int irq)
646 {
647         struct irq_desc *desc = irq_to_desc(irq);
648
649         unmask_IO_APIC_irq_desc(desc);
650 }
651
652 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
653 {
654         struct IO_APIC_route_entry entry;
655
656         /* Check delivery_mode to be sure we're not clearing an SMI pin */
657         entry = ioapic_read_entry(apic, pin);
658         if (entry.delivery_mode == dest_SMI)
659                 return;
660         /*
661          * Disable it in the IO-APIC irq-routing table:
662          */
663         ioapic_mask_entry(apic, pin);
664 }
665
666 static void clear_IO_APIC (void)
667 {
668         int apic, pin;
669
670         for (apic = 0; apic < nr_ioapics; apic++)
671                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
672                         clear_IO_APIC_pin(apic, pin);
673 }
674
675 #ifdef CONFIG_X86_32
676 /*
677  * support for broken MP BIOSs, enables hand-redirection of PIRQ0-7 to
678  * specific CPU-side IRQs.
679  */
680
681 #define MAX_PIRQS 8
682 static int pirq_entries[MAX_PIRQS] = {
683         [0 ... MAX_PIRQS - 1] = -1
684 };
685
686 static int __init ioapic_pirq_setup(char *str)
687 {
688         int i, max;
689         int ints[MAX_PIRQS+1];
690
691         get_options(str, ARRAY_SIZE(ints), ints);
692
693         apic_printk(APIC_VERBOSE, KERN_INFO
694                         "PIRQ redirection, working around broken MP-BIOS.\n");
695         max = MAX_PIRQS;
696         if (ints[0] < MAX_PIRQS)
697                 max = ints[0];
698
699         for (i = 0; i < max; i++) {
700                 apic_printk(APIC_VERBOSE, KERN_DEBUG
701                                 "... PIRQ%d -> IRQ %d\n", i, ints[i+1]);
702                 /*
703                  * PIRQs are mapped upside down, usually.
704                  */
705                 pirq_entries[MAX_PIRQS-i-1] = ints[i+1];
706         }
707         return 1;
708 }
709
710 __setup("pirq=", ioapic_pirq_setup);
711 #endif /* CONFIG_X86_32 */
712
713 struct IO_APIC_route_entry **alloc_ioapic_entries(void)
714 {
715         int apic;
716         struct IO_APIC_route_entry **ioapic_entries;
717
718         ioapic_entries = kzalloc(sizeof(*ioapic_entries) * nr_ioapics,
719                                 GFP_ATOMIC);
720         if (!ioapic_entries)
721                 return 0;
722
723         for (apic = 0; apic < nr_ioapics; apic++) {
724                 ioapic_entries[apic] =
725                         kzalloc(sizeof(struct IO_APIC_route_entry) *
726                                 nr_ioapic_registers[apic], GFP_ATOMIC);
727                 if (!ioapic_entries[apic])
728                         goto nomem;
729         }
730
731         return ioapic_entries;
732
733 nomem:
734         while (--apic >= 0)
735                 kfree(ioapic_entries[apic]);
736         kfree(ioapic_entries);
737
738         return 0;
739 }
740
741 /*
742  * Saves all the IO-APIC RTE's
743  */
744 int save_IO_APIC_setup(struct IO_APIC_route_entry **ioapic_entries)
745 {
746         int apic, pin;
747
748         if (!ioapic_entries)
749                 return -ENOMEM;
750
751         for (apic = 0; apic < nr_ioapics; apic++) {
752                 if (!ioapic_entries[apic])
753                         return -ENOMEM;
754
755                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
756                         ioapic_entries[apic][pin] =
757                                 ioapic_read_entry(apic, pin);
758         }
759
760         return 0;
761 }
762
763 /*
764  * Mask all IO APIC entries.
765  */
766 void mask_IO_APIC_setup(struct IO_APIC_route_entry **ioapic_entries)
767 {
768         int apic, pin;
769
770         if (!ioapic_entries)
771                 return;
772
773         for (apic = 0; apic < nr_ioapics; apic++) {
774                 if (!ioapic_entries[apic])
775                         break;
776
777                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
778                         struct IO_APIC_route_entry entry;
779
780                         entry = ioapic_entries[apic][pin];
781                         if (!entry.mask) {
782                                 entry.mask = 1;
783                                 ioapic_write_entry(apic, pin, entry);
784                         }
785                 }
786         }
787 }
788
789 /*
790  * Restore IO APIC entries which was saved in ioapic_entries.
791  */
792 int restore_IO_APIC_setup(struct IO_APIC_route_entry **ioapic_entries)
793 {
794         int apic, pin;
795
796         if (!ioapic_entries)
797                 return -ENOMEM;
798
799         for (apic = 0; apic < nr_ioapics; apic++) {
800                 if (!ioapic_entries[apic])
801                         return -ENOMEM;
802
803                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
804                         ioapic_write_entry(apic, pin,
805                                         ioapic_entries[apic][pin]);
806         }
807         return 0;
808 }
809
810 void free_ioapic_entries(struct IO_APIC_route_entry **ioapic_entries)
811 {
812         int apic;
813
814         for (apic = 0; apic < nr_ioapics; apic++)
815                 kfree(ioapic_entries[apic]);
816
817         kfree(ioapic_entries);
818 }
819
820 /*
821  * Find the IRQ entry number of a certain pin.
822  */
823 static int find_irq_entry(int apic, int pin, int type)
824 {
825         int i;
826
827         for (i = 0; i < mp_irq_entries; i++)
828                 if (mp_irqs[i].irqtype == type &&
829                     (mp_irqs[i].dstapic == mp_ioapics[apic].apicid ||
830                      mp_irqs[i].dstapic == MP_APIC_ALL) &&
831                     mp_irqs[i].dstirq == pin)
832                         return i;
833
834         return -1;
835 }
836
837 /*
838  * Find the pin to which IRQ[irq] (ISA) is connected
839  */
840 static int __init find_isa_irq_pin(int irq, int type)
841 {
842         int i;
843
844         for (i = 0; i < mp_irq_entries; i++) {
845                 int lbus = mp_irqs[i].srcbus;
846
847                 if (test_bit(lbus, mp_bus_not_pci) &&
848                     (mp_irqs[i].irqtype == type) &&
849                     (mp_irqs[i].srcbusirq == irq))
850
851                         return mp_irqs[i].dstirq;
852         }
853         return -1;
854 }
855
856 static int __init find_isa_irq_apic(int irq, int type)
857 {
858         int i;
859
860         for (i = 0; i < mp_irq_entries; i++) {
861                 int lbus = mp_irqs[i].srcbus;
862
863                 if (test_bit(lbus, mp_bus_not_pci) &&
864                     (mp_irqs[i].irqtype == type) &&
865                     (mp_irqs[i].srcbusirq == irq))
866                         break;
867         }
868         if (i < mp_irq_entries) {
869                 int apic;
870                 for(apic = 0; apic < nr_ioapics; apic++) {
871                         if (mp_ioapics[apic].apicid == mp_irqs[i].dstapic)
872                                 return apic;
873                 }
874         }
875
876         return -1;
877 }
878
879 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
880 /*
881  * EISA Edge/Level control register, ELCR
882  */
883 static int EISA_ELCR(unsigned int irq)
884 {
885         if (irq < NR_IRQS_LEGACY) {
886                 unsigned int port = 0x4d0 + (irq >> 3);
887                 return (inb(port) >> (irq & 7)) & 1;
888         }
889         apic_printk(APIC_VERBOSE, KERN_INFO
890                         "Broken MPtable reports ISA irq %d\n", irq);
891         return 0;
892 }
893
894 #endif
895
896 /* ISA interrupts are always polarity zero edge triggered,
897  * when listed as conforming in the MP table. */
898
899 #define default_ISA_trigger(idx)        (0)
900 #define default_ISA_polarity(idx)       (0)
901
902 /* EISA interrupts are always polarity zero and can be edge or level
903  * trigger depending on the ELCR value.  If an interrupt is listed as
904  * EISA conforming in the MP table, that means its trigger type must
905  * be read in from the ELCR */
906
907 #define default_EISA_trigger(idx)       (EISA_ELCR(mp_irqs[idx].srcbusirq))
908 #define default_EISA_polarity(idx)      default_ISA_polarity(idx)
909
910 /* PCI interrupts are always polarity one level triggered,
911  * when listed as conforming in the MP table. */
912
913 #define default_PCI_trigger(idx)        (1)
914 #define default_PCI_polarity(idx)       (1)
915
916 /* MCA interrupts are always polarity zero level triggered,
917  * when listed as conforming in the MP table. */
918
919 #define default_MCA_trigger(idx)        (1)
920 #define default_MCA_polarity(idx)       default_ISA_polarity(idx)
921
922 static int MPBIOS_polarity(int idx)
923 {
924         int bus = mp_irqs[idx].srcbus;
925         int polarity;
926
927         /*
928          * Determine IRQ line polarity (high active or low active):
929          */
930         switch (mp_irqs[idx].irqflag & 3)
931         {
932                 case 0: /* conforms, ie. bus-type dependent polarity */
933                         if (test_bit(bus, mp_bus_not_pci))
934                                 polarity = default_ISA_polarity(idx);
935                         else
936                                 polarity = default_PCI_polarity(idx);
937                         break;
938                 case 1: /* high active */
939                 {
940                         polarity = 0;
941                         break;
942                 }
943                 case 2: /* reserved */
944                 {
945                         printk(KERN_WARNING "broken BIOS!!\n");
946                         polarity = 1;
947                         break;
948                 }
949                 case 3: /* low active */
950                 {
951                         polarity = 1;
952                         break;
953                 }
954                 default: /* invalid */
955                 {
956                         printk(KERN_WARNING "broken BIOS!!\n");
957                         polarity = 1;
958                         break;
959                 }
960         }
961         return polarity;
962 }
963
964 static int MPBIOS_trigger(int idx)
965 {
966         int bus = mp_irqs[idx].srcbus;
967         int trigger;
968
969         /*
970          * Determine IRQ trigger mode (edge or level sensitive):
971          */
972         switch ((mp_irqs[idx].irqflag>>2) & 3)
973         {
974                 case 0: /* conforms, ie. bus-type dependent */
975                         if (test_bit(bus, mp_bus_not_pci))
976                                 trigger = default_ISA_trigger(idx);
977                         else
978                                 trigger = default_PCI_trigger(idx);
979 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
980                         switch (mp_bus_id_to_type[bus]) {
981                                 case MP_BUS_ISA: /* ISA pin */
982                                 {
983                                         /* set before the switch */
984                                         break;
985                                 }
986                                 case MP_BUS_EISA: /* EISA pin */
987                                 {
988                                         trigger = default_EISA_trigger(idx);
989                                         break;
990                                 }
991                                 case MP_BUS_PCI: /* PCI pin */
992                                 {
993                                         /* set before the switch */
994                                         break;
995                                 }
996                                 case MP_BUS_MCA: /* MCA pin */
997                                 {
998                                         trigger = default_MCA_trigger(idx);
999                                         break;
1000                                 }
1001                                 default:
1002                                 {
1003                                         printk(KERN_WARNING "broken BIOS!!\n");
1004                                         trigger = 1;
1005                                         break;
1006                                 }
1007                         }
1008 #endif
1009                         break;
1010                 case 1: /* edge */
1011                 {
1012                         trigger = 0;
1013                         break;
1014                 }
1015                 case 2: /* reserved */
1016                 {
1017                         printk(KERN_WARNING "broken BIOS!!\n");
1018                         trigger = 1;
1019                         break;
1020                 }
1021                 case 3: /* level */
1022                 {
1023                         trigger = 1;
1024                         break;
1025                 }
1026                 default: /* invalid */
1027                 {
1028                         printk(KERN_WARNING "broken BIOS!!\n");
1029                         trigger = 0;
1030                         break;
1031                 }
1032         }
1033         return trigger;
1034 }
1035
1036 static inline int irq_polarity(int idx)
1037 {
1038         return MPBIOS_polarity(idx);
1039 }
1040
1041 static inline int irq_trigger(int idx)
1042 {
1043         return MPBIOS_trigger(idx);
1044 }
1045
1046 int (*ioapic_renumber_irq)(int ioapic, int irq);
1047 static int pin_2_irq(int idx, int apic, int pin)
1048 {
1049         int irq, i;
1050         int bus = mp_irqs[idx].srcbus;
1051
1052         /*
1053          * Debugging check, we are in big trouble if this message pops up!
1054          */
1055         if (mp_irqs[idx].dstirq != pin)
1056                 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
1057
1058         if (test_bit(bus, mp_bus_not_pci)) {
1059                 irq = mp_irqs[idx].srcbusirq;
1060         } else {
1061                 /*
1062                  * PCI IRQs are mapped in order
1063                  */
1064                 i = irq = 0;
1065                 while (i < apic)
1066                         irq += nr_ioapic_registers[i++];
1067                 irq += pin;
1068                 /*
1069                  * For MPS mode, so far only needed by ES7000 platform
1070                  */
1071                 if (ioapic_renumber_irq)
1072                         irq = ioapic_renumber_irq(apic, irq);
1073         }
1074
1075 #ifdef CONFIG_X86_32
1076         /*
1077          * PCI IRQ command line redirection. Yes, limits are hardcoded.
1078          */
1079         if ((pin >= 16) && (pin <= 23)) {
1080                 if (pirq_entries[pin-16] != -1) {
1081                         if (!pirq_entries[pin-16]) {
1082                                 apic_printk(APIC_VERBOSE, KERN_DEBUG
1083                                                 "disabling PIRQ%d\n", pin-16);
1084                         } else {
1085                                 irq = pirq_entries[pin-16];
1086                                 apic_printk(APIC_VERBOSE, KERN_DEBUG
1087                                                 "using PIRQ%d -> IRQ %d\n",
1088                                                 pin-16, irq);
1089                         }
1090                 }
1091         }
1092 #endif
1093
1094         return irq;
1095 }
1096
1097 /*
1098  * Find a specific PCI IRQ entry.
1099  * Not an __init, possibly needed by modules
1100  */
1101 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin,
1102                                 struct io_apic_irq_attr *irq_attr)
1103 {
1104         int apic, i, best_guess = -1;
1105
1106         apic_printk(APIC_DEBUG,
1107                     "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
1108                     bus, slot, pin);
1109         if (test_bit(bus, mp_bus_not_pci)) {
1110                 apic_printk(APIC_VERBOSE,
1111                             "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
1112                 return -1;
1113         }
1114         for (i = 0; i < mp_irq_entries; i++) {
1115                 int lbus = mp_irqs[i].srcbus;
1116
1117                 for (apic = 0; apic < nr_ioapics; apic++)
1118                         if (mp_ioapics[apic].apicid == mp_irqs[i].dstapic ||
1119                             mp_irqs[i].dstapic == MP_APIC_ALL)
1120                                 break;
1121
1122                 if (!test_bit(lbus, mp_bus_not_pci) &&
1123                     !mp_irqs[i].irqtype &&
1124                     (bus == lbus) &&
1125                     (slot == ((mp_irqs[i].srcbusirq >> 2) & 0x1f))) {
1126                         int irq = pin_2_irq(i, apic, mp_irqs[i].dstirq);
1127
1128                         if (!(apic || IO_APIC_IRQ(irq)))
1129                                 continue;
1130
1131                         if (pin == (mp_irqs[i].srcbusirq & 3)) {
1132                                 set_io_apic_irq_attr(irq_attr, apic,
1133                                                      mp_irqs[i].dstirq,
1134                                                      irq_trigger(i),
1135                                                      irq_polarity(i));
1136                                 return irq;
1137                         }
1138                         /*
1139                          * Use the first all-but-pin matching entry as a
1140                          * best-guess fuzzy result for broken mptables.
1141                          */
1142                         if (best_guess < 0) {
1143                                 set_io_apic_irq_attr(irq_attr, apic,
1144                                                      mp_irqs[i].dstirq,
1145                                                      irq_trigger(i),
1146                                                      irq_polarity(i));
1147                                 best_guess = irq;
1148                         }
1149                 }
1150         }
1151         return best_guess;
1152 }
1153 EXPORT_SYMBOL(IO_APIC_get_PCI_irq_vector);
1154
1155 void lock_vector_lock(void)
1156 {
1157         /* Used to the online set of cpus does not change
1158          * during assign_irq_vector.
1159          */
1160         spin_lock(&vector_lock);
1161 }
1162
1163 void unlock_vector_lock(void)
1164 {
1165         spin_unlock(&vector_lock);
1166 }
1167
1168 static int
1169 __assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask)
1170 {
1171         /*
1172          * NOTE! The local APIC isn't very good at handling
1173          * multiple interrupts at the same interrupt level.
1174          * As the interrupt level is determined by taking the
1175          * vector number and shifting that right by 4, we
1176          * want to spread these out a bit so that they don't
1177          * all fall in the same interrupt level.
1178          *
1179          * Also, we've got to be careful not to trash gate
1180          * 0x80, because int 0x80 is hm, kind of importantish. ;)
1181          */
1182         static int current_vector = FIRST_DEVICE_VECTOR, current_offset = 0;
1183         unsigned int old_vector;
1184         int cpu, err;
1185         cpumask_var_t tmp_mask;
1186
1187         if ((cfg->move_in_progress) || cfg->move_cleanup_count)
1188                 return -EBUSY;
1189
1190         if (!alloc_cpumask_var(&tmp_mask, GFP_ATOMIC))
1191                 return -ENOMEM;
1192
1193         old_vector = cfg->vector;
1194         if (old_vector) {
1195                 cpumask_and(tmp_mask, mask, cpu_online_mask);
1196                 cpumask_and(tmp_mask, cfg->domain, tmp_mask);
1197                 if (!cpumask_empty(tmp_mask)) {
1198                         free_cpumask_var(tmp_mask);
1199                         return 0;
1200                 }
1201         }
1202
1203         /* Only try and allocate irqs on cpus that are present */
1204         err = -ENOSPC;
1205         for_each_cpu_and(cpu, mask, cpu_online_mask) {
1206                 int new_cpu;
1207                 int vector, offset;
1208
1209                 apic->vector_allocation_domain(cpu, tmp_mask);
1210
1211                 vector = current_vector;
1212                 offset = current_offset;
1213 next:
1214                 vector += 8;
1215                 if (vector >= first_system_vector) {
1216                         /* If out of vectors on large boxen, must share them. */
1217                         offset = (offset + 1) % 8;
1218                         vector = FIRST_DEVICE_VECTOR + offset;
1219                 }
1220                 if (unlikely(current_vector == vector))
1221                         continue;
1222
1223                 if (test_bit(vector, used_vectors))
1224                         goto next;
1225
1226                 for_each_cpu_and(new_cpu, tmp_mask, cpu_online_mask)
1227                         if (per_cpu(vector_irq, new_cpu)[vector] != -1)
1228                                 goto next;
1229                 /* Found one! */
1230                 current_vector = vector;
1231                 current_offset = offset;
1232                 if (old_vector) {
1233                         cfg->move_in_progress = 1;
1234                         cpumask_copy(cfg->old_domain, cfg->domain);
1235                 }
1236                 for_each_cpu_and(new_cpu, tmp_mask, cpu_online_mask)
1237                         per_cpu(vector_irq, new_cpu)[vector] = irq;
1238                 cfg->vector = vector;
1239                 cpumask_copy(cfg->domain, tmp_mask);
1240                 err = 0;
1241                 break;
1242         }
1243         free_cpumask_var(tmp_mask);
1244         return err;
1245 }
1246
1247 static int
1248 assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask)
1249 {
1250         int err;
1251         unsigned long flags;
1252
1253         spin_lock_irqsave(&vector_lock, flags);
1254         err = __assign_irq_vector(irq, cfg, mask);
1255         spin_unlock_irqrestore(&vector_lock, flags);
1256         return err;
1257 }
1258
1259 static void __clear_irq_vector(int irq, struct irq_cfg *cfg)
1260 {
1261         int cpu, vector;
1262
1263         BUG_ON(!cfg->vector);
1264
1265         vector = cfg->vector;
1266         for_each_cpu_and(cpu, cfg->domain, cpu_online_mask)
1267                 per_cpu(vector_irq, cpu)[vector] = -1;
1268
1269         cfg->vector = 0;
1270         cpumask_clear(cfg->domain);
1271
1272         if (likely(!cfg->move_in_progress))
1273                 return;
1274         for_each_cpu_and(cpu, cfg->old_domain, cpu_online_mask) {
1275                 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS;
1276                                                                 vector++) {
1277                         if (per_cpu(vector_irq, cpu)[vector] != irq)
1278                                 continue;
1279                         per_cpu(vector_irq, cpu)[vector] = -1;
1280                         break;
1281                 }
1282         }
1283         cfg->move_in_progress = 0;
1284 }
1285
1286 void __setup_vector_irq(int cpu)
1287 {
1288         /* Initialize vector_irq on a new cpu */
1289         /* This function must be called with vector_lock held */
1290         int irq, vector;
1291         struct irq_cfg *cfg;
1292         struct irq_desc *desc;
1293
1294         /* Mark the inuse vectors */
1295         for_each_irq_desc(irq, desc) {
1296                 cfg = desc->chip_data;
1297                 if (!cpumask_test_cpu(cpu, cfg->domain))
1298                         continue;
1299                 vector = cfg->vector;
1300                 per_cpu(vector_irq, cpu)[vector] = irq;
1301         }
1302         /* Mark the free vectors */
1303         for (vector = 0; vector < NR_VECTORS; ++vector) {
1304                 irq = per_cpu(vector_irq, cpu)[vector];
1305                 if (irq < 0)
1306                         continue;
1307
1308                 cfg = irq_cfg(irq);
1309                 if (!cpumask_test_cpu(cpu, cfg->domain))
1310                         per_cpu(vector_irq, cpu)[vector] = -1;
1311         }
1312 }
1313
1314 static struct irq_chip ioapic_chip;
1315 static struct irq_chip ir_ioapic_chip;
1316
1317 #define IOAPIC_AUTO     -1
1318 #define IOAPIC_EDGE     0
1319 #define IOAPIC_LEVEL    1
1320
1321 #ifdef CONFIG_X86_32
1322 static inline int IO_APIC_irq_trigger(int irq)
1323 {
1324         int apic, idx, pin;
1325
1326         for (apic = 0; apic < nr_ioapics; apic++) {
1327                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1328                         idx = find_irq_entry(apic, pin, mp_INT);
1329                         if ((idx != -1) && (irq == pin_2_irq(idx, apic, pin)))
1330                                 return irq_trigger(idx);
1331                 }
1332         }
1333         /*
1334          * nonexistent IRQs are edge default
1335          */
1336         return 0;
1337 }
1338 #else
1339 static inline int IO_APIC_irq_trigger(int irq)
1340 {
1341         return 1;
1342 }
1343 #endif
1344
1345 static void ioapic_register_intr(int irq, struct irq_desc *desc, unsigned long trigger)
1346 {
1347
1348         if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1349             trigger == IOAPIC_LEVEL)
1350                 desc->status |= IRQ_LEVEL;
1351         else
1352                 desc->status &= ~IRQ_LEVEL;
1353
1354         if (irq_remapped(irq)) {
1355                 desc->status |= IRQ_MOVE_PCNTXT;
1356                 if (trigger)
1357                         set_irq_chip_and_handler_name(irq, &ir_ioapic_chip,
1358                                                       handle_fasteoi_irq,
1359                                                      "fasteoi");
1360                 else
1361                         set_irq_chip_and_handler_name(irq, &ir_ioapic_chip,
1362                                                       handle_edge_irq, "edge");
1363                 return;
1364         }
1365
1366         if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1367             trigger == IOAPIC_LEVEL)
1368                 set_irq_chip_and_handler_name(irq, &ioapic_chip,
1369                                               handle_fasteoi_irq,
1370                                               "fasteoi");
1371         else
1372                 set_irq_chip_and_handler_name(irq, &ioapic_chip,
1373                                               handle_edge_irq, "edge");
1374 }
1375
1376 int setup_ioapic_entry(int apic_id, int irq,
1377                        struct IO_APIC_route_entry *entry,
1378                        unsigned int destination, int trigger,
1379                        int polarity, int vector, int pin)
1380 {
1381         /*
1382          * add it to the IO-APIC irq-routing table:
1383          */
1384         memset(entry,0,sizeof(*entry));
1385
1386         if (intr_remapping_enabled) {
1387                 struct intel_iommu *iommu = map_ioapic_to_ir(apic_id);
1388                 struct irte irte;
1389                 struct IR_IO_APIC_route_entry *ir_entry =
1390                         (struct IR_IO_APIC_route_entry *) entry;
1391                 int index;
1392
1393                 if (!iommu)
1394                         panic("No mapping iommu for ioapic %d\n", apic_id);
1395
1396                 index = alloc_irte(iommu, irq, 1);
1397                 if (index < 0)
1398                         panic("Failed to allocate IRTE for ioapic %d\n", apic_id);
1399
1400                 memset(&irte, 0, sizeof(irte));
1401
1402                 irte.present = 1;
1403                 irte.dst_mode = apic->irq_dest_mode;
1404                 /*
1405                  * Trigger mode in the IRTE will always be edge, and the
1406                  * actual level or edge trigger will be setup in the IO-APIC
1407                  * RTE. This will help simplify level triggered irq migration.
1408                  * For more details, see the comments above explainig IO-APIC
1409                  * irq migration in the presence of interrupt-remapping.
1410                  */
1411                 irte.trigger_mode = 0;
1412                 irte.dlvry_mode = apic->irq_delivery_mode;
1413                 irte.vector = vector;
1414                 irte.dest_id = IRTE_DEST(destination);
1415
1416                 modify_irte(irq, &irte);
1417
1418                 ir_entry->index2 = (index >> 15) & 0x1;
1419                 ir_entry->zero = 0;
1420                 ir_entry->format = 1;
1421                 ir_entry->index = (index & 0x7fff);
1422                 /*
1423                  * IO-APIC RTE will be configured with virtual vector.
1424                  * irq handler will do the explicit EOI to the io-apic.
1425                  */
1426                 ir_entry->vector = pin;
1427         } else {
1428                 entry->delivery_mode = apic->irq_delivery_mode;
1429                 entry->dest_mode = apic->irq_dest_mode;
1430                 entry->dest = destination;
1431                 entry->vector = vector;
1432         }
1433
1434         entry->mask = 0;                                /* enable IRQ */
1435         entry->trigger = trigger;
1436         entry->polarity = polarity;
1437
1438         /* Mask level triggered irqs.
1439          * Use IRQ_DELAYED_DISABLE for edge triggered irqs.
1440          */
1441         if (trigger)
1442                 entry->mask = 1;
1443         return 0;
1444 }
1445
1446 static void setup_IO_APIC_irq(int apic_id, int pin, unsigned int irq, struct irq_desc *desc,
1447                               int trigger, int polarity)
1448 {
1449         struct irq_cfg *cfg;
1450         struct IO_APIC_route_entry entry;
1451         unsigned int dest;
1452
1453         if (!IO_APIC_IRQ(irq))
1454                 return;
1455
1456         cfg = desc->chip_data;
1457
1458         if (assign_irq_vector(irq, cfg, apic->target_cpus()))
1459                 return;
1460
1461         dest = apic->cpu_mask_to_apicid_and(cfg->domain, apic->target_cpus());
1462
1463         apic_printk(APIC_VERBOSE,KERN_DEBUG
1464                     "IOAPIC[%d]: Set routing entry (%d-%d -> 0x%x -> "
1465                     "IRQ %d Mode:%i Active:%i)\n",
1466                     apic_id, mp_ioapics[apic_id].apicid, pin, cfg->vector,
1467                     irq, trigger, polarity);
1468
1469
1470         if (setup_ioapic_entry(mp_ioapics[apic_id].apicid, irq, &entry,
1471                                dest, trigger, polarity, cfg->vector, pin)) {
1472                 printk("Failed to setup ioapic entry for ioapic  %d, pin %d\n",
1473                        mp_ioapics[apic_id].apicid, pin);
1474                 __clear_irq_vector(irq, cfg);
1475                 return;
1476         }
1477
1478         ioapic_register_intr(irq, desc, trigger);
1479         if (irq < NR_IRQS_LEGACY)
1480                 disable_8259A_irq(irq);
1481
1482         ioapic_write_entry(apic_id, pin, entry);
1483 }
1484
1485 static struct {
1486         DECLARE_BITMAP(pin_programmed, MP_MAX_IOAPIC_PIN + 1);
1487 } mp_ioapic_routing[MAX_IO_APICS];
1488
1489 static void __init setup_IO_APIC_irqs(void)
1490 {
1491         int apic_id = 0, pin, idx, irq;
1492         int notcon = 0;
1493         struct irq_desc *desc;
1494         struct irq_cfg *cfg;
1495         int node = cpu_to_node(boot_cpu_id);
1496
1497         apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
1498
1499 #ifdef CONFIG_ACPI
1500         if (!acpi_disabled && acpi_ioapic) {
1501                 apic_id = mp_find_ioapic(0);
1502                 if (apic_id < 0)
1503                         apic_id = 0;
1504         }
1505 #endif
1506
1507         for (pin = 0; pin < nr_ioapic_registers[apic_id]; pin++) {
1508                 idx = find_irq_entry(apic_id, pin, mp_INT);
1509                 if (idx == -1) {
1510                         if (!notcon) {
1511                                 notcon = 1;
1512                                 apic_printk(APIC_VERBOSE,
1513                                         KERN_DEBUG " %d-%d",
1514                                         mp_ioapics[apic_id].apicid, pin);
1515                         } else
1516                                 apic_printk(APIC_VERBOSE, " %d-%d",
1517                                         mp_ioapics[apic_id].apicid, pin);
1518                         continue;
1519                 }
1520                 if (notcon) {
1521                         apic_printk(APIC_VERBOSE,
1522                                 " (apicid-pin) not connected\n");
1523                         notcon = 0;
1524                 }
1525
1526                 irq = pin_2_irq(idx, apic_id, pin);
1527
1528                 /*
1529                  * Skip the timer IRQ if there's a quirk handler
1530                  * installed and if it returns 1:
1531                  */
1532                 if (apic->multi_timer_check &&
1533                                 apic->multi_timer_check(apic_id, irq))
1534                         continue;
1535
1536                 desc = irq_to_desc_alloc_node(irq, node);
1537                 if (!desc) {
1538                         printk(KERN_INFO "can not get irq_desc for %d\n", irq);
1539                         continue;
1540                 }
1541                 cfg = desc->chip_data;
1542                 add_pin_to_irq_node(cfg, node, apic_id, pin);
1543                 /*
1544                  * don't mark it in pin_programmed, so later acpi could
1545                  * set it correctly when irq < 16
1546                  */
1547                 setup_IO_APIC_irq(apic_id, pin, irq, desc,
1548                                 irq_trigger(idx), irq_polarity(idx));
1549         }
1550
1551         if (notcon)
1552                 apic_printk(APIC_VERBOSE,
1553                         " (apicid-pin) not connected\n");
1554 }
1555
1556 /*
1557  * Set up the timer pin, possibly with the 8259A-master behind.
1558  */
1559 static void __init setup_timer_IRQ0_pin(unsigned int apic_id, unsigned int pin,
1560                                         int vector)
1561 {
1562         struct IO_APIC_route_entry entry;
1563
1564         if (intr_remapping_enabled)
1565                 return;
1566
1567         memset(&entry, 0, sizeof(entry));
1568
1569         /*
1570          * We use logical delivery to get the timer IRQ
1571          * to the first CPU.
1572          */
1573         entry.dest_mode = apic->irq_dest_mode;
1574         entry.mask = 0;                 /* don't mask IRQ for edge */
1575         entry.dest = apic->cpu_mask_to_apicid(apic->target_cpus());
1576         entry.delivery_mode = apic->irq_delivery_mode;
1577         entry.polarity = 0;
1578         entry.trigger = 0;
1579         entry.vector = vector;
1580
1581         /*
1582          * The timer IRQ doesn't have to know that behind the
1583          * scene we may have a 8259A-master in AEOI mode ...
1584          */
1585         set_irq_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq, "edge");
1586
1587         /*
1588          * Add it to the IO-APIC irq-routing table:
1589          */
1590         ioapic_write_entry(apic_id, pin, entry);
1591 }
1592
1593
1594 __apicdebuginit(void) print_IO_APIC(void)
1595 {
1596         int apic, i;
1597         union IO_APIC_reg_00 reg_00;
1598         union IO_APIC_reg_01 reg_01;
1599         union IO_APIC_reg_02 reg_02;
1600         union IO_APIC_reg_03 reg_03;
1601         unsigned long flags;
1602         struct irq_cfg *cfg;
1603         struct irq_desc *desc;
1604         unsigned int irq;
1605
1606         if (apic_verbosity == APIC_QUIET)
1607                 return;
1608
1609         printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
1610         for (i = 0; i < nr_ioapics; i++)
1611                 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
1612                        mp_ioapics[i].apicid, nr_ioapic_registers[i]);
1613
1614         /*
1615          * We are a bit conservative about what we expect.  We have to
1616          * know about every hardware change ASAP.
1617          */
1618         printk(KERN_INFO "testing the IO APIC.......................\n");
1619
1620         for (apic = 0; apic < nr_ioapics; apic++) {
1621
1622         spin_lock_irqsave(&ioapic_lock, flags);
1623         reg_00.raw = io_apic_read(apic, 0);
1624         reg_01.raw = io_apic_read(apic, 1);
1625         if (reg_01.bits.version >= 0x10)
1626                 reg_02.raw = io_apic_read(apic, 2);
1627         if (reg_01.bits.version >= 0x20)
1628                 reg_03.raw = io_apic_read(apic, 3);
1629         spin_unlock_irqrestore(&ioapic_lock, flags);
1630
1631         printk("\n");
1632         printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].apicid);
1633         printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
1634         printk(KERN_DEBUG ".......    : physical APIC id: %02X\n", reg_00.bits.ID);
1635         printk(KERN_DEBUG ".......    : Delivery Type: %X\n", reg_00.bits.delivery_type);
1636         printk(KERN_DEBUG ".......    : LTS          : %X\n", reg_00.bits.LTS);
1637
1638         printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)&reg_01);
1639         printk(KERN_DEBUG ".......     : max redirection entries: %04X\n", reg_01.bits.entries);
1640
1641         printk(KERN_DEBUG ".......     : PRQ implemented: %X\n", reg_01.bits.PRQ);
1642         printk(KERN_DEBUG ".......     : IO APIC version: %04X\n", reg_01.bits.version);
1643
1644         /*
1645          * Some Intel chipsets with IO APIC VERSION of 0x1? don't have reg_02,
1646          * but the value of reg_02 is read as the previous read register
1647          * value, so ignore it if reg_02 == reg_01.
1648          */
1649         if (reg_01.bits.version >= 0x10 && reg_02.raw != reg_01.raw) {
1650                 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
1651                 printk(KERN_DEBUG ".......     : arbitration: %02X\n", reg_02.bits.arbitration);
1652         }
1653
1654         /*
1655          * Some Intel chipsets with IO APIC VERSION of 0x2? don't have reg_02
1656          * or reg_03, but the value of reg_0[23] is read as the previous read
1657          * register value, so ignore it if reg_03 == reg_0[12].
1658          */
1659         if (reg_01.bits.version >= 0x20 && reg_03.raw != reg_02.raw &&
1660             reg_03.raw != reg_01.raw) {
1661                 printk(KERN_DEBUG ".... register #03: %08X\n", reg_03.raw);
1662                 printk(KERN_DEBUG ".......     : Boot DT    : %X\n", reg_03.bits.boot_DT);
1663         }
1664
1665         printk(KERN_DEBUG ".... IRQ redirection table:\n");
1666
1667         printk(KERN_DEBUG " NR Dst Mask Trig IRR Pol"
1668                           " Stat Dmod Deli Vect:   \n");
1669
1670         for (i = 0; i <= reg_01.bits.entries; i++) {
1671                 struct IO_APIC_route_entry entry;
1672
1673                 entry = ioapic_read_entry(apic, i);
1674
1675                 printk(KERN_DEBUG " %02x %03X ",
1676                         i,
1677                         entry.dest
1678                 );
1679
1680                 printk("%1d    %1d    %1d   %1d   %1d    %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         printk(KERN_DEBUG "IRQ to pin mappings:\n");
1693         for_each_irq_desc(irq, desc) {
1694                 struct irq_pin_list *entry;
1695
1696                 cfg = desc->chip_data;
1697                 entry = cfg->irq_2_pin;
1698                 if (!entry)
1699                         continue;
1700                 printk(KERN_DEBUG "IRQ%d ", irq);
1701                 for (;;) {
1702                         printk("-> %d:%d", entry->apic, entry->pin);
1703                         if (!entry->next)
1704                                 break;
1705                         entry = entry->next;
1706                 }
1707                 printk("\n");
1708         }
1709
1710         printk(KERN_INFO ".................................... done.\n");
1711
1712         return;
1713 }
1714
1715 __apicdebuginit(void) print_APIC_bitfield(int base)
1716 {
1717         unsigned int v;
1718         int i, j;
1719
1720         if (apic_verbosity == APIC_QUIET)
1721                 return;
1722
1723         printk(KERN_DEBUG "0123456789abcdef0123456789abcdef\n" KERN_DEBUG);
1724         for (i = 0; i < 8; i++) {
1725                 v = apic_read(base + i*0x10);
1726                 for (j = 0; j < 32; j++) {
1727                         if (v & (1<<j))
1728                                 printk("1");
1729                         else
1730                                 printk("0");
1731                 }
1732                 printk("\n");
1733         }
1734 }
1735
1736 __apicdebuginit(void) print_local_APIC(void *dummy)
1737 {
1738         unsigned int i, v, ver, maxlvt;
1739         u64 icr;
1740
1741         if (apic_verbosity == APIC_QUIET)
1742                 return;
1743
1744         printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1745                 smp_processor_id(), hard_smp_processor_id());
1746         v = apic_read(APIC_ID);
1747         printk(KERN_INFO "... APIC ID:      %08x (%01x)\n", v, read_apic_id());
1748         v = apic_read(APIC_LVR);
1749         printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1750         ver = GET_APIC_VERSION(v);
1751         maxlvt = lapic_get_maxlvt();
1752
1753         v = apic_read(APIC_TASKPRI);
1754         printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1755
1756         if (APIC_INTEGRATED(ver)) {                     /* !82489DX */
1757                 if (!APIC_XAPIC(ver)) {
1758                         v = apic_read(APIC_ARBPRI);
1759                         printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
1760                                v & APIC_ARBPRI_MASK);
1761                 }
1762                 v = apic_read(APIC_PROCPRI);
1763                 printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1764         }
1765
1766         /*
1767          * Remote read supported only in the 82489DX and local APIC for
1768          * Pentium processors.
1769          */
1770         if (!APIC_INTEGRATED(ver) || maxlvt == 3) {
1771                 v = apic_read(APIC_RRR);
1772                 printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1773         }
1774
1775         v = apic_read(APIC_LDR);
1776         printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1777         if (!x2apic_enabled()) {
1778                 v = apic_read(APIC_DFR);
1779                 printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1780         }
1781         v = apic_read(APIC_SPIV);
1782         printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1783
1784         printk(KERN_DEBUG "... APIC ISR field:\n");
1785         print_APIC_bitfield(APIC_ISR);
1786         printk(KERN_DEBUG "... APIC TMR field:\n");
1787         print_APIC_bitfield(APIC_TMR);
1788         printk(KERN_DEBUG "... APIC IRR field:\n");
1789         print_APIC_bitfield(APIC_IRR);
1790
1791         if (APIC_INTEGRATED(ver)) {             /* !82489DX */
1792                 if (maxlvt > 3)         /* Due to the Pentium erratum 3AP. */
1793                         apic_write(APIC_ESR, 0);
1794
1795                 v = apic_read(APIC_ESR);
1796                 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1797         }
1798
1799         icr = apic_icr_read();
1800         printk(KERN_DEBUG "... APIC ICR: %08x\n", (u32)icr);
1801         printk(KERN_DEBUG "... APIC ICR2: %08x\n", (u32)(icr >> 32));
1802
1803         v = apic_read(APIC_LVTT);
1804         printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1805
1806         if (maxlvt > 3) {                       /* PC is LVT#4. */
1807                 v = apic_read(APIC_LVTPC);
1808                 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1809         }
1810         v = apic_read(APIC_LVT0);
1811         printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1812         v = apic_read(APIC_LVT1);
1813         printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1814
1815         if (maxlvt > 2) {                       /* ERR is LVT#3. */
1816                 v = apic_read(APIC_LVTERR);
1817                 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1818         }
1819
1820         v = apic_read(APIC_TMICT);
1821         printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1822         v = apic_read(APIC_TMCCT);
1823         printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1824         v = apic_read(APIC_TDCR);
1825         printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1826
1827         if (boot_cpu_has(X86_FEATURE_EXTAPIC)) {
1828                 v = apic_read(APIC_EFEAT);
1829                 maxlvt = (v >> 16) & 0xff;
1830                 printk(KERN_DEBUG "... APIC EFEAT: %08x\n", v);
1831                 v = apic_read(APIC_ECTRL);
1832                 printk(KERN_DEBUG "... APIC ECTRL: %08x\n", v);
1833                 for (i = 0; i < maxlvt; i++) {
1834                         v = apic_read(APIC_EILVTn(i));
1835                         printk(KERN_DEBUG "... APIC EILVT%d: %08x\n", i, v);
1836                 }
1837         }
1838         printk("\n");
1839 }
1840
1841 __apicdebuginit(void) print_all_local_APICs(void)
1842 {
1843         int cpu;
1844
1845         preempt_disable();
1846         for_each_online_cpu(cpu)
1847                 smp_call_function_single(cpu, print_local_APIC, NULL, 1);
1848         preempt_enable();
1849 }
1850
1851 __apicdebuginit(void) print_PIC(void)
1852 {
1853         unsigned int v;
1854         unsigned long flags;
1855
1856         if (apic_verbosity == APIC_QUIET)
1857                 return;
1858
1859         printk(KERN_DEBUG "\nprinting PIC contents\n");
1860
1861         spin_lock_irqsave(&i8259A_lock, flags);
1862
1863         v = inb(0xa1) << 8 | inb(0x21);
1864         printk(KERN_DEBUG "... PIC  IMR: %04x\n", v);
1865
1866         v = inb(0xa0) << 8 | inb(0x20);
1867         printk(KERN_DEBUG "... PIC  IRR: %04x\n", v);
1868
1869         outb(0x0b,0xa0);
1870         outb(0x0b,0x20);
1871         v = inb(0xa0) << 8 | inb(0x20);
1872         outb(0x0a,0xa0);
1873         outb(0x0a,0x20);
1874
1875         spin_unlock_irqrestore(&i8259A_lock, flags);
1876
1877         printk(KERN_DEBUG "... PIC  ISR: %04x\n", v);
1878
1879         v = inb(0x4d1) << 8 | inb(0x4d0);
1880         printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1881 }
1882
1883 __apicdebuginit(int) print_all_ICs(void)
1884 {
1885         print_PIC();
1886
1887         /* don't print out if apic is not there */
1888         if (!cpu_has_apic || disable_apic)
1889                 return 0;
1890
1891         print_all_local_APICs();
1892         print_IO_APIC();
1893
1894         return 0;
1895 }
1896
1897 fs_initcall(print_all_ICs);
1898
1899
1900 /* Where if anywhere is the i8259 connect in external int mode */
1901 static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
1902
1903 void __init enable_IO_APIC(void)
1904 {
1905         union IO_APIC_reg_01 reg_01;
1906         int i8259_apic, i8259_pin;
1907         int apic;
1908         unsigned long flags;
1909
1910         /*
1911          * The number of IO-APIC IRQ registers (== #pins):
1912          */
1913         for (apic = 0; apic < nr_ioapics; apic++) {
1914                 spin_lock_irqsave(&ioapic_lock, flags);
1915                 reg_01.raw = io_apic_read(apic, 1);
1916                 spin_unlock_irqrestore(&ioapic_lock, flags);
1917                 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
1918         }
1919         for(apic = 0; apic < nr_ioapics; apic++) {
1920                 int pin;
1921                 /* See if any of the pins is in ExtINT mode */
1922                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1923                         struct IO_APIC_route_entry entry;
1924                         entry = ioapic_read_entry(apic, pin);
1925
1926                         /* If the interrupt line is enabled and in ExtInt mode
1927                          * I have found the pin where the i8259 is connected.
1928                          */
1929                         if ((entry.mask == 0) && (entry.delivery_mode == dest_ExtINT)) {
1930                                 ioapic_i8259.apic = apic;
1931                                 ioapic_i8259.pin  = pin;
1932                                 goto found_i8259;
1933                         }
1934                 }
1935         }
1936  found_i8259:
1937         /* Look to see what if the MP table has reported the ExtINT */
1938         /* If we could not find the appropriate pin by looking at the ioapic
1939          * the i8259 probably is not connected the ioapic but give the
1940          * mptable a chance anyway.
1941          */
1942         i8259_pin  = find_isa_irq_pin(0, mp_ExtINT);
1943         i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
1944         /* Trust the MP table if nothing is setup in the hardware */
1945         if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
1946                 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
1947                 ioapic_i8259.pin  = i8259_pin;
1948                 ioapic_i8259.apic = i8259_apic;
1949         }
1950         /* Complain if the MP table and the hardware disagree */
1951         if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
1952                 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
1953         {
1954                 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
1955         }
1956
1957         /*
1958          * Do not trust the IO-APIC being empty at bootup
1959          */
1960         clear_IO_APIC();
1961 }
1962
1963 /*
1964  * Not an __init, needed by the reboot code
1965  */
1966 void disable_IO_APIC(void)
1967 {
1968         /*
1969          * Clear the IO-APIC before rebooting:
1970          */
1971         clear_IO_APIC();
1972
1973         /*
1974          * If the i8259 is routed through an IOAPIC
1975          * Put that IOAPIC in virtual wire mode
1976          * so legacy interrupts can be delivered.
1977          *
1978          * With interrupt-remapping, for now we will use virtual wire A mode,
1979          * as virtual wire B is little complex (need to configure both
1980          * IOAPIC RTE aswell as interrupt-remapping table entry).
1981          * As this gets called during crash dump, keep this simple for now.
1982          */
1983         if (ioapic_i8259.pin != -1 && !intr_remapping_enabled) {
1984                 struct IO_APIC_route_entry entry;
1985
1986                 memset(&entry, 0, sizeof(entry));
1987                 entry.mask            = 0; /* Enabled */
1988                 entry.trigger         = 0; /* Edge */
1989                 entry.irr             = 0;
1990                 entry.polarity        = 0; /* High */
1991                 entry.delivery_status = 0;
1992                 entry.dest_mode       = 0; /* Physical */
1993                 entry.delivery_mode   = dest_ExtINT; /* ExtInt */
1994                 entry.vector          = 0;
1995                 entry.dest            = read_apic_id();
1996
1997                 /*
1998                  * Add it to the IO-APIC irq-routing table:
1999                  */
2000                 ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
2001         }
2002
2003         /*
2004          * Use virtual wire A mode when interrupt remapping is enabled.
2005          */
2006         disconnect_bsp_APIC(!intr_remapping_enabled && ioapic_i8259.pin != -1);
2007 }
2008
2009 #ifdef CONFIG_X86_32
2010 /*
2011  * function to set the IO-APIC physical IDs based on the
2012  * values stored in the MPC table.
2013  *
2014  * by Matt Domsch <Matt_Domsch@dell.com>  Tue Dec 21 12:25:05 CST 1999
2015  */
2016
2017 static void __init setup_ioapic_ids_from_mpc(void)
2018 {
2019         union IO_APIC_reg_00 reg_00;
2020         physid_mask_t phys_id_present_map;
2021         int apic_id;
2022         int i;
2023         unsigned char old_id;
2024         unsigned long flags;
2025
2026         if (x86_quirks->setup_ioapic_ids && x86_quirks->setup_ioapic_ids())
2027                 return;
2028
2029         /*
2030          * Don't check I/O APIC IDs for xAPIC systems.  They have
2031          * no meaning without the serial APIC bus.
2032          */
2033         if (!(boot_cpu_data.x86_vendor == X86_VENDOR_INTEL)
2034                 || APIC_XAPIC(apic_version[boot_cpu_physical_apicid]))
2035                 return;
2036         /*
2037          * This is broken; anything with a real cpu count has to
2038          * circumvent this idiocy regardless.
2039          */
2040         phys_id_present_map = apic->ioapic_phys_id_map(phys_cpu_present_map);
2041
2042         /*
2043          * Set the IOAPIC ID to the value stored in the MPC table.
2044          */
2045         for (apic_id = 0; apic_id < nr_ioapics; apic_id++) {
2046
2047                 /* Read the register 0 value */
2048                 spin_lock_irqsave(&ioapic_lock, flags);
2049                 reg_00.raw = io_apic_read(apic_id, 0);
2050                 spin_unlock_irqrestore(&ioapic_lock, flags);
2051
2052                 old_id = mp_ioapics[apic_id].apicid;
2053
2054                 if (mp_ioapics[apic_id].apicid >= get_physical_broadcast()) {
2055                         printk(KERN_ERR "BIOS bug, IO-APIC#%d ID is %d in the MPC table!...\n",
2056                                 apic_id, mp_ioapics[apic_id].apicid);
2057                         printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
2058                                 reg_00.bits.ID);
2059                         mp_ioapics[apic_id].apicid = reg_00.bits.ID;
2060                 }
2061
2062                 /*
2063                  * Sanity check, is the ID really free? Every APIC in a
2064                  * system must have a unique ID or we get lots of nice
2065                  * 'stuck on smp_invalidate_needed IPI wait' messages.
2066                  */
2067                 if (apic->check_apicid_used(phys_id_present_map,
2068                                         mp_ioapics[apic_id].apicid)) {
2069                         printk(KERN_ERR "BIOS bug, IO-APIC#%d ID %d is already used!...\n",
2070                                 apic_id, mp_ioapics[apic_id].apicid);
2071                         for (i = 0; i < get_physical_broadcast(); i++)
2072                                 if (!physid_isset(i, phys_id_present_map))
2073                                         break;
2074                         if (i >= get_physical_broadcast())
2075                                 panic("Max APIC ID exceeded!\n");
2076                         printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
2077                                 i);
2078                         physid_set(i, phys_id_present_map);
2079                         mp_ioapics[apic_id].apicid = i;
2080                 } else {
2081                         physid_mask_t tmp;
2082                         tmp = apic->apicid_to_cpu_present(mp_ioapics[apic_id].apicid);
2083                         apic_printk(APIC_VERBOSE, "Setting %d in the "
2084                                         "phys_id_present_map\n",
2085                                         mp_ioapics[apic_id].apicid);
2086                         physids_or(phys_id_present_map, phys_id_present_map, tmp);
2087                 }
2088
2089
2090                 /*
2091                  * We need to adjust the IRQ routing table
2092                  * if the ID changed.
2093                  */
2094                 if (old_id != mp_ioapics[apic_id].apicid)
2095                         for (i = 0; i < mp_irq_entries; i++)
2096                                 if (mp_irqs[i].dstapic == old_id)
2097                                         mp_irqs[i].dstapic
2098                                                 = mp_ioapics[apic_id].apicid;
2099
2100                 /*
2101                  * Read the right value from the MPC table and
2102                  * write it into the ID register.
2103                  */
2104                 apic_printk(APIC_VERBOSE, KERN_INFO
2105                         "...changing IO-APIC physical APIC ID to %d ...",
2106                         mp_ioapics[apic_id].apicid);
2107
2108                 reg_00.bits.ID = mp_ioapics[apic_id].apicid;
2109                 spin_lock_irqsave(&ioapic_lock, flags);
2110                 io_apic_write(apic_id, 0, reg_00.raw);
2111                 spin_unlock_irqrestore(&ioapic_lock, flags);
2112
2113                 /*
2114                  * Sanity check
2115                  */
2116                 spin_lock_irqsave(&ioapic_lock, flags);
2117                 reg_00.raw = io_apic_read(apic_id, 0);
2118                 spin_unlock_irqrestore(&ioapic_lock, flags);
2119                 if (reg_00.bits.ID != mp_ioapics[apic_id].apicid)
2120                         printk("could not set ID!\n");
2121                 else
2122                         apic_printk(APIC_VERBOSE, " ok.\n");
2123         }
2124 }
2125 #endif
2126
2127 int no_timer_check __initdata;
2128
2129 static int __init notimercheck(char *s)
2130 {
2131         no_timer_check = 1;
2132         return 1;
2133 }
2134 __setup("no_timer_check", notimercheck);
2135
2136 /*
2137  * There is a nasty bug in some older SMP boards, their mptable lies
2138  * about the timer IRQ. We do the following to work around the situation:
2139  *
2140  *      - timer IRQ defaults to IO-APIC IRQ
2141  *      - if this function detects that timer IRQs are defunct, then we fall
2142  *        back to ISA timer IRQs
2143  */
2144 static int __init timer_irq_works(void)
2145 {
2146         unsigned long t1 = jiffies;
2147         unsigned long flags;
2148
2149         if (no_timer_check)
2150                 return 1;
2151
2152         local_save_flags(flags);
2153         local_irq_enable();
2154         /* Let ten ticks pass... */
2155         mdelay((10 * 1000) / HZ);
2156         local_irq_restore(flags);
2157
2158         /*
2159          * Expect a few ticks at least, to be sure some possible
2160          * glue logic does not lock up after one or two first
2161          * ticks in a non-ExtINT mode.  Also the local APIC
2162          * might have cached one ExtINT interrupt.  Finally, at
2163          * least one tick may be lost due to delays.
2164          */
2165
2166         /* jiffies wrap? */
2167         if (time_after(jiffies, t1 + 4))
2168                 return 1;
2169         return 0;
2170 }
2171
2172 /*
2173  * In the SMP+IOAPIC case it might happen that there are an unspecified
2174  * number of pending IRQ events unhandled. These cases are very rare,
2175  * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
2176  * better to do it this way as thus we do not have to be aware of
2177  * 'pending' interrupts in the IRQ path, except at this point.
2178  */
2179 /*
2180  * Edge triggered needs to resend any interrupt
2181  * that was delayed but this is now handled in the device
2182  * independent code.
2183  */
2184
2185 /*
2186  * Starting up a edge-triggered IO-APIC interrupt is
2187  * nasty - we need to make sure that we get the edge.
2188  * If it is already asserted for some reason, we need
2189  * return 1 to indicate that is was pending.
2190  *
2191  * This is not complete - we should be able to fake
2192  * an edge even if it isn't on the 8259A...
2193  */
2194
2195 static unsigned int startup_ioapic_irq(unsigned int irq)
2196 {
2197         int was_pending = 0;
2198         unsigned long flags;
2199         struct irq_cfg *cfg;
2200
2201         spin_lock_irqsave(&ioapic_lock, flags);
2202         if (irq < NR_IRQS_LEGACY) {
2203                 disable_8259A_irq(irq);
2204                 if (i8259A_irq_pending(irq))
2205                         was_pending = 1;
2206         }
2207         cfg = irq_cfg(irq);
2208         __unmask_IO_APIC_irq(cfg);
2209         spin_unlock_irqrestore(&ioapic_lock, flags);
2210
2211         return was_pending;
2212 }
2213
2214 #ifdef CONFIG_X86_64
2215 static int ioapic_retrigger_irq(unsigned int irq)
2216 {
2217
2218         struct irq_cfg *cfg = irq_cfg(irq);
2219         unsigned long flags;
2220
2221         spin_lock_irqsave(&vector_lock, flags);
2222         apic->send_IPI_mask(cpumask_of(cpumask_first(cfg->domain)), cfg->vector);
2223         spin_unlock_irqrestore(&vector_lock, flags);
2224
2225         return 1;
2226 }
2227 #else
2228 static int ioapic_retrigger_irq(unsigned int irq)
2229 {
2230         apic->send_IPI_self(irq_cfg(irq)->vector);
2231
2232         return 1;
2233 }
2234 #endif
2235
2236 /*
2237  * Level and edge triggered IO-APIC interrupts need different handling,
2238  * so we use two separate IRQ descriptors. Edge triggered IRQs can be
2239  * handled with the level-triggered descriptor, but that one has slightly
2240  * more overhead. Level-triggered interrupts cannot be handled with the
2241  * edge-triggered handler, without risking IRQ storms and other ugly
2242  * races.
2243  */
2244
2245 #ifdef CONFIG_SMP
2246 static void send_cleanup_vector(struct irq_cfg *cfg)
2247 {
2248         cpumask_var_t cleanup_mask;
2249
2250         if (unlikely(!alloc_cpumask_var(&cleanup_mask, GFP_ATOMIC))) {
2251                 unsigned int i;
2252                 cfg->move_cleanup_count = 0;
2253                 for_each_cpu_and(i, cfg->old_domain, cpu_online_mask)
2254                         cfg->move_cleanup_count++;
2255                 for_each_cpu_and(i, cfg->old_domain, cpu_online_mask)
2256                         apic->send_IPI_mask(cpumask_of(i), IRQ_MOVE_CLEANUP_VECTOR);
2257         } else {
2258                 cpumask_and(cleanup_mask, cfg->old_domain, cpu_online_mask);
2259                 cfg->move_cleanup_count = cpumask_weight(cleanup_mask);
2260                 apic->send_IPI_mask(cleanup_mask, IRQ_MOVE_CLEANUP_VECTOR);
2261                 free_cpumask_var(cleanup_mask);
2262         }
2263         cfg->move_in_progress = 0;
2264 }
2265
2266 static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, struct irq_cfg *cfg)
2267 {
2268         int apic, pin;
2269         struct irq_pin_list *entry;
2270         u8 vector = cfg->vector;
2271
2272         entry = cfg->irq_2_pin;
2273         for (;;) {
2274                 unsigned int reg;
2275
2276                 if (!entry)
2277                         break;
2278
2279                 apic = entry->apic;
2280                 pin = entry->pin;
2281                 /*
2282                  * With interrupt-remapping, destination information comes
2283                  * from interrupt-remapping table entry.
2284                  */
2285                 if (!irq_remapped(irq))
2286                         io_apic_write(apic, 0x11 + pin*2, dest);
2287                 reg = io_apic_read(apic, 0x10 + pin*2);
2288                 reg &= ~IO_APIC_REDIR_VECTOR_MASK;
2289                 reg |= vector;
2290                 io_apic_modify(apic, 0x10 + pin*2, reg);
2291                 if (!entry->next)
2292                         break;
2293                 entry = entry->next;
2294         }
2295 }
2296
2297 static int
2298 assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask);
2299
2300 /*
2301  * Either sets desc->affinity to a valid value, and returns
2302  * ->cpu_mask_to_apicid of that, or returns BAD_APICID and
2303  * leaves desc->affinity untouched.
2304  */
2305 static unsigned int
2306 set_desc_affinity(struct irq_desc *desc, const struct cpumask *mask)
2307 {
2308         struct irq_cfg *cfg;
2309         unsigned int irq;
2310
2311         if (!cpumask_intersects(mask, cpu_online_mask))
2312                 return BAD_APICID;
2313
2314         irq = desc->irq;
2315         cfg = desc->chip_data;
2316         if (assign_irq_vector(irq, cfg, mask))
2317                 return BAD_APICID;
2318
2319         cpumask_copy(desc->affinity, mask);
2320
2321         return apic->cpu_mask_to_apicid_and(desc->affinity, cfg->domain);
2322 }
2323
2324 static int
2325 set_ioapic_affinity_irq_desc(struct irq_desc *desc, const struct cpumask *mask)
2326 {
2327         struct irq_cfg *cfg;
2328         unsigned long flags;
2329         unsigned int dest;
2330         unsigned int irq;
2331         int ret = -1;
2332
2333         irq = desc->irq;
2334         cfg = desc->chip_data;
2335
2336         spin_lock_irqsave(&ioapic_lock, flags);
2337         dest = set_desc_affinity(desc, mask);
2338         if (dest != BAD_APICID) {
2339                 /* Only the high 8 bits are valid. */
2340                 dest = SET_APIC_LOGICAL_ID(dest);
2341                 __target_IO_APIC_irq(irq, dest, cfg);
2342                 ret = 0;
2343         }
2344         spin_unlock_irqrestore(&ioapic_lock, flags);
2345
2346         return ret;
2347 }
2348
2349 static int
2350 set_ioapic_affinity_irq(unsigned int irq, const struct cpumask *mask)
2351 {
2352         struct irq_desc *desc;
2353
2354         desc = irq_to_desc(irq);
2355
2356         return set_ioapic_affinity_irq_desc(desc, mask);
2357 }
2358
2359 #ifdef CONFIG_INTR_REMAP
2360
2361 /*
2362  * Migrate the IO-APIC irq in the presence of intr-remapping.
2363  *
2364  * For both level and edge triggered, irq migration is a simple atomic
2365  * update(of vector and cpu destination) of IRTE and flush the hardware cache.
2366  *
2367  * For level triggered, we eliminate the io-apic RTE modification (with the
2368  * updated vector information), by using a virtual vector (io-apic pin number).
2369  * Real vector that is used for interrupting cpu will be coming from
2370  * the interrupt-remapping table entry.
2371  */
2372 static int
2373 migrate_ioapic_irq_desc(struct irq_desc *desc, const struct cpumask *mask)
2374 {
2375         struct irq_cfg *cfg;
2376         struct irte irte;
2377         unsigned int dest;
2378         unsigned int irq;
2379         int ret = -1;
2380
2381         if (!cpumask_intersects(mask, cpu_online_mask))
2382                 return ret;
2383
2384         irq = desc->irq;
2385         if (get_irte(irq, &irte))
2386                 return ret;
2387
2388         cfg = desc->chip_data;
2389         if (assign_irq_vector(irq, cfg, mask))
2390                 return ret;
2391
2392         dest = apic->cpu_mask_to_apicid_and(cfg->domain, mask);
2393
2394         irte.vector = cfg->vector;
2395         irte.dest_id = IRTE_DEST(dest);
2396
2397         /*
2398          * Modified the IRTE and flushes the Interrupt entry cache.
2399          */
2400         modify_irte(irq, &irte);
2401
2402         if (cfg->move_in_progress)
2403                 send_cleanup_vector(cfg);
2404
2405         cpumask_copy(desc->affinity, mask);
2406
2407         return 0;
2408 }
2409
2410 /*
2411  * Migrates the IRQ destination in the process context.
2412  */
2413 static int set_ir_ioapic_affinity_irq_desc(struct irq_desc *desc,
2414                                             const struct cpumask *mask)
2415 {
2416         return migrate_ioapic_irq_desc(desc, mask);
2417 }
2418 static int set_ir_ioapic_affinity_irq(unsigned int irq,
2419                                        const struct cpumask *mask)
2420 {
2421         struct irq_desc *desc = irq_to_desc(irq);
2422
2423         return set_ir_ioapic_affinity_irq_desc(desc, mask);
2424 }
2425 #else
2426 static inline int set_ir_ioapic_affinity_irq_desc(struct irq_desc *desc,
2427                                                    const struct cpumask *mask)
2428 {
2429         return 0;
2430 }
2431 #endif
2432
2433 asmlinkage void smp_irq_move_cleanup_interrupt(void)
2434 {
2435         unsigned vector, me;
2436
2437         ack_APIC_irq();
2438         exit_idle();
2439         irq_enter();
2440
2441         me = smp_processor_id();
2442         for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
2443                 unsigned int irq;
2444                 unsigned int irr;
2445                 struct irq_desc *desc;
2446                 struct irq_cfg *cfg;
2447                 irq = __get_cpu_var(vector_irq)[vector];
2448
2449                 if (irq == -1)
2450                         continue;
2451
2452                 desc = irq_to_desc(irq);
2453                 if (!desc)
2454                         continue;
2455
2456                 cfg = irq_cfg(irq);
2457                 spin_lock(&desc->lock);
2458                 if (!cfg->move_cleanup_count)
2459                         goto unlock;
2460
2461                 if (vector == cfg->vector && cpumask_test_cpu(me, cfg->domain))
2462                         goto unlock;
2463
2464                 irr = apic_read(APIC_IRR + (vector / 32 * 0x10));
2465                 /*
2466                  * Check if the vector that needs to be cleanedup is
2467                  * registered at the cpu's IRR. If so, then this is not
2468                  * the best time to clean it up. Lets clean it up in the
2469                  * next attempt by sending another IRQ_MOVE_CLEANUP_VECTOR
2470                  * to myself.
2471                  */
2472                 if (irr  & (1 << (vector % 32))) {
2473                         apic->send_IPI_self(IRQ_MOVE_CLEANUP_VECTOR);
2474                         goto unlock;
2475                 }
2476                 __get_cpu_var(vector_irq)[vector] = -1;
2477                 cfg->move_cleanup_count--;
2478 unlock:
2479                 spin_unlock(&desc->lock);
2480         }
2481
2482         irq_exit();
2483 }
2484
2485 static void irq_complete_move(struct irq_desc **descp)
2486 {
2487         struct irq_desc *desc = *descp;
2488         struct irq_cfg *cfg = desc->chip_data;
2489         unsigned vector, me;
2490
2491         if (likely(!cfg->move_in_progress))
2492                 return;
2493
2494         vector = ~get_irq_regs()->orig_ax;
2495         me = smp_processor_id();
2496
2497         if (vector == cfg->vector && cpumask_test_cpu(me, cfg->domain))
2498                 send_cleanup_vector(cfg);
2499 }
2500 #else
2501 static inline void irq_complete_move(struct irq_desc **descp) {}
2502 #endif
2503
2504 static void ack_apic_edge(unsigned int irq)
2505 {
2506         struct irq_desc *desc = irq_to_desc(irq);
2507
2508         irq_complete_move(&desc);
2509         move_native_irq(irq);
2510         ack_APIC_irq();
2511 }
2512
2513 atomic_t irq_mis_count;
2514
2515 static void ack_apic_level(unsigned int irq)
2516 {
2517         struct irq_desc *desc = irq_to_desc(irq);
2518
2519 #ifdef CONFIG_X86_32
2520         unsigned long v;
2521         int i;
2522 #endif
2523         struct irq_cfg *cfg;
2524         int do_unmask_irq = 0;
2525
2526         irq_complete_move(&desc);
2527 #ifdef CONFIG_GENERIC_PENDING_IRQ
2528         /* If we are moving the irq we need to mask it */
2529         if (unlikely(desc->status & IRQ_MOVE_PENDING)) {
2530                 do_unmask_irq = 1;
2531                 mask_IO_APIC_irq_desc(desc);
2532         }
2533 #endif
2534
2535 #ifdef CONFIG_X86_32
2536         /*
2537         * It appears there is an erratum which affects at least version 0x11
2538         * of I/O APIC (that's the 82093AA and cores integrated into various
2539         * chipsets).  Under certain conditions a level-triggered interrupt is
2540         * erroneously delivered as edge-triggered one but the respective IRR
2541         * bit gets set nevertheless.  As a result the I/O unit expects an EOI
2542         * message but it will never arrive and further interrupts are blocked
2543         * from the source.  The exact reason is so far unknown, but the
2544         * phenomenon was observed when two consecutive interrupt requests
2545         * from a given source get delivered to the same CPU and the source is
2546         * temporarily disabled in between.
2547         *
2548         * A workaround is to simulate an EOI message manually.  We achieve it
2549         * by setting the trigger mode to edge and then to level when the edge
2550         * trigger mode gets detected in the TMR of a local APIC for a
2551         * level-triggered interrupt.  We mask the source for the time of the
2552         * operation to prevent an edge-triggered interrupt escaping meanwhile.
2553         * The idea is from Manfred Spraul.  --macro
2554         */
2555         cfg = desc->chip_data;
2556         i = cfg->vector;
2557
2558         v = apic_read(APIC_TMR + ((i & ~0x1f) >> 1));
2559 #endif
2560
2561         /*
2562          * We must acknowledge the irq before we move it or the acknowledge will
2563          * not propagate properly.
2564          */
2565         ack_APIC_irq();
2566
2567         /* Now we can move and renable the irq */
2568         if (unlikely(do_unmask_irq)) {
2569                 /* Only migrate the irq if the ack has been received.
2570                  *
2571                  * On rare occasions the broadcast level triggered ack gets
2572                  * delayed going to ioapics, and if we reprogram the
2573                  * vector while Remote IRR is still set the irq will never
2574                  * fire again.
2575                  *
2576                  * To prevent this scenario we read the Remote IRR bit
2577                  * of the ioapic.  This has two effects.
2578                  * - On any sane system the read of the ioapic will
2579                  *   flush writes (and acks) going to the ioapic from
2580                  *   this cpu.
2581                  * - We get to see if the ACK has actually been delivered.
2582                  *
2583                  * Based on failed experiments of reprogramming the
2584                  * ioapic entry from outside of irq context starting
2585                  * with masking the ioapic entry and then polling until
2586                  * Remote IRR was clear before reprogramming the
2587                  * ioapic I don't trust the Remote IRR bit to be
2588                  * completey accurate.
2589                  *
2590                  * However there appears to be no other way to plug
2591                  * this race, so if the Remote IRR bit is not
2592                  * accurate and is causing problems then it is a hardware bug
2593                  * and you can go talk to the chipset vendor about it.
2594                  */
2595                 cfg = desc->chip_data;
2596                 if (!io_apic_level_ack_pending(cfg))
2597                         move_masked_irq(irq);
2598                 unmask_IO_APIC_irq_desc(desc);
2599         }
2600
2601 #ifdef CONFIG_X86_32
2602         if (!(v & (1 << (i & 0x1f)))) {
2603                 atomic_inc(&irq_mis_count);
2604                 spin_lock(&ioapic_lock);
2605                 __mask_and_edge_IO_APIC_irq(cfg);
2606                 __unmask_and_level_IO_APIC_irq(cfg);
2607                 spin_unlock(&ioapic_lock);
2608         }
2609 #endif
2610 }
2611
2612 #ifdef CONFIG_INTR_REMAP
2613 static void __eoi_ioapic_irq(unsigned int irq, struct irq_cfg *cfg)
2614 {
2615         int apic, pin;
2616         struct irq_pin_list *entry;
2617
2618         entry = cfg->irq_2_pin;
2619         for (;;) {
2620
2621                 if (!entry)
2622                         break;
2623
2624                 apic = entry->apic;
2625                 pin = entry->pin;
2626                 io_apic_eoi(apic, pin);
2627                 entry = entry->next;
2628         }
2629 }
2630
2631 static void
2632 eoi_ioapic_irq(struct irq_desc *desc)
2633 {
2634         struct irq_cfg *cfg;
2635         unsigned long flags;
2636         unsigned int irq;
2637
2638         irq = desc->irq;
2639         cfg = desc->chip_data;
2640
2641         spin_lock_irqsave(&ioapic_lock, flags);
2642         __eoi_ioapic_irq(irq, cfg);
2643         spin_unlock_irqrestore(&ioapic_lock, flags);
2644 }
2645
2646 static void ir_ack_apic_edge(unsigned int irq)
2647 {
2648         ack_APIC_irq();
2649 }
2650
2651 static void ir_ack_apic_level(unsigned int irq)
2652 {
2653         struct irq_desc *desc = irq_to_desc(irq);
2654
2655         ack_APIC_irq();
2656         eoi_ioapic_irq(desc);
2657 }
2658 #endif /* CONFIG_INTR_REMAP */
2659
2660 static struct irq_chip ioapic_chip __read_mostly = {
2661         .name           = "IO-APIC",
2662         .startup        = startup_ioapic_irq,
2663         .mask           = mask_IO_APIC_irq,
2664         .unmask         = unmask_IO_APIC_irq,
2665         .ack            = ack_apic_edge,
2666         .eoi            = ack_apic_level,
2667 #ifdef CONFIG_SMP
2668         .set_affinity   = set_ioapic_affinity_irq,
2669 #endif
2670         .retrigger      = ioapic_retrigger_irq,
2671 };
2672
2673 static struct irq_chip ir_ioapic_chip __read_mostly = {
2674         .name           = "IR-IO-APIC",
2675         .startup        = startup_ioapic_irq,
2676         .mask           = mask_IO_APIC_irq,
2677         .unmask         = unmask_IO_APIC_irq,
2678 #ifdef CONFIG_INTR_REMAP
2679         .ack            = ir_ack_apic_edge,
2680         .eoi            = ir_ack_apic_level,
2681 #ifdef CONFIG_SMP
2682         .set_affinity   = set_ir_ioapic_affinity_irq,
2683 #endif
2684 #endif
2685         .retrigger      = ioapic_retrigger_irq,
2686 };
2687
2688 static inline void init_IO_APIC_traps(void)
2689 {
2690         int irq;
2691         struct irq_desc *desc;
2692         struct irq_cfg *cfg;
2693
2694         /*
2695          * NOTE! The local APIC isn't very good at handling
2696          * multiple interrupts at the same interrupt level.
2697          * As the interrupt level is determined by taking the
2698          * vector number and shifting that right by 4, we
2699          * want to spread these out a bit so that they don't
2700          * all fall in the same interrupt level.
2701          *
2702          * Also, we've got to be careful not to trash gate
2703          * 0x80, because int 0x80 is hm, kind of importantish. ;)
2704          */
2705         for_each_irq_desc(irq, desc) {
2706                 cfg = desc->chip_data;
2707                 if (IO_APIC_IRQ(irq) && cfg && !cfg->vector) {
2708                         /*
2709                          * Hmm.. We don't have an entry for this,
2710                          * so default to an old-fashioned 8259
2711                          * interrupt if we can..
2712                          */
2713                         if (irq < NR_IRQS_LEGACY)
2714                                 make_8259A_irq(irq);
2715                         else
2716                                 /* Strange. Oh, well.. */
2717                                 desc->chip = &no_irq_chip;
2718                 }
2719         }
2720 }
2721
2722 /*
2723  * The local APIC irq-chip implementation:
2724  */
2725
2726 static void mask_lapic_irq(unsigned int irq)
2727 {
2728         unsigned long v;
2729
2730         v = apic_read(APIC_LVT0);
2731         apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
2732 }
2733
2734 static void unmask_lapic_irq(unsigned int irq)
2735 {
2736         unsigned long v;
2737
2738         v = apic_read(APIC_LVT0);
2739         apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
2740 }
2741
2742 static void ack_lapic_irq(unsigned int irq)
2743 {
2744         ack_APIC_irq();
2745 }
2746
2747 static struct irq_chip lapic_chip __read_mostly = {
2748         .name           = "local-APIC",
2749         .mask           = mask_lapic_irq,
2750         .unmask         = unmask_lapic_irq,
2751         .ack            = ack_lapic_irq,
2752 };
2753
2754 static void lapic_register_intr(int irq, struct irq_desc *desc)
2755 {
2756         desc->status &= ~IRQ_LEVEL;
2757         set_irq_chip_and_handler_name(irq, &lapic_chip, handle_edge_irq,
2758                                       "edge");
2759 }
2760
2761 static void __init setup_nmi(void)
2762 {
2763         /*
2764          * Dirty trick to enable the NMI watchdog ...
2765          * We put the 8259A master into AEOI mode and
2766          * unmask on all local APICs LVT0 as NMI.
2767          *
2768          * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
2769          * is from Maciej W. Rozycki - so we do not have to EOI from
2770          * the NMI handler or the timer interrupt.
2771          */
2772         apic_printk(APIC_VERBOSE, KERN_INFO "activating NMI Watchdog ...");
2773
2774         enable_NMI_through_LVT0();
2775
2776         apic_printk(APIC_VERBOSE, " done.\n");
2777 }
2778
2779 /*
2780  * This looks a bit hackish but it's about the only one way of sending
2781  * a few INTA cycles to 8259As and any associated glue logic.  ICR does
2782  * not support the ExtINT mode, unfortunately.  We need to send these
2783  * cycles as some i82489DX-based boards have glue logic that keeps the
2784  * 8259A interrupt line asserted until INTA.  --macro
2785  */
2786 static inline void __init unlock_ExtINT_logic(void)
2787 {
2788         int apic, pin, i;
2789         struct IO_APIC_route_entry entry0, entry1;
2790         unsigned char save_control, save_freq_select;
2791
2792         pin  = find_isa_irq_pin(8, mp_INT);
2793         if (pin == -1) {
2794                 WARN_ON_ONCE(1);
2795                 return;
2796         }
2797         apic = find_isa_irq_apic(8, mp_INT);
2798         if (apic == -1) {
2799                 WARN_ON_ONCE(1);
2800                 return;
2801         }
2802
2803         entry0 = ioapic_read_entry(apic, pin);
2804         clear_IO_APIC_pin(apic, pin);
2805
2806         memset(&entry1, 0, sizeof(entry1));
2807
2808         entry1.dest_mode = 0;                   /* physical delivery */
2809         entry1.mask = 0;                        /* unmask IRQ now */
2810         entry1.dest = hard_smp_processor_id();
2811         entry1.delivery_mode = dest_ExtINT;
2812         entry1.polarity = entry0.polarity;
2813         entry1.trigger = 0;
2814         entry1.vector = 0;
2815
2816         ioapic_write_entry(apic, pin, entry1);
2817
2818         save_control = CMOS_READ(RTC_CONTROL);
2819         save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
2820         CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
2821                    RTC_FREQ_SELECT);
2822         CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
2823
2824         i = 100;
2825         while (i-- > 0) {
2826                 mdelay(10);
2827                 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
2828                         i -= 10;
2829         }
2830
2831         CMOS_WRITE(save_control, RTC_CONTROL);
2832         CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
2833         clear_IO_APIC_pin(apic, pin);
2834
2835         ioapic_write_entry(apic, pin, entry0);
2836 }
2837
2838 static int disable_timer_pin_1 __initdata;
2839 /* Actually the next is obsolete, but keep it for paranoid reasons -AK */
2840 static int __init disable_timer_pin_setup(char *arg)
2841 {
2842         disable_timer_pin_1 = 1;
2843         return 0;
2844 }
2845 early_param("disable_timer_pin_1", disable_timer_pin_setup);
2846
2847 int timer_through_8259 __initdata;
2848
2849 /*
2850  * This code may look a bit paranoid, but it's supposed to cooperate with
2851  * a wide range of boards and BIOS bugs.  Fortunately only the timer IRQ
2852  * is so screwy.  Thanks to Brian Perkins for testing/hacking this beast
2853  * fanatically on his truly buggy board.
2854  *
2855  * FIXME: really need to revamp this for all platforms.
2856  */
2857 static inline void __init check_timer(void)
2858 {
2859         struct irq_desc *desc = irq_to_desc(0);
2860         struct irq_cfg *cfg = desc->chip_data;
2861         int node = cpu_to_node(boot_cpu_id);
2862         int apic1, pin1, apic2, pin2;
2863         unsigned long flags;
2864         int no_pin1 = 0;
2865
2866         local_irq_save(flags);
2867
2868         /*
2869          * get/set the timer IRQ vector:
2870          */
2871         disable_8259A_irq(0);
2872         assign_irq_vector(0, cfg, apic->target_cpus());
2873
2874         /*
2875          * As IRQ0 is to be enabled in the 8259A, the virtual
2876          * wire has to be disabled in the local APIC.  Also
2877          * timer interrupts need to be acknowledged manually in
2878          * the 8259A for the i82489DX when using the NMI
2879          * watchdog as that APIC treats NMIs as level-triggered.
2880          * The AEOI mode will finish them in the 8259A
2881          * automatically.
2882          */
2883         apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
2884         init_8259A(1);
2885 #ifdef CONFIG_X86_32
2886         {
2887                 unsigned int ver;
2888
2889                 ver = apic_read(APIC_LVR);
2890                 ver = GET_APIC_VERSION(ver);
2891                 timer_ack = (nmi_watchdog == NMI_IO_APIC && !APIC_INTEGRATED(ver));
2892         }
2893 #endif
2894
2895         pin1  = find_isa_irq_pin(0, mp_INT);
2896         apic1 = find_isa_irq_apic(0, mp_INT);
2897         pin2  = ioapic_i8259.pin;
2898         apic2 = ioapic_i8259.apic;
2899
2900         apic_printk(APIC_QUIET, KERN_INFO "..TIMER: vector=0x%02X "
2901                     "apic1=%d pin1=%d apic2=%d pin2=%d\n",
2902                     cfg->vector, apic1, pin1, apic2, pin2);
2903
2904         /*
2905          * Some BIOS writers are clueless and report the ExtINTA
2906          * I/O APIC input from the cascaded 8259A as the timer
2907          * interrupt input.  So just in case, if only one pin
2908          * was found above, try it both directly and through the
2909          * 8259A.
2910          */
2911         if (pin1 == -1) {
2912                 if (intr_remapping_enabled)
2913                         panic("BIOS bug: timer not connected to IO-APIC");
2914                 pin1 = pin2;
2915                 apic1 = apic2;
2916                 no_pin1 = 1;
2917         } else if (pin2 == -1) {
2918                 pin2 = pin1;
2919                 apic2 = apic1;
2920         }
2921
2922         if (pin1 != -1) {
2923                 /*
2924                  * Ok, does IRQ0 through the IOAPIC work?
2925                  */
2926                 if (no_pin1) {
2927                         add_pin_to_irq_node(cfg, node, apic1, pin1);
2928                         setup_timer_IRQ0_pin(apic1, pin1, cfg->vector);
2929                 } else {
2930                         /* for edge trigger, setup_IO_APIC_irq already
2931                          * leave it unmasked.
2932                          * so only need to unmask if it is level-trigger
2933                          * do we really have level trigger timer?
2934                          */
2935                         int idx;
2936                         idx = find_irq_entry(apic1, pin1, mp_INT);
2937                         if (idx != -1 && irq_trigger(idx))
2938                                 unmask_IO_APIC_irq_desc(desc);
2939                 }
2940                 if (timer_irq_works()) {
2941                         if (nmi_watchdog == NMI_IO_APIC) {
2942                                 setup_nmi();
2943                                 enable_8259A_irq(0);
2944                         }
2945                         if (disable_timer_pin_1 > 0)
2946                                 clear_IO_APIC_pin(0, pin1);
2947                         goto out;
2948                 }
2949                 if (intr_remapping_enabled)
2950                         panic("timer doesn't work through Interrupt-remapped IO-APIC");
2951                 local_irq_disable();
2952                 clear_IO_APIC_pin(apic1, pin1);
2953                 if (!no_pin1)
2954                         apic_printk(APIC_QUIET, KERN_ERR "..MP-BIOS bug: "
2955                                     "8254 timer not connected to IO-APIC\n");
2956
2957                 apic_printk(APIC_QUIET, KERN_INFO "...trying to set up timer "
2958                             "(IRQ0) through the 8259A ...\n");
2959                 apic_printk(APIC_QUIET, KERN_INFO
2960                             "..... (found apic %d pin %d) ...\n", apic2, pin2);
2961                 /*
2962                  * legacy devices should be connected to IO APIC #0
2963                  */
2964                 replace_pin_at_irq_node(cfg, node, apic1, pin1, apic2, pin2);
2965                 setup_timer_IRQ0_pin(apic2, pin2, cfg->vector);
2966                 enable_8259A_irq(0);
2967                 if (timer_irq_works()) {
2968                         apic_printk(APIC_QUIET, KERN_INFO "....... works.\n");
2969                         timer_through_8259 = 1;
2970                         if (nmi_watchdog == NMI_IO_APIC) {
2971                                 disable_8259A_irq(0);
2972                                 setup_nmi();
2973                                 enable_8259A_irq(0);
2974                         }
2975                         goto out;
2976                 }
2977                 /*
2978                  * Cleanup, just in case ...
2979                  */
2980                 local_irq_disable();
2981                 disable_8259A_irq(0);
2982                 clear_IO_APIC_pin(apic2, pin2);
2983                 apic_printk(APIC_QUIET, KERN_INFO "....... failed.\n");
2984         }
2985
2986         if (nmi_watchdog == NMI_IO_APIC) {
2987                 apic_printk(APIC_QUIET, KERN_WARNING "timer doesn't work "
2988                             "through the IO-APIC - disabling NMI Watchdog!\n");
2989                 nmi_watchdog = NMI_NONE;
2990         }
2991 #ifdef CONFIG_X86_32
2992         timer_ack = 0;
2993 #endif
2994
2995         apic_printk(APIC_QUIET, KERN_INFO
2996                     "...trying to set up timer as Virtual Wire IRQ...\n");
2997
2998         lapic_register_intr(0, desc);
2999         apic_write(APIC_LVT0, APIC_DM_FIXED | cfg->vector);     /* Fixed mode */
3000         enable_8259A_irq(0);
3001
3002         if (timer_irq_works()) {
3003                 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
3004                 goto out;
3005         }
3006         local_irq_disable();
3007         disable_8259A_irq(0);
3008         apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | cfg->vector);
3009         apic_printk(APIC_QUIET, KERN_INFO "..... failed.\n");
3010
3011         apic_printk(APIC_QUIET, KERN_INFO
3012                     "...trying to set up timer as ExtINT IRQ...\n");
3013
3014         init_8259A(0);
3015         make_8259A_irq(0);
3016         apic_write(APIC_LVT0, APIC_DM_EXTINT);
3017
3018         unlock_ExtINT_logic();
3019
3020         if (timer_irq_works()) {
3021                 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
3022                 goto out;
3023         }
3024         local_irq_disable();
3025         apic_printk(APIC_QUIET, KERN_INFO "..... failed :(.\n");
3026         panic("IO-APIC + timer doesn't work!  Boot with apic=debug and send a "
3027                 "report.  Then try booting with the 'noapic' option.\n");
3028 out:
3029         local_irq_restore(flags);
3030 }
3031
3032 /*
3033  * Traditionally ISA IRQ2 is the cascade IRQ, and is not available
3034  * to devices.  However there may be an I/O APIC pin available for
3035  * this interrupt regardless.  The pin may be left unconnected, but
3036  * typically it will be reused as an ExtINT cascade interrupt for
3037  * the master 8259A.  In the MPS case such a pin will normally be
3038  * reported as an ExtINT interrupt in the MP table.  With ACPI
3039  * there is no provision for ExtINT interrupts, and in the absence
3040  * of an override it would be treated as an ordinary ISA I/O APIC
3041  * interrupt, that is edge-triggered and unmasked by default.  We
3042  * used to do this, but it caused problems on some systems because
3043  * of the NMI watchdog and sometimes IRQ0 of the 8254 timer using
3044  * the same ExtINT cascade interrupt to drive the local APIC of the
3045  * bootstrap processor.  Therefore we refrain from routing IRQ2 to
3046  * the I/O APIC in all cases now.  No actual device should request
3047  * it anyway.  --macro
3048  */
3049 #define PIC_IRQS        (1 << PIC_CASCADE_IR)
3050
3051 void __init setup_IO_APIC(void)
3052 {
3053
3054         /*
3055          * calling enable_IO_APIC() is moved to setup_local_APIC for BP
3056          */
3057
3058         io_apic_irqs = ~PIC_IRQS;
3059
3060         apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
3061         /*
3062          * Set up IO-APIC IRQ routing.
3063          */
3064 #ifdef CONFIG_X86_32
3065         if (!acpi_ioapic)
3066                 setup_ioapic_ids_from_mpc();
3067 #endif
3068         sync_Arb_IDs();
3069         setup_IO_APIC_irqs();
3070         init_IO_APIC_traps();
3071         check_timer();
3072 }
3073
3074 /*
3075  *      Called after all the initialization is done. If we didnt find any
3076  *      APIC bugs then we can allow the modify fast path
3077  */
3078
3079 static int __init io_apic_bug_finalize(void)
3080 {
3081         if (sis_apic_bug == -1)
3082                 sis_apic_bug = 0;
3083         return 0;
3084 }
3085
3086 late_initcall(io_apic_bug_finalize);
3087
3088 struct sysfs_ioapic_data {
3089         struct sys_device dev;
3090         struct IO_APIC_route_entry entry[0];
3091 };
3092 static struct sysfs_ioapic_data * mp_ioapic_data[MAX_IO_APICS];
3093
3094 static int ioapic_suspend(struct sys_device *dev, pm_message_t state)
3095 {
3096         struct IO_APIC_route_entry *entry;
3097         struct sysfs_ioapic_data *data;
3098         int i;
3099
3100         data = container_of(dev, struct sysfs_ioapic_data, dev);
3101         entry = data->entry;
3102         for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ )
3103                 *entry = ioapic_read_entry(dev->id, i);
3104
3105         return 0;
3106 }
3107
3108 static int ioapic_resume(struct sys_device *dev)
3109 {
3110         struct IO_APIC_route_entry *entry;
3111         struct sysfs_ioapic_data *data;
3112         unsigned long flags;
3113         union IO_APIC_reg_00 reg_00;
3114         int i;
3115
3116         data = container_of(dev, struct sysfs_ioapic_data, dev);
3117         entry = data->entry;
3118
3119         spin_lock_irqsave(&ioapic_lock, flags);
3120         reg_00.raw = io_apic_read(dev->id, 0);
3121         if (reg_00.bits.ID != mp_ioapics[dev->id].apicid) {
3122                 reg_00.bits.ID = mp_ioapics[dev->id].apicid;
3123                 io_apic_write(dev->id, 0, reg_00.raw);
3124         }
3125         spin_unlock_irqrestore(&ioapic_lock, flags);
3126         for (i = 0; i < nr_ioapic_registers[dev->id]; i++)
3127                 ioapic_write_entry(dev->id, i, entry[i]);
3128
3129         return 0;
3130 }
3131
3132 static struct sysdev_class ioapic_sysdev_class = {
3133         .name = "ioapic",
3134         .suspend = ioapic_suspend,
3135         .resume = ioapic_resume,
3136 };
3137
3138 static int __init ioapic_init_sysfs(void)
3139 {
3140         struct sys_device * dev;
3141         int i, size, error;
3142
3143         error = sysdev_class_register(&ioapic_sysdev_class);
3144         if (error)
3145                 return error;
3146
3147         for (i = 0; i < nr_ioapics; i++ ) {
3148                 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
3149                         * sizeof(struct IO_APIC_route_entry);
3150                 mp_ioapic_data[i] = kzalloc(size, GFP_KERNEL);
3151                 if (!mp_ioapic_data[i]) {
3152                         printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
3153                         continue;
3154                 }
3155                 dev = &mp_ioapic_data[i]->dev;
3156                 dev->id = i;
3157                 dev->cls = &ioapic_sysdev_class;
3158                 error = sysdev_register(dev);
3159                 if (error) {
3160                         kfree(mp_ioapic_data[i]);
3161                         mp_ioapic_data[i] = NULL;
3162                         printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
3163                         continue;
3164                 }
3165         }
3166
3167         return 0;
3168 }
3169
3170 device_initcall(ioapic_init_sysfs);
3171
3172 static int nr_irqs_gsi = NR_IRQS_LEGACY;
3173 /*
3174  * Dynamic irq allocate and deallocation
3175  */
3176 unsigned int create_irq_nr(unsigned int irq_want, int node)
3177 {
3178         /* Allocate an unused irq */
3179         unsigned int irq;
3180         unsigned int new;
3181         unsigned long flags;
3182         struct irq_cfg *cfg_new = NULL;
3183         struct irq_desc *desc_new = NULL;
3184
3185         irq = 0;
3186         if (irq_want < nr_irqs_gsi)
3187                 irq_want = nr_irqs_gsi;
3188
3189         spin_lock_irqsave(&vector_lock, flags);
3190         for (new = irq_want; new < nr_irqs; new++) {
3191                 desc_new = irq_to_desc_alloc_node(new, node);
3192                 if (!desc_new) {
3193                         printk(KERN_INFO "can not get irq_desc for %d\n", new);
3194                         continue;
3195                 }
3196                 cfg_new = desc_new->chip_data;
3197
3198                 if (cfg_new->vector != 0)
3199                         continue;
3200
3201                 desc_new = move_irq_desc(desc_new, node);
3202
3203                 if (__assign_irq_vector(new, cfg_new, apic->target_cpus()) == 0)
3204                         irq = new;
3205                 break;
3206         }
3207         spin_unlock_irqrestore(&vector_lock, flags);
3208
3209         if (irq > 0) {
3210                 dynamic_irq_init(irq);
3211                 /* restore it, in case dynamic_irq_init clear it */
3212                 if (desc_new)
3213                         desc_new->chip_data = cfg_new;
3214         }
3215         return irq;
3216 }
3217
3218 int create_irq(void)
3219 {
3220         int node = cpu_to_node(boot_cpu_id);
3221         unsigned int irq_want;
3222         int irq;
3223
3224         irq_want = nr_irqs_gsi;
3225         irq = create_irq_nr(irq_want, node);
3226
3227         if (irq == 0)
3228                 irq = -1;
3229
3230         return irq;
3231 }
3232
3233 void destroy_irq(unsigned int irq)
3234 {
3235         unsigned long flags;
3236         struct irq_cfg *cfg;
3237         struct irq_desc *desc;
3238
3239         /* store it, in case dynamic_irq_cleanup clear it */
3240         desc = irq_to_desc(irq);
3241         cfg = desc->chip_data;
3242         dynamic_irq_cleanup(irq);
3243         /* connect back irq_cfg */
3244         if (desc)
3245                 desc->chip_data = cfg;
3246
3247         free_irte(irq);
3248         spin_lock_irqsave(&vector_lock, flags);
3249         __clear_irq_vector(irq, cfg);
3250         spin_unlock_irqrestore(&vector_lock, flags);
3251 }
3252
3253 /*
3254  * MSI message composition
3255  */
3256 #ifdef CONFIG_PCI_MSI
3257 static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq, struct msi_msg *msg)
3258 {
3259         struct irq_cfg *cfg;
3260         int err;
3261         unsigned dest;
3262
3263         if (disable_apic)
3264                 return -ENXIO;
3265
3266         cfg = irq_cfg(irq);
3267         err = assign_irq_vector(irq, cfg, apic->target_cpus());
3268         if (err)
3269                 return err;
3270
3271         dest = apic->cpu_mask_to_apicid_and(cfg->domain, apic->target_cpus());
3272
3273         if (irq_remapped(irq)) {
3274                 struct irte irte;
3275                 int ir_index;
3276                 u16 sub_handle;
3277
3278                 ir_index = map_irq_to_irte_handle(irq, &sub_handle);
3279                 BUG_ON(ir_index == -1);
3280
3281                 memset (&irte, 0, sizeof(irte));
3282
3283                 irte.present = 1;
3284                 irte.dst_mode = apic->irq_dest_mode;
3285                 irte.trigger_mode = 0; /* edge */
3286                 irte.dlvry_mode = apic->irq_delivery_mode;
3287                 irte.vector = cfg->vector;
3288                 irte.dest_id = IRTE_DEST(dest);
3289
3290                 modify_irte(irq, &irte);
3291
3292                 msg->address_hi = MSI_ADDR_BASE_HI;
3293                 msg->data = sub_handle;
3294                 msg->address_lo = MSI_ADDR_BASE_LO | MSI_ADDR_IR_EXT_INT |
3295                                   MSI_ADDR_IR_SHV |
3296                                   MSI_ADDR_IR_INDEX1(ir_index) |
3297                                   MSI_ADDR_IR_INDEX2(ir_index);
3298         } else {
3299                 if (x2apic_enabled())
3300                         msg->address_hi = MSI_ADDR_BASE_HI |
3301                                           MSI_ADDR_EXT_DEST_ID(dest);
3302                 else
3303                         msg->address_hi = MSI_ADDR_BASE_HI;
3304
3305                 msg->address_lo =
3306                         MSI_ADDR_BASE_LO |
3307                         ((apic->irq_dest_mode == 0) ?
3308                                 MSI_ADDR_DEST_MODE_PHYSICAL:
3309                                 MSI_ADDR_DEST_MODE_LOGICAL) |
3310                         ((apic->irq_delivery_mode != dest_LowestPrio) ?
3311                                 MSI_ADDR_REDIRECTION_CPU:
3312                                 MSI_ADDR_REDIRECTION_LOWPRI) |
3313                         MSI_ADDR_DEST_ID(dest);
3314
3315                 msg->data =
3316                         MSI_DATA_TRIGGER_EDGE |
3317                         MSI_DATA_LEVEL_ASSERT |
3318                         ((apic->irq_delivery_mode != dest_LowestPrio) ?
3319                                 MSI_DATA_DELIVERY_FIXED:
3320                                 MSI_DATA_DELIVERY_LOWPRI) |
3321                         MSI_DATA_VECTOR(cfg->vector);
3322         }
3323         return err;
3324 }
3325
3326 #ifdef CONFIG_SMP
3327 static int set_msi_irq_affinity(unsigned int irq, const struct cpumask *mask)
3328 {
3329         struct irq_desc *desc = irq_to_desc(irq);
3330         struct irq_cfg *cfg;
3331         struct msi_msg msg;
3332         unsigned int dest;
3333
3334         dest = set_desc_affinity(desc, mask);
3335         if (dest == BAD_APICID)
3336                 return -1;
3337
3338         cfg = desc->chip_data;
3339
3340         read_msi_msg_desc(desc, &msg);
3341
3342         msg.data &= ~MSI_DATA_VECTOR_MASK;
3343         msg.data |= MSI_DATA_VECTOR(cfg->vector);
3344         msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3345         msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3346
3347         write_msi_msg_desc(desc, &msg);
3348
3349         return 0;
3350 }
3351 #ifdef CONFIG_INTR_REMAP
3352 /*
3353  * Migrate the MSI irq to another cpumask. This migration is
3354  * done in the process context using interrupt-remapping hardware.
3355  */
3356 static int
3357 ir_set_msi_irq_affinity(unsigned int irq, const struct cpumask *mask)
3358 {
3359         struct irq_desc *desc = irq_to_desc(irq);
3360         struct irq_cfg *cfg = desc->chip_data;
3361         unsigned int dest;
3362         struct irte irte;
3363
3364         if (get_irte(irq, &irte))
3365                 return -1;
3366
3367         dest = set_desc_affinity(desc, mask);
3368         if (dest == BAD_APICID)
3369                 return -1;
3370
3371         irte.vector = cfg->vector;
3372         irte.dest_id = IRTE_DEST(dest);
3373
3374         /*
3375          * atomically update the IRTE with the new destination and vector.
3376          */
3377         modify_irte(irq, &irte);
3378
3379         /*
3380          * After this point, all the interrupts will start arriving
3381          * at the new destination. So, time to cleanup the previous
3382          * vector allocation.
3383          */
3384         if (cfg->move_in_progress)
3385                 send_cleanup_vector(cfg);
3386
3387         return 0;
3388 }
3389
3390 #endif
3391 #endif /* CONFIG_SMP */
3392
3393 /*
3394  * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
3395  * which implement the MSI or MSI-X Capability Structure.
3396  */
3397 static struct irq_chip msi_chip = {
3398         .name           = "PCI-MSI",
3399         .unmask         = unmask_msi_irq,
3400         .mask           = mask_msi_irq,
3401         .ack            = ack_apic_edge,
3402 #ifdef CONFIG_SMP
3403         .set_affinity   = set_msi_irq_affinity,
3404 #endif
3405         .retrigger      = ioapic_retrigger_irq,
3406 };
3407
3408 static struct irq_chip msi_ir_chip = {
3409         .name           = "IR-PCI-MSI",
3410         .unmask         = unmask_msi_irq,
3411         .mask           = mask_msi_irq,
3412 #ifdef CONFIG_INTR_REMAP
3413         .ack            = ir_ack_apic_edge,
3414 #ifdef CONFIG_SMP
3415         .set_affinity   = ir_set_msi_irq_affinity,
3416 #endif
3417 #endif
3418         .retrigger      = ioapic_retrigger_irq,
3419 };
3420
3421 /*
3422  * Map the PCI dev to the corresponding remapping hardware unit
3423  * and allocate 'nvec' consecutive interrupt-remapping table entries
3424  * in it.
3425  */
3426 static int msi_alloc_irte(struct pci_dev *dev, int irq, int nvec)
3427 {
3428         struct intel_iommu *iommu;
3429         int index;
3430
3431         iommu = map_dev_to_ir(dev);
3432         if (!iommu) {
3433                 printk(KERN_ERR
3434                        "Unable to map PCI %s to iommu\n", pci_name(dev));
3435                 return -ENOENT;
3436         }
3437
3438         index = alloc_irte(iommu, irq, nvec);
3439         if (index < 0) {
3440                 printk(KERN_ERR
3441                        "Unable to allocate %d IRTE for PCI %s\n", nvec,
3442                        pci_name(dev));
3443                 return -ENOSPC;
3444         }
3445         return index;
3446 }
3447
3448 static int setup_msi_irq(struct pci_dev *dev, struct msi_desc *msidesc, int irq)
3449 {
3450         int ret;
3451         struct msi_msg msg;
3452
3453         ret = msi_compose_msg(dev, irq, &msg);
3454         if (ret < 0)
3455                 return ret;
3456
3457         set_irq_msi(irq, msidesc);
3458         write_msi_msg(irq, &msg);
3459
3460         if (irq_remapped(irq)) {
3461                 struct irq_desc *desc = irq_to_desc(irq);
3462                 /*
3463                  * irq migration in process context
3464                  */
3465                 desc->status |= IRQ_MOVE_PCNTXT;
3466                 set_irq_chip_and_handler_name(irq, &msi_ir_chip, handle_edge_irq, "edge");
3467         } else
3468                 set_irq_chip_and_handler_name(irq, &msi_chip, handle_edge_irq, "edge");
3469
3470         dev_printk(KERN_DEBUG, &dev->dev, "irq %d for MSI/MSI-X\n", irq);
3471
3472         return 0;
3473 }
3474
3475 int arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
3476 {
3477         unsigned int irq;
3478         int ret, sub_handle;
3479         struct msi_desc *msidesc;
3480         unsigned int irq_want;
3481         struct intel_iommu *iommu = NULL;
3482         int index = 0;
3483         int node;
3484
3485         /* x86 doesn't support multiple MSI yet */
3486         if (type == PCI_CAP_ID_MSI && nvec > 1)
3487                 return 1;
3488
3489         node = dev_to_node(&dev->dev);
3490         irq_want = nr_irqs_gsi;
3491         sub_handle = 0;
3492         list_for_each_entry(msidesc, &dev->msi_list, list) {
3493                 irq = create_irq_nr(irq_want, node);
3494                 if (irq == 0)
3495                         return -1;
3496                 irq_want = irq + 1;
3497                 if (!intr_remapping_enabled)
3498                         goto no_ir;
3499
3500                 if (!sub_handle) {
3501                         /*
3502                          * allocate the consecutive block of IRTE's
3503                          * for 'nvec'
3504                          */
3505                         index = msi_alloc_irte(dev, irq, nvec);
3506                         if (index < 0) {
3507                                 ret = index;
3508                                 goto error;
3509                         }
3510                 } else {
3511                         iommu = map_dev_to_ir(dev);
3512                         if (!iommu) {
3513                                 ret = -ENOENT;
3514                                 goto error;
3515                         }
3516                         /*
3517                          * setup the mapping between the irq and the IRTE
3518                          * base index, the sub_handle pointing to the
3519                          * appropriate interrupt remap table entry.
3520                          */
3521                         set_irte_irq(irq, iommu, index, sub_handle);
3522                 }
3523 no_ir:
3524                 ret = setup_msi_irq(dev, msidesc, irq);
3525                 if (ret < 0)
3526                         goto error;
3527                 sub_handle++;
3528         }
3529         return 0;
3530
3531 error:
3532         destroy_irq(irq);
3533         return ret;
3534 }
3535
3536 void arch_teardown_msi_irq(unsigned int irq)
3537 {
3538         destroy_irq(irq);
3539 }
3540
3541 #if defined (CONFIG_DMAR) || defined (CONFIG_INTR_REMAP)
3542 #ifdef CONFIG_SMP
3543 static int dmar_msi_set_affinity(unsigned int irq, const struct cpumask *mask)
3544 {
3545         struct irq_desc *desc = irq_to_desc(irq);
3546         struct irq_cfg *cfg;
3547         struct msi_msg msg;
3548         unsigned int dest;
3549
3550         dest = set_desc_affinity(desc, mask);
3551         if (dest == BAD_APICID)
3552                 return -1;
3553
3554         cfg = desc->chip_data;
3555
3556         dmar_msi_read(irq, &msg);
3557
3558         msg.data &= ~MSI_DATA_VECTOR_MASK;
3559         msg.data |= MSI_DATA_VECTOR(cfg->vector);
3560         msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3561         msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3562
3563         dmar_msi_write(irq, &msg);
3564
3565         return 0;
3566 }
3567
3568 #endif /* CONFIG_SMP */
3569
3570 struct irq_chip dmar_msi_type = {
3571         .name = "DMAR_MSI",
3572         .unmask = dmar_msi_unmask,
3573         .mask = dmar_msi_mask,
3574         .ack = ack_apic_edge,
3575 #ifdef CONFIG_SMP
3576         .set_affinity = dmar_msi_set_affinity,
3577 #endif
3578         .retrigger = ioapic_retrigger_irq,
3579 };
3580
3581 int arch_setup_dmar_msi(unsigned int irq)
3582 {
3583         int ret;
3584         struct msi_msg msg;
3585
3586         ret = msi_compose_msg(NULL, irq, &msg);
3587         if (ret < 0)
3588                 return ret;
3589         dmar_msi_write(irq, &msg);
3590         set_irq_chip_and_handler_name(irq, &dmar_msi_type, handle_edge_irq,
3591                 "edge");
3592         return 0;
3593 }
3594 #endif
3595
3596 #ifdef CONFIG_HPET_TIMER
3597
3598 #ifdef CONFIG_SMP
3599 static int hpet_msi_set_affinity(unsigned int irq, const struct cpumask *mask)
3600 {
3601         struct irq_desc *desc = irq_to_desc(irq);
3602         struct irq_cfg *cfg;
3603         struct msi_msg msg;
3604         unsigned int dest;
3605
3606         dest = set_desc_affinity(desc, mask);
3607         if (dest == BAD_APICID)
3608                 return -1;
3609
3610         cfg = desc->chip_data;
3611
3612         hpet_msi_read(irq, &msg);
3613
3614         msg.data &= ~MSI_DATA_VECTOR_MASK;
3615         msg.data |= MSI_DATA_VECTOR(cfg->vector);
3616         msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3617         msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3618
3619         hpet_msi_write(irq, &msg);
3620
3621         return 0;
3622 }
3623
3624 #endif /* CONFIG_SMP */
3625
3626 static struct irq_chip hpet_msi_type = {
3627         .name = "HPET_MSI",
3628         .unmask = hpet_msi_unmask,
3629         .mask = hpet_msi_mask,
3630         .ack = ack_apic_edge,
3631 #ifdef CONFIG_SMP
3632         .set_affinity = hpet_msi_set_affinity,
3633 #endif
3634         .retrigger = ioapic_retrigger_irq,
3635 };
3636
3637 int arch_setup_hpet_msi(unsigned int irq)
3638 {
3639         int ret;
3640         struct msi_msg msg;
3641         struct irq_desc *desc = irq_to_desc(irq);
3642
3643         ret = msi_compose_msg(NULL, irq, &msg);
3644         if (ret < 0)
3645                 return ret;
3646
3647         hpet_msi_write(irq, &msg);
3648         desc->status |= IRQ_MOVE_PCNTXT;
3649         set_irq_chip_and_handler_name(irq, &hpet_msi_type, handle_edge_irq,
3650                 "edge");
3651
3652         return 0;
3653 }
3654 #endif
3655
3656 #endif /* CONFIG_PCI_MSI */
3657 /*
3658  * Hypertransport interrupt support
3659  */
3660 #ifdef CONFIG_HT_IRQ
3661
3662 #ifdef CONFIG_SMP
3663
3664 static void target_ht_irq(unsigned int irq, unsigned int dest, u8 vector)
3665 {
3666         struct ht_irq_msg msg;
3667         fetch_ht_irq_msg(irq, &msg);
3668
3669         msg.address_lo &= ~(HT_IRQ_LOW_VECTOR_MASK | HT_IRQ_LOW_DEST_ID_MASK);
3670         msg.address_hi &= ~(HT_IRQ_HIGH_DEST_ID_MASK);
3671
3672         msg.address_lo |= HT_IRQ_LOW_VECTOR(vector) | HT_IRQ_LOW_DEST_ID(dest);
3673         msg.address_hi |= HT_IRQ_HIGH_DEST_ID(dest);
3674
3675         write_ht_irq_msg(irq, &msg);
3676 }
3677
3678 static int set_ht_irq_affinity(unsigned int irq, const struct cpumask *mask)
3679 {
3680         struct irq_desc *desc = irq_to_desc(irq);
3681         struct irq_cfg *cfg;
3682         unsigned int dest;
3683
3684         dest = set_desc_affinity(desc, mask);
3685         if (dest == BAD_APICID)
3686                 return -1;
3687
3688         cfg = desc->chip_data;
3689
3690         target_ht_irq(irq, dest, cfg->vector);
3691
3692         return 0;
3693 }
3694
3695 #endif
3696
3697 static struct irq_chip ht_irq_chip = {
3698         .name           = "PCI-HT",
3699         .mask           = mask_ht_irq,
3700         .unmask         = unmask_ht_irq,
3701         .ack            = ack_apic_edge,
3702 #ifdef CONFIG_SMP
3703         .set_affinity   = set_ht_irq_affinity,
3704 #endif
3705         .retrigger      = ioapic_retrigger_irq,
3706 };
3707
3708 int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev)
3709 {
3710         struct irq_cfg *cfg;
3711         int err;
3712
3713         if (disable_apic)
3714                 return -ENXIO;
3715
3716         cfg = irq_cfg(irq);
3717         err = assign_irq_vector(irq, cfg, apic->target_cpus());
3718         if (!err) {
3719                 struct ht_irq_msg msg;
3720                 unsigned dest;
3721
3722                 dest = apic->cpu_mask_to_apicid_and(cfg->domain,
3723                                                     apic->target_cpus());
3724
3725                 msg.address_hi = HT_IRQ_HIGH_DEST_ID(dest);
3726
3727                 msg.address_lo =
3728                         HT_IRQ_LOW_BASE |
3729                         HT_IRQ_LOW_DEST_ID(dest) |
3730                         HT_IRQ_LOW_VECTOR(cfg->vector) |
3731                         ((apic->irq_dest_mode == 0) ?
3732                                 HT_IRQ_LOW_DM_PHYSICAL :
3733                                 HT_IRQ_LOW_DM_LOGICAL) |
3734                         HT_IRQ_LOW_RQEOI_EDGE |
3735                         ((apic->irq_delivery_mode != dest_LowestPrio) ?
3736                                 HT_IRQ_LOW_MT_FIXED :
3737                                 HT_IRQ_LOW_MT_ARBITRATED) |
3738                         HT_IRQ_LOW_IRQ_MASKED;
3739
3740                 write_ht_irq_msg(irq, &msg);
3741
3742                 set_irq_chip_and_handler_name(irq, &ht_irq_chip,
3743                                               handle_edge_irq, "edge");
3744
3745                 dev_printk(KERN_DEBUG, &dev->dev, "irq %d for HT\n", irq);
3746         }
3747         return err;
3748 }
3749 #endif /* CONFIG_HT_IRQ */
3750
3751 #ifdef CONFIG_X86_UV
3752 /*
3753  * Re-target the irq to the specified CPU and enable the specified MMR located
3754  * on the specified blade to allow the sending of MSIs to the specified CPU.
3755  */
3756 int arch_enable_uv_irq(char *irq_name, unsigned int irq, int cpu, int mmr_blade,
3757                        unsigned long mmr_offset)
3758 {
3759         const struct cpumask *eligible_cpu = cpumask_of(cpu);
3760         struct irq_cfg *cfg;
3761         int mmr_pnode;
3762         unsigned long mmr_value;
3763         struct uv_IO_APIC_route_entry *entry;
3764         unsigned long flags;
3765         int err;
3766
3767         BUILD_BUG_ON(sizeof(struct uv_IO_APIC_route_entry) != sizeof(unsigned long));
3768
3769         cfg = irq_cfg(irq);
3770
3771         err = assign_irq_vector(irq, cfg, eligible_cpu);
3772         if (err != 0)
3773                 return err;
3774
3775         spin_lock_irqsave(&vector_lock, flags);
3776         set_irq_chip_and_handler_name(irq, &uv_irq_chip, handle_percpu_irq,
3777                                       irq_name);
3778         spin_unlock_irqrestore(&vector_lock, flags);
3779
3780         mmr_value = 0;
3781         entry = (struct uv_IO_APIC_route_entry *)&mmr_value;
3782         entry->vector           = cfg->vector;
3783         entry->delivery_mode    = apic->irq_delivery_mode;
3784         entry->dest_mode        = apic->irq_dest_mode;
3785         entry->polarity         = 0;
3786         entry->trigger          = 0;
3787         entry->mask             = 0;
3788         entry->dest             = apic->cpu_mask_to_apicid(eligible_cpu);
3789
3790         mmr_pnode = uv_blade_to_pnode(mmr_blade);
3791         uv_write_global_mmr64(mmr_pnode, mmr_offset, mmr_value);
3792
3793         return irq;
3794 }
3795
3796 /*
3797  * Disable the specified MMR located on the specified blade so that MSIs are
3798  * longer allowed to be sent.
3799  */
3800 void arch_disable_uv_irq(int mmr_blade, unsigned long mmr_offset)
3801 {
3802         unsigned long mmr_value;
3803         struct uv_IO_APIC_route_entry *entry;
3804         int mmr_pnode;
3805
3806         BUILD_BUG_ON(sizeof(struct uv_IO_APIC_route_entry) != sizeof(unsigned long));
3807
3808         mmr_value = 0;
3809         entry = (struct uv_IO_APIC_route_entry *)&mmr_value;
3810         entry->mask = 1;
3811
3812         mmr_pnode = uv_blade_to_pnode(mmr_blade);
3813         uv_write_global_mmr64(mmr_pnode, mmr_offset, mmr_value);
3814 }
3815 #endif /* CONFIG_X86_64 */
3816
3817 int __init io_apic_get_redir_entries (int ioapic)
3818 {
3819         union IO_APIC_reg_01    reg_01;
3820         unsigned long flags;
3821
3822         spin_lock_irqsave(&ioapic_lock, flags);
3823         reg_01.raw = io_apic_read(ioapic, 1);
3824         spin_unlock_irqrestore(&ioapic_lock, flags);
3825
3826         return reg_01.bits.entries;
3827 }
3828
3829 void __init probe_nr_irqs_gsi(void)
3830 {
3831         int nr = 0;
3832
3833         nr = acpi_probe_gsi();
3834         if (nr > nr_irqs_gsi) {
3835                 nr_irqs_gsi = nr;
3836         } else {
3837                 /* for acpi=off or acpi is not compiled in */
3838                 int idx;
3839
3840                 nr = 0;
3841                 for (idx = 0; idx < nr_ioapics; idx++)
3842                         nr += io_apic_get_redir_entries(idx) + 1;
3843
3844                 if (nr > nr_irqs_gsi)
3845                         nr_irqs_gsi = nr;
3846         }
3847
3848         printk(KERN_DEBUG "nr_irqs_gsi: %d\n", nr_irqs_gsi);
3849 }
3850
3851 #ifdef CONFIG_SPARSE_IRQ
3852 int __init arch_probe_nr_irqs(void)
3853 {
3854         int nr;
3855
3856         if (nr_irqs > (NR_VECTORS * nr_cpu_ids))
3857                 nr_irqs = NR_VECTORS * nr_cpu_ids;
3858
3859         nr = nr_irqs_gsi + 8 * nr_cpu_ids;
3860 #if defined(CONFIG_PCI_MSI) || defined(CONFIG_HT_IRQ)
3861         /*
3862          * for MSI and HT dyn irq
3863          */
3864         nr += nr_irqs_gsi * 16;
3865 #endif
3866         if (nr < nr_irqs)
3867                 nr_irqs = nr;
3868
3869         return 0;
3870 }
3871 #endif
3872
3873 static int __io_apic_set_pci_routing(struct device *dev, int irq,
3874                                 struct io_apic_irq_attr *irq_attr)
3875 {
3876         struct irq_desc *desc;
3877         struct irq_cfg *cfg;
3878         int node;
3879         int ioapic, pin;
3880         int trigger, polarity;
3881
3882         ioapic = irq_attr->ioapic;
3883         if (!IO_APIC_IRQ(irq)) {
3884                 apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
3885                         ioapic);
3886                 return -EINVAL;
3887         }
3888
3889         if (dev)
3890                 node = dev_to_node(dev);
3891         else
3892                 node = cpu_to_node(boot_cpu_id);
3893
3894         desc = irq_to_desc_alloc_node(irq, node);
3895         if (!desc) {
3896                 printk(KERN_INFO "can not get irq_desc %d\n", irq);
3897                 return 0;
3898         }
3899
3900         pin = irq_attr->ioapic_pin;
3901         trigger = irq_attr->trigger;
3902         polarity = irq_attr->polarity;
3903
3904         /*
3905          * IRQs < 16 are already in the irq_2_pin[] map
3906          */
3907         if (irq >= NR_IRQS_LEGACY) {
3908                 cfg = desc->chip_data;
3909                 add_pin_to_irq_node(cfg, node, ioapic, pin);
3910         }
3911
3912         setup_IO_APIC_irq(ioapic, pin, irq, desc, trigger, polarity);
3913
3914         return 0;
3915 }
3916
3917 int io_apic_set_pci_routing(struct device *dev, int irq,
3918                                 struct io_apic_irq_attr *irq_attr)
3919 {
3920         int ioapic, pin;
3921         /*
3922          * Avoid pin reprogramming.  PRTs typically include entries
3923          * with redundant pin->gsi mappings (but unique PCI devices);
3924          * we only program the IOAPIC on the first.
3925          */
3926         ioapic = irq_attr->ioapic;
3927         pin = irq_attr->ioapic_pin;
3928         if (test_bit(pin, mp_ioapic_routing[ioapic].pin_programmed)) {
3929                 pr_debug("Pin %d-%d already programmed\n",
3930                          mp_ioapics[ioapic].apicid, pin);
3931                 return 0;
3932         }
3933         set_bit(pin, mp_ioapic_routing[ioapic].pin_programmed);
3934
3935         return __io_apic_set_pci_routing(dev, irq, irq_attr);
3936 }
3937
3938 /* --------------------------------------------------------------------------
3939                           ACPI-based IOAPIC Configuration
3940    -------------------------------------------------------------------------- */
3941
3942 #ifdef CONFIG_ACPI
3943
3944 #ifdef CONFIG_X86_32
3945 int __init io_apic_get_unique_id(int ioapic, int apic_id)
3946 {
3947         union IO_APIC_reg_00 reg_00;
3948         static physid_mask_t apic_id_map = PHYSID_MASK_NONE;
3949         physid_mask_t tmp;
3950         unsigned long flags;
3951         int i = 0;
3952
3953         /*
3954          * The P4 platform supports up to 256 APIC IDs on two separate APIC
3955          * buses (one for LAPICs, one for IOAPICs), where predecessors only
3956          * supports up to 16 on one shared APIC bus.
3957          *
3958          * TBD: Expand LAPIC/IOAPIC support on P4-class systems to take full
3959          *      advantage of new APIC bus architecture.
3960          */
3961
3962         if (physids_empty(apic_id_map))
3963                 apic_id_map = apic->ioapic_phys_id_map(phys_cpu_present_map);
3964
3965         spin_lock_irqsave(&ioapic_lock, flags);
3966         reg_00.raw = io_apic_read(ioapic, 0);
3967         spin_unlock_irqrestore(&ioapic_lock, flags);
3968
3969         if (apic_id >= get_physical_broadcast()) {
3970                 printk(KERN_WARNING "IOAPIC[%d]: Invalid apic_id %d, trying "
3971                         "%d\n", ioapic, apic_id, reg_00.bits.ID);
3972                 apic_id = reg_00.bits.ID;
3973         }
3974
3975         /*
3976          * Every APIC in a system must have a unique ID or we get lots of nice
3977          * 'stuck on smp_invalidate_needed IPI wait' messages.
3978          */
3979         if (apic->check_apicid_used(apic_id_map, apic_id)) {
3980
3981                 for (i = 0; i < get_physical_broadcast(); i++) {
3982                         if (!apic->check_apicid_used(apic_id_map, i))
3983                                 break;
3984                 }
3985
3986                 if (i == get_physical_broadcast())
3987                         panic("Max apic_id exceeded!\n");
3988
3989                 printk(KERN_WARNING "IOAPIC[%d]: apic_id %d already used, "
3990                         "trying %d\n", ioapic, apic_id, i);
3991
3992                 apic_id = i;
3993         }
3994
3995         tmp = apic->apicid_to_cpu_present(apic_id);
3996         physids_or(apic_id_map, apic_id_map, tmp);
3997
3998         if (reg_00.bits.ID != apic_id) {
3999                 reg_00.bits.ID = apic_id;
4000
4001                 spin_lock_irqsave(&ioapic_lock, flags);
4002                 io_apic_write(ioapic, 0, reg_00.raw);
4003                 reg_00.raw = io_apic_read(ioapic, 0);
4004                 spin_unlock_irqrestore(&ioapic_lock, flags);
4005
4006                 /* Sanity check */
4007                 if (reg_00.bits.ID != apic_id) {
4008                         printk("IOAPIC[%d]: Unable to change apic_id!\n", ioapic);
4009                         return -1;
4010                 }
4011         }
4012
4013         apic_printk(APIC_VERBOSE, KERN_INFO
4014                         "IOAPIC[%d]: Assigned apic_id %d\n", ioapic, apic_id);
4015
4016         return apic_id;
4017 }
4018 #endif
4019
4020 int __init io_apic_get_version(int ioapic)
4021 {
4022         union IO_APIC_reg_01    reg_01;
4023         unsigned long flags;
4024
4025         spin_lock_irqsave(&ioapic_lock, flags);
4026         reg_01.raw = io_apic_read(ioapic, 1);
4027         spin_unlock_irqrestore(&ioapic_lock, flags);
4028
4029         return reg_01.bits.version;
4030 }
4031
4032 int acpi_get_override_irq(int bus_irq, int *trigger, int *polarity)
4033 {
4034         int i;
4035
4036         if (skip_ioapic_setup)
4037                 return -1;
4038
4039         for (i = 0; i < mp_irq_entries; i++)
4040                 if (mp_irqs[i].irqtype == mp_INT &&
4041                     mp_irqs[i].srcbusirq == bus_irq)
4042                         break;
4043         if (i >= mp_irq_entries)
4044                 return -1;
4045
4046         *trigger = irq_trigger(i);
4047         *polarity = irq_polarity(i);
4048         return 0;
4049 }
4050
4051 #endif /* CONFIG_ACPI */
4052
4053 /*
4054  * This function currently is only a helper for the i386 smp boot process where
4055  * we need to reprogram the ioredtbls to cater for the cpus which have come online
4056  * so mask in all cases should simply be apic->target_cpus()
4057  */
4058 #ifdef CONFIG_SMP
4059 void __init setup_ioapic_dest(void)
4060 {
4061         int pin, ioapic = 0, irq, irq_entry;
4062         struct irq_desc *desc;
4063         const struct cpumask *mask;
4064
4065         if (skip_ioapic_setup == 1)
4066                 return;
4067
4068 #ifdef CONFIG_ACPI
4069         if (!acpi_disabled && acpi_ioapic) {
4070                 ioapic = mp_find_ioapic(0);
4071                 if (ioapic < 0)
4072                         ioapic = 0;
4073         }
4074 #endif
4075
4076         for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
4077                 irq_entry = find_irq_entry(ioapic, pin, mp_INT);
4078                 if (irq_entry == -1)
4079                         continue;
4080                 irq = pin_2_irq(irq_entry, ioapic, pin);
4081
4082                 desc = irq_to_desc(irq);
4083
4084                 /*
4085                  * Honour affinities which have been set in early boot
4086                  */
4087                 if (desc->status &
4088                     (IRQ_NO_BALANCING | IRQ_AFFINITY_SET))
4089                         mask = desc->affinity;
4090                 else
4091                         mask = apic->target_cpus();
4092
4093                 if (intr_remapping_enabled)
4094                         set_ir_ioapic_affinity_irq_desc(desc, mask);
4095                 else
4096                         set_ioapic_affinity_irq_desc(desc, mask);
4097         }
4098
4099 }
4100 #endif
4101
4102 #define IOAPIC_RESOURCE_NAME_SIZE 11
4103
4104 static struct resource *ioapic_resources;
4105
4106 static struct resource * __init ioapic_setup_resources(void)
4107 {
4108         unsigned long n;
4109         struct resource *res;
4110         char *mem;
4111         int i;
4112
4113         if (nr_ioapics <= 0)
4114                 return NULL;
4115
4116         n = IOAPIC_RESOURCE_NAME_SIZE + sizeof(struct resource);
4117         n *= nr_ioapics;
4118
4119         mem = alloc_bootmem(n);
4120         res = (void *)mem;
4121
4122         if (mem != NULL) {
4123                 mem += sizeof(struct resource) * nr_ioapics;
4124
4125                 for (i = 0; i < nr_ioapics; i++) {
4126                         res[i].name = mem;
4127                         res[i].flags = IORESOURCE_MEM | IORESOURCE_BUSY;
4128                         sprintf(mem,  "IOAPIC %u", i);
4129                         mem += IOAPIC_RESOURCE_NAME_SIZE;
4130                 }
4131         }
4132
4133         ioapic_resources = res;
4134
4135         return res;
4136 }
4137
4138 void __init ioapic_init_mappings(void)
4139 {
4140         unsigned long ioapic_phys, idx = FIX_IO_APIC_BASE_0;
4141         struct resource *ioapic_res;
4142         int i;
4143
4144         ioapic_res = ioapic_setup_resources();
4145         for (i = 0; i < nr_ioapics; i++) {
4146                 if (smp_found_config) {
4147                         ioapic_phys = mp_ioapics[i].apicaddr;
4148 #ifdef CONFIG_X86_32
4149                         if (!ioapic_phys) {
4150                                 printk(KERN_ERR
4151                                        "WARNING: bogus zero IO-APIC "
4152                                        "address found in MPTABLE, "
4153                                        "disabling IO/APIC support!\n");
4154                                 smp_found_config = 0;
4155                                 skip_ioapic_setup = 1;
4156                                 goto fake_ioapic_page;
4157                         }
4158 #endif
4159                 } else {
4160 #ifdef CONFIG_X86_32
4161 fake_ioapic_page:
4162 #endif
4163                         ioapic_phys = (unsigned long)
4164                                 alloc_bootmem_pages(PAGE_SIZE);
4165                         ioapic_phys = __pa(ioapic_phys);
4166                 }
4167                 set_fixmap_nocache(idx, ioapic_phys);
4168                 apic_printk(APIC_VERBOSE,
4169                             "mapped IOAPIC to %08lx (%08lx)\n",
4170                             __fix_to_virt(idx), ioapic_phys);
4171                 idx++;
4172
4173                 if (ioapic_res != NULL) {
4174                         ioapic_res->start = ioapic_phys;
4175                         ioapic_res->end = ioapic_phys + (4 * 1024) - 1;
4176                         ioapic_res++;
4177                 }
4178         }
4179 }
4180
4181 static int __init ioapic_insert_resources(void)
4182 {
4183         int i;
4184         struct resource *r = ioapic_resources;
4185
4186         if (!r) {
4187                 if (nr_ioapics > 0) {
4188                         printk(KERN_ERR
4189                                 "IO APIC resources couldn't be allocated.\n");
4190                         return -1;
4191                 }
4192                 return 0;
4193         }
4194
4195         for (i = 0; i < nr_ioapics; i++) {
4196                 insert_resource(&iomem_resource, r);
4197                 r++;
4198         }
4199
4200         return 0;
4201 }
4202
4203 /* Insert the IO APIC resources after PCI initialization has occured to handle
4204  * IO APICS that are mapped in on a BAR in PCI space. */
4205 late_initcall(ioapic_insert_resources);