Merge branch 'tracing-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[pandora-kernel.git] / 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 = {{0, 0}};
466
467         eu.entry = e;
468         io_apic_write(apic, 0x11 + 2*pin, eu.w2);
469         io_apic_write(apic, 0x10 + 2*pin, eu.w1);
470 }
471
472 void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
473 {
474         unsigned long flags;
475         spin_lock_irqsave(&ioapic_lock, flags);
476         __ioapic_write_entry(apic, pin, e);
477         spin_unlock_irqrestore(&ioapic_lock, flags);
478 }
479
480 /*
481  * When we mask an IO APIC routing entry, we need to write the low
482  * word first, in order to set the mask bit before we change the
483  * high bits!
484  */
485 static void ioapic_mask_entry(int apic, int pin)
486 {
487         unsigned long flags;
488         union entry_union eu = { .entry.mask = 1 };
489
490         spin_lock_irqsave(&ioapic_lock, flags);
491         io_apic_write(apic, 0x10 + 2*pin, eu.w1);
492         io_apic_write(apic, 0x11 + 2*pin, eu.w2);
493         spin_unlock_irqrestore(&ioapic_lock, flags);
494 }
495
496 /*
497  * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
498  * shared ISA-space IRQs, so we have to support them. We are super
499  * fast in the common case, and fast for shared ISA-space IRQs.
500  */
501 static void add_pin_to_irq_node(struct irq_cfg *cfg, int node, int apic, int pin)
502 {
503         struct irq_pin_list *entry;
504
505         entry = cfg->irq_2_pin;
506         if (!entry) {
507                 entry = get_one_free_irq_2_pin(node);
508                 if (!entry) {
509                         printk(KERN_ERR "can not alloc irq_2_pin to add %d - %d\n",
510                                         apic, pin);
511                         return;
512                 }
513                 cfg->irq_2_pin = entry;
514                 entry->apic = apic;
515                 entry->pin = pin;
516                 return;
517         }
518
519         while (entry->next) {
520                 /* not again, please */
521                 if (entry->apic == apic && entry->pin == pin)
522                         return;
523
524                 entry = entry->next;
525         }
526
527         entry->next = get_one_free_irq_2_pin(node);
528         entry = entry->next;
529         entry->apic = apic;
530         entry->pin = pin;
531 }
532
533 /*
534  * Reroute an IRQ to a different pin.
535  */
536 static void __init replace_pin_at_irq_node(struct irq_cfg *cfg, int node,
537                                       int oldapic, int oldpin,
538                                       int newapic, int newpin)
539 {
540         struct irq_pin_list *entry = cfg->irq_2_pin;
541         int replaced = 0;
542
543         while (entry) {
544                 if (entry->apic == oldapic && entry->pin == oldpin) {
545                         entry->apic = newapic;
546                         entry->pin = newpin;
547                         replaced = 1;
548                         /* every one is different, right? */
549                         break;
550                 }
551                 entry = entry->next;
552         }
553
554         /* why? call replace before add? */
555         if (!replaced)
556                 add_pin_to_irq_node(cfg, node, newapic, newpin);
557 }
558
559 static inline void io_apic_modify_irq(struct irq_cfg *cfg,
560                                 int mask_and, int mask_or,
561                                 void (*final)(struct irq_pin_list *entry))
562 {
563         int pin;
564         struct irq_pin_list *entry;
565
566         for (entry = cfg->irq_2_pin; entry != NULL; entry = entry->next) {
567                 unsigned int reg;
568                 pin = entry->pin;
569                 reg = io_apic_read(entry->apic, 0x10 + pin * 2);
570                 reg &= mask_and;
571                 reg |= mask_or;
572                 io_apic_modify(entry->apic, 0x10 + pin * 2, reg);
573                 if (final)
574                         final(entry);
575         }
576 }
577
578 static void __unmask_IO_APIC_irq(struct irq_cfg *cfg)
579 {
580         io_apic_modify_irq(cfg, ~IO_APIC_REDIR_MASKED, 0, NULL);
581 }
582
583 #ifdef CONFIG_X86_64
584 static void io_apic_sync(struct irq_pin_list *entry)
585 {
586         /*
587          * Synchronize the IO-APIC and the CPU by doing
588          * a dummy read from the IO-APIC
589          */
590         struct io_apic __iomem *io_apic;
591         io_apic = io_apic_base(entry->apic);
592         readl(&io_apic->data);
593 }
594
595 static void __mask_IO_APIC_irq(struct irq_cfg *cfg)
596 {
597         io_apic_modify_irq(cfg, ~0, IO_APIC_REDIR_MASKED, &io_apic_sync);
598 }
599 #else /* CONFIG_X86_32 */
600 static void __mask_IO_APIC_irq(struct irq_cfg *cfg)
601 {
602         io_apic_modify_irq(cfg, ~0, IO_APIC_REDIR_MASKED, NULL);
603 }
604
605 static void __mask_and_edge_IO_APIC_irq(struct irq_cfg *cfg)
606 {
607         io_apic_modify_irq(cfg, ~IO_APIC_REDIR_LEVEL_TRIGGER,
608                         IO_APIC_REDIR_MASKED, NULL);
609 }
610
611 static void __unmask_and_level_IO_APIC_irq(struct irq_cfg *cfg)
612 {
613         io_apic_modify_irq(cfg, ~IO_APIC_REDIR_MASKED,
614                         IO_APIC_REDIR_LEVEL_TRIGGER, NULL);
615 }
616 #endif /* CONFIG_X86_32 */
617
618 static void mask_IO_APIC_irq_desc(struct irq_desc *desc)
619 {
620         struct irq_cfg *cfg = desc->chip_data;
621         unsigned long flags;
622
623         BUG_ON(!cfg);
624
625         spin_lock_irqsave(&ioapic_lock, flags);
626         __mask_IO_APIC_irq(cfg);
627         spin_unlock_irqrestore(&ioapic_lock, flags);
628 }
629
630 static void unmask_IO_APIC_irq_desc(struct irq_desc *desc)
631 {
632         struct irq_cfg *cfg = desc->chip_data;
633         unsigned long flags;
634
635         spin_lock_irqsave(&ioapic_lock, flags);
636         __unmask_IO_APIC_irq(cfg);
637         spin_unlock_irqrestore(&ioapic_lock, flags);
638 }
639
640 static void mask_IO_APIC_irq(unsigned int irq)
641 {
642         struct irq_desc *desc = irq_to_desc(irq);
643
644         mask_IO_APIC_irq_desc(desc);
645 }
646 static void unmask_IO_APIC_irq(unsigned int irq)
647 {
648         struct irq_desc *desc = irq_to_desc(irq);
649
650         unmask_IO_APIC_irq_desc(desc);
651 }
652
653 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
654 {
655         struct IO_APIC_route_entry entry;
656
657         /* Check delivery_mode to be sure we're not clearing an SMI pin */
658         entry = ioapic_read_entry(apic, pin);
659         if (entry.delivery_mode == dest_SMI)
660                 return;
661         /*
662          * Disable it in the IO-APIC irq-routing table:
663          */
664         ioapic_mask_entry(apic, pin);
665 }
666
667 static void clear_IO_APIC (void)
668 {
669         int apic, pin;
670
671         for (apic = 0; apic < nr_ioapics; apic++)
672                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
673                         clear_IO_APIC_pin(apic, pin);
674 }
675
676 #ifdef CONFIG_X86_32
677 /*
678  * support for broken MP BIOSs, enables hand-redirection of PIRQ0-7 to
679  * specific CPU-side IRQs.
680  */
681
682 #define MAX_PIRQS 8
683 static int pirq_entries[MAX_PIRQS] = {
684         [0 ... MAX_PIRQS - 1] = -1
685 };
686
687 static int __init ioapic_pirq_setup(char *str)
688 {
689         int i, max;
690         int ints[MAX_PIRQS+1];
691
692         get_options(str, ARRAY_SIZE(ints), ints);
693
694         apic_printk(APIC_VERBOSE, KERN_INFO
695                         "PIRQ redirection, working around broken MP-BIOS.\n");
696         max = MAX_PIRQS;
697         if (ints[0] < MAX_PIRQS)
698                 max = ints[0];
699
700         for (i = 0; i < max; i++) {
701                 apic_printk(APIC_VERBOSE, KERN_DEBUG
702                                 "... PIRQ%d -> IRQ %d\n", i, ints[i+1]);
703                 /*
704                  * PIRQs are mapped upside down, usually.
705                  */
706                 pirq_entries[MAX_PIRQS-i-1] = ints[i+1];
707         }
708         return 1;
709 }
710
711 __setup("pirq=", ioapic_pirq_setup);
712 #endif /* CONFIG_X86_32 */
713
714 struct IO_APIC_route_entry **alloc_ioapic_entries(void)
715 {
716         int apic;
717         struct IO_APIC_route_entry **ioapic_entries;
718
719         ioapic_entries = kzalloc(sizeof(*ioapic_entries) * nr_ioapics,
720                                 GFP_ATOMIC);
721         if (!ioapic_entries)
722                 return 0;
723
724         for (apic = 0; apic < nr_ioapics; apic++) {
725                 ioapic_entries[apic] =
726                         kzalloc(sizeof(struct IO_APIC_route_entry) *
727                                 nr_ioapic_registers[apic], GFP_ATOMIC);
728                 if (!ioapic_entries[apic])
729                         goto nomem;
730         }
731
732         return ioapic_entries;
733
734 nomem:
735         while (--apic >= 0)
736                 kfree(ioapic_entries[apic]);
737         kfree(ioapic_entries);
738
739         return 0;
740 }
741
742 /*
743  * Saves all the IO-APIC RTE's
744  */
745 int save_IO_APIC_setup(struct IO_APIC_route_entry **ioapic_entries)
746 {
747         int apic, pin;
748
749         if (!ioapic_entries)
750                 return -ENOMEM;
751
752         for (apic = 0; apic < nr_ioapics; apic++) {
753                 if (!ioapic_entries[apic])
754                         return -ENOMEM;
755
756                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
757                         ioapic_entries[apic][pin] =
758                                 ioapic_read_entry(apic, pin);
759         }
760
761         return 0;
762 }
763
764 /*
765  * Mask all IO APIC entries.
766  */
767 void mask_IO_APIC_setup(struct IO_APIC_route_entry **ioapic_entries)
768 {
769         int apic, pin;
770
771         if (!ioapic_entries)
772                 return;
773
774         for (apic = 0; apic < nr_ioapics; apic++) {
775                 if (!ioapic_entries[apic])
776                         break;
777
778                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
779                         struct IO_APIC_route_entry entry;
780
781                         entry = ioapic_entries[apic][pin];
782                         if (!entry.mask) {
783                                 entry.mask = 1;
784                                 ioapic_write_entry(apic, pin, entry);
785                         }
786                 }
787         }
788 }
789
790 /*
791  * Restore IO APIC entries which was saved in ioapic_entries.
792  */
793 int restore_IO_APIC_setup(struct IO_APIC_route_entry **ioapic_entries)
794 {
795         int apic, pin;
796
797         if (!ioapic_entries)
798                 return -ENOMEM;
799
800         for (apic = 0; apic < nr_ioapics; apic++) {
801                 if (!ioapic_entries[apic])
802                         return -ENOMEM;
803
804                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
805                         ioapic_write_entry(apic, pin,
806                                         ioapic_entries[apic][pin]);
807         }
808         return 0;
809 }
810
811 void free_ioapic_entries(struct IO_APIC_route_entry **ioapic_entries)
812 {
813         int apic;
814
815         for (apic = 0; apic < nr_ioapics; apic++)
816                 kfree(ioapic_entries[apic]);
817
818         kfree(ioapic_entries);
819 }
820
821 /*
822  * Find the IRQ entry number of a certain pin.
823  */
824 static int find_irq_entry(int apic, int pin, int type)
825 {
826         int i;
827
828         for (i = 0; i < mp_irq_entries; i++)
829                 if (mp_irqs[i].irqtype == type &&
830                     (mp_irqs[i].dstapic == mp_ioapics[apic].apicid ||
831                      mp_irqs[i].dstapic == MP_APIC_ALL) &&
832                     mp_irqs[i].dstirq == pin)
833                         return i;
834
835         return -1;
836 }
837
838 /*
839  * Find the pin to which IRQ[irq] (ISA) is connected
840  */
841 static int __init find_isa_irq_pin(int irq, int type)
842 {
843         int i;
844
845         for (i = 0; i < mp_irq_entries; i++) {
846                 int lbus = mp_irqs[i].srcbus;
847
848                 if (test_bit(lbus, mp_bus_not_pci) &&
849                     (mp_irqs[i].irqtype == type) &&
850                     (mp_irqs[i].srcbusirq == irq))
851
852                         return mp_irqs[i].dstirq;
853         }
854         return -1;
855 }
856
857 static int __init find_isa_irq_apic(int irq, int type)
858 {
859         int i;
860
861         for (i = 0; i < mp_irq_entries; i++) {
862                 int lbus = mp_irqs[i].srcbus;
863
864                 if (test_bit(lbus, mp_bus_not_pci) &&
865                     (mp_irqs[i].irqtype == type) &&
866                     (mp_irqs[i].srcbusirq == irq))
867                         break;
868         }
869         if (i < mp_irq_entries) {
870                 int apic;
871                 for(apic = 0; apic < nr_ioapics; apic++) {
872                         if (mp_ioapics[apic].apicid == mp_irqs[i].dstapic)
873                                 return apic;
874                 }
875         }
876
877         return -1;
878 }
879
880 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
881 /*
882  * EISA Edge/Level control register, ELCR
883  */
884 static int EISA_ELCR(unsigned int irq)
885 {
886         if (irq < NR_IRQS_LEGACY) {
887                 unsigned int port = 0x4d0 + (irq >> 3);
888                 return (inb(port) >> (irq & 7)) & 1;
889         }
890         apic_printk(APIC_VERBOSE, KERN_INFO
891                         "Broken MPtable reports ISA irq %d\n", irq);
892         return 0;
893 }
894
895 #endif
896
897 /* ISA interrupts are always polarity zero edge triggered,
898  * when listed as conforming in the MP table. */
899
900 #define default_ISA_trigger(idx)        (0)
901 #define default_ISA_polarity(idx)       (0)
902
903 /* EISA interrupts are always polarity zero and can be edge or level
904  * trigger depending on the ELCR value.  If an interrupt is listed as
905  * EISA conforming in the MP table, that means its trigger type must
906  * be read in from the ELCR */
907
908 #define default_EISA_trigger(idx)       (EISA_ELCR(mp_irqs[idx].srcbusirq))
909 #define default_EISA_polarity(idx)      default_ISA_polarity(idx)
910
911 /* PCI interrupts are always polarity one level triggered,
912  * when listed as conforming in the MP table. */
913
914 #define default_PCI_trigger(idx)        (1)
915 #define default_PCI_polarity(idx)       (1)
916
917 /* MCA interrupts are always polarity zero level triggered,
918  * when listed as conforming in the MP table. */
919
920 #define default_MCA_trigger(idx)        (1)
921 #define default_MCA_polarity(idx)       default_ISA_polarity(idx)
922
923 static int MPBIOS_polarity(int idx)
924 {
925         int bus = mp_irqs[idx].srcbus;
926         int polarity;
927
928         /*
929          * Determine IRQ line polarity (high active or low active):
930          */
931         switch (mp_irqs[idx].irqflag & 3)
932         {
933                 case 0: /* conforms, ie. bus-type dependent polarity */
934                         if (test_bit(bus, mp_bus_not_pci))
935                                 polarity = default_ISA_polarity(idx);
936                         else
937                                 polarity = default_PCI_polarity(idx);
938                         break;
939                 case 1: /* high active */
940                 {
941                         polarity = 0;
942                         break;
943                 }
944                 case 2: /* reserved */
945                 {
946                         printk(KERN_WARNING "broken BIOS!!\n");
947                         polarity = 1;
948                         break;
949                 }
950                 case 3: /* low active */
951                 {
952                         polarity = 1;
953                         break;
954                 }
955                 default: /* invalid */
956                 {
957                         printk(KERN_WARNING "broken BIOS!!\n");
958                         polarity = 1;
959                         break;
960                 }
961         }
962         return polarity;
963 }
964
965 static int MPBIOS_trigger(int idx)
966 {
967         int bus = mp_irqs[idx].srcbus;
968         int trigger;
969
970         /*
971          * Determine IRQ trigger mode (edge or level sensitive):
972          */
973         switch ((mp_irqs[idx].irqflag>>2) & 3)
974         {
975                 case 0: /* conforms, ie. bus-type dependent */
976                         if (test_bit(bus, mp_bus_not_pci))
977                                 trigger = default_ISA_trigger(idx);
978                         else
979                                 trigger = default_PCI_trigger(idx);
980 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
981                         switch (mp_bus_id_to_type[bus]) {
982                                 case MP_BUS_ISA: /* ISA pin */
983                                 {
984                                         /* set before the switch */
985                                         break;
986                                 }
987                                 case MP_BUS_EISA: /* EISA pin */
988                                 {
989                                         trigger = default_EISA_trigger(idx);
990                                         break;
991                                 }
992                                 case MP_BUS_PCI: /* PCI pin */
993                                 {
994                                         /* set before the switch */
995                                         break;
996                                 }
997                                 case MP_BUS_MCA: /* MCA pin */
998                                 {
999                                         trigger = default_MCA_trigger(idx);
1000                                         break;
1001                                 }
1002                                 default:
1003                                 {
1004                                         printk(KERN_WARNING "broken BIOS!!\n");
1005                                         trigger = 1;
1006                                         break;
1007                                 }
1008                         }
1009 #endif
1010                         break;
1011                 case 1: /* edge */
1012                 {
1013                         trigger = 0;
1014                         break;
1015                 }
1016                 case 2: /* reserved */
1017                 {
1018                         printk(KERN_WARNING "broken BIOS!!\n");
1019                         trigger = 1;
1020                         break;
1021                 }
1022                 case 3: /* level */
1023                 {
1024                         trigger = 1;
1025                         break;
1026                 }
1027                 default: /* invalid */
1028                 {
1029                         printk(KERN_WARNING "broken BIOS!!\n");
1030                         trigger = 0;
1031                         break;
1032                 }
1033         }
1034         return trigger;
1035 }
1036
1037 static inline int irq_polarity(int idx)
1038 {
1039         return MPBIOS_polarity(idx);
1040 }
1041
1042 static inline int irq_trigger(int idx)
1043 {
1044         return MPBIOS_trigger(idx);
1045 }
1046
1047 int (*ioapic_renumber_irq)(int ioapic, int irq);
1048 static int pin_2_irq(int idx, int apic, int pin)
1049 {
1050         int irq, i;
1051         int bus = mp_irqs[idx].srcbus;
1052
1053         /*
1054          * Debugging check, we are in big trouble if this message pops up!
1055          */
1056         if (mp_irqs[idx].dstirq != pin)
1057                 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
1058
1059         if (test_bit(bus, mp_bus_not_pci)) {
1060                 irq = mp_irqs[idx].srcbusirq;
1061         } else {
1062                 /*
1063                  * PCI IRQs are mapped in order
1064                  */
1065                 i = irq = 0;
1066                 while (i < apic)
1067                         irq += nr_ioapic_registers[i++];
1068                 irq += pin;
1069                 /*
1070                  * For MPS mode, so far only needed by ES7000 platform
1071                  */
1072                 if (ioapic_renumber_irq)
1073                         irq = ioapic_renumber_irq(apic, irq);
1074         }
1075
1076 #ifdef CONFIG_X86_32
1077         /*
1078          * PCI IRQ command line redirection. Yes, limits are hardcoded.
1079          */
1080         if ((pin >= 16) && (pin <= 23)) {
1081                 if (pirq_entries[pin-16] != -1) {
1082                         if (!pirq_entries[pin-16]) {
1083                                 apic_printk(APIC_VERBOSE, KERN_DEBUG
1084                                                 "disabling PIRQ%d\n", pin-16);
1085                         } else {
1086                                 irq = pirq_entries[pin-16];
1087                                 apic_printk(APIC_VERBOSE, KERN_DEBUG
1088                                                 "using PIRQ%d -> IRQ %d\n",
1089                                                 pin-16, irq);
1090                         }
1091                 }
1092         }
1093 #endif
1094
1095         return irq;
1096 }
1097
1098 /*
1099  * Find a specific PCI IRQ entry.
1100  * Not an __init, possibly needed by modules
1101  */
1102 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin,
1103                                 struct io_apic_irq_attr *irq_attr)
1104 {
1105         int apic, i, best_guess = -1;
1106
1107         apic_printk(APIC_DEBUG,
1108                     "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
1109                     bus, slot, pin);
1110         if (test_bit(bus, mp_bus_not_pci)) {
1111                 apic_printk(APIC_VERBOSE,
1112                             "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
1113                 return -1;
1114         }
1115         for (i = 0; i < mp_irq_entries; i++) {
1116                 int lbus = mp_irqs[i].srcbus;
1117
1118                 for (apic = 0; apic < nr_ioapics; apic++)
1119                         if (mp_ioapics[apic].apicid == mp_irqs[i].dstapic ||
1120                             mp_irqs[i].dstapic == MP_APIC_ALL)
1121                                 break;
1122
1123                 if (!test_bit(lbus, mp_bus_not_pci) &&
1124                     !mp_irqs[i].irqtype &&
1125                     (bus == lbus) &&
1126                     (slot == ((mp_irqs[i].srcbusirq >> 2) & 0x1f))) {
1127                         int irq = pin_2_irq(i, apic, mp_irqs[i].dstirq);
1128
1129                         if (!(apic || IO_APIC_IRQ(irq)))
1130                                 continue;
1131
1132                         if (pin == (mp_irqs[i].srcbusirq & 3)) {
1133                                 set_io_apic_irq_attr(irq_attr, apic,
1134                                                      mp_irqs[i].dstirq,
1135                                                      irq_trigger(i),
1136                                                      irq_polarity(i));
1137                                 return irq;
1138                         }
1139                         /*
1140                          * Use the first all-but-pin matching entry as a
1141                          * best-guess fuzzy result for broken mptables.
1142                          */
1143                         if (best_guess < 0) {
1144                                 set_io_apic_irq_attr(irq_attr, apic,
1145                                                      mp_irqs[i].dstirq,
1146                                                      irq_trigger(i),
1147                                                      irq_polarity(i));
1148                                 best_guess = irq;
1149                         }
1150                 }
1151         }
1152         return best_guess;
1153 }
1154 EXPORT_SYMBOL(IO_APIC_get_PCI_irq_vector);
1155
1156 void lock_vector_lock(void)
1157 {
1158         /* Used to the online set of cpus does not change
1159          * during assign_irq_vector.
1160          */
1161         spin_lock(&vector_lock);
1162 }
1163
1164 void unlock_vector_lock(void)
1165 {
1166         spin_unlock(&vector_lock);
1167 }
1168
1169 static int
1170 __assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask)
1171 {
1172         /*
1173          * NOTE! The local APIC isn't very good at handling
1174          * multiple interrupts at the same interrupt level.
1175          * As the interrupt level is determined by taking the
1176          * vector number and shifting that right by 4, we
1177          * want to spread these out a bit so that they don't
1178          * all fall in the same interrupt level.
1179          *
1180          * Also, we've got to be careful not to trash gate
1181          * 0x80, because int 0x80 is hm, kind of importantish. ;)
1182          */
1183         static int current_vector = FIRST_DEVICE_VECTOR, current_offset = 0;
1184         unsigned int old_vector;
1185         int cpu, err;
1186         cpumask_var_t tmp_mask;
1187
1188         if ((cfg->move_in_progress) || cfg->move_cleanup_count)
1189                 return -EBUSY;
1190
1191         if (!alloc_cpumask_var(&tmp_mask, GFP_ATOMIC))
1192                 return -ENOMEM;
1193
1194         old_vector = cfg->vector;
1195         if (old_vector) {
1196                 cpumask_and(tmp_mask, mask, cpu_online_mask);
1197                 cpumask_and(tmp_mask, cfg->domain, tmp_mask);
1198                 if (!cpumask_empty(tmp_mask)) {
1199                         free_cpumask_var(tmp_mask);
1200                         return 0;
1201                 }
1202         }
1203
1204         /* Only try and allocate irqs on cpus that are present */
1205         err = -ENOSPC;
1206         for_each_cpu_and(cpu, mask, cpu_online_mask) {
1207                 int new_cpu;
1208                 int vector, offset;
1209
1210                 apic->vector_allocation_domain(cpu, tmp_mask);
1211
1212                 vector = current_vector;
1213                 offset = current_offset;
1214 next:
1215                 vector += 8;
1216                 if (vector >= first_system_vector) {
1217                         /* If out of vectors on large boxen, must share them. */
1218                         offset = (offset + 1) % 8;
1219                         vector = FIRST_DEVICE_VECTOR + offset;
1220                 }
1221                 if (unlikely(current_vector == vector))
1222                         continue;
1223
1224                 if (test_bit(vector, used_vectors))
1225                         goto next;
1226
1227                 for_each_cpu_and(new_cpu, tmp_mask, cpu_online_mask)
1228                         if (per_cpu(vector_irq, new_cpu)[vector] != -1)
1229                                 goto next;
1230                 /* Found one! */
1231                 current_vector = vector;
1232                 current_offset = offset;
1233                 if (old_vector) {
1234                         cfg->move_in_progress = 1;
1235                         cpumask_copy(cfg->old_domain, cfg->domain);
1236                 }
1237                 for_each_cpu_and(new_cpu, tmp_mask, cpu_online_mask)
1238                         per_cpu(vector_irq, new_cpu)[vector] = irq;
1239                 cfg->vector = vector;
1240                 cpumask_copy(cfg->domain, tmp_mask);
1241                 err = 0;
1242                 break;
1243         }
1244         free_cpumask_var(tmp_mask);
1245         return err;
1246 }
1247
1248 static int
1249 assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask)
1250 {
1251         int err;
1252         unsigned long flags;
1253
1254         spin_lock_irqsave(&vector_lock, flags);
1255         err = __assign_irq_vector(irq, cfg, mask);
1256         spin_unlock_irqrestore(&vector_lock, flags);
1257         return err;
1258 }
1259
1260 static void __clear_irq_vector(int irq, struct irq_cfg *cfg)
1261 {
1262         int cpu, vector;
1263
1264         BUG_ON(!cfg->vector);
1265
1266         vector = cfg->vector;
1267         for_each_cpu_and(cpu, cfg->domain, cpu_online_mask)
1268                 per_cpu(vector_irq, cpu)[vector] = -1;
1269
1270         cfg->vector = 0;
1271         cpumask_clear(cfg->domain);
1272
1273         if (likely(!cfg->move_in_progress))
1274                 return;
1275         for_each_cpu_and(cpu, cfg->old_domain, cpu_online_mask) {
1276                 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS;
1277                                                                 vector++) {
1278                         if (per_cpu(vector_irq, cpu)[vector] != irq)
1279                                 continue;
1280                         per_cpu(vector_irq, cpu)[vector] = -1;
1281                         break;
1282                 }
1283         }
1284         cfg->move_in_progress = 0;
1285 }
1286
1287 void __setup_vector_irq(int cpu)
1288 {
1289         /* Initialize vector_irq on a new cpu */
1290         /* This function must be called with vector_lock held */
1291         int irq, vector;
1292         struct irq_cfg *cfg;
1293         struct irq_desc *desc;
1294
1295         /* Mark the inuse vectors */
1296         for_each_irq_desc(irq, desc) {
1297                 cfg = desc->chip_data;
1298                 if (!cpumask_test_cpu(cpu, cfg->domain))
1299                         continue;
1300                 vector = cfg->vector;
1301                 per_cpu(vector_irq, cpu)[vector] = irq;
1302         }
1303         /* Mark the free vectors */
1304         for (vector = 0; vector < NR_VECTORS; ++vector) {
1305                 irq = per_cpu(vector_irq, cpu)[vector];
1306                 if (irq < 0)
1307                         continue;
1308
1309                 cfg = irq_cfg(irq);
1310                 if (!cpumask_test_cpu(cpu, cfg->domain))
1311                         per_cpu(vector_irq, cpu)[vector] = -1;
1312         }
1313 }
1314
1315 static struct irq_chip ioapic_chip;
1316 static struct irq_chip ir_ioapic_chip;
1317
1318 #define IOAPIC_AUTO     -1
1319 #define IOAPIC_EDGE     0
1320 #define IOAPIC_LEVEL    1
1321
1322 #ifdef CONFIG_X86_32
1323 static inline int IO_APIC_irq_trigger(int irq)
1324 {
1325         int apic, idx, pin;
1326
1327         for (apic = 0; apic < nr_ioapics; apic++) {
1328                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1329                         idx = find_irq_entry(apic, pin, mp_INT);
1330                         if ((idx != -1) && (irq == pin_2_irq(idx, apic, pin)))
1331                                 return irq_trigger(idx);
1332                 }
1333         }
1334         /*
1335          * nonexistent IRQs are edge default
1336          */
1337         return 0;
1338 }
1339 #else
1340 static inline int IO_APIC_irq_trigger(int irq)
1341 {
1342         return 1;
1343 }
1344 #endif
1345
1346 static void ioapic_register_intr(int irq, struct irq_desc *desc, unsigned long trigger)
1347 {
1348
1349         if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1350             trigger == IOAPIC_LEVEL)
1351                 desc->status |= IRQ_LEVEL;
1352         else
1353                 desc->status &= ~IRQ_LEVEL;
1354
1355         if (irq_remapped(irq)) {
1356                 desc->status |= IRQ_MOVE_PCNTXT;
1357                 if (trigger)
1358                         set_irq_chip_and_handler_name(irq, &ir_ioapic_chip,
1359                                                       handle_fasteoi_irq,
1360                                                      "fasteoi");
1361                 else
1362                         set_irq_chip_and_handler_name(irq, &ir_ioapic_chip,
1363                                                       handle_edge_irq, "edge");
1364                 return;
1365         }
1366
1367         if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1368             trigger == IOAPIC_LEVEL)
1369                 set_irq_chip_and_handler_name(irq, &ioapic_chip,
1370                                               handle_fasteoi_irq,
1371                                               "fasteoi");
1372         else
1373                 set_irq_chip_and_handler_name(irq, &ioapic_chip,
1374                                               handle_edge_irq, "edge");
1375 }
1376
1377 int setup_ioapic_entry(int apic_id, int irq,
1378                        struct IO_APIC_route_entry *entry,
1379                        unsigned int destination, int trigger,
1380                        int polarity, int vector, int pin)
1381 {
1382         /*
1383          * add it to the IO-APIC irq-routing table:
1384          */
1385         memset(entry,0,sizeof(*entry));
1386
1387         if (intr_remapping_enabled) {
1388                 struct intel_iommu *iommu = map_ioapic_to_ir(apic_id);
1389                 struct irte irte;
1390                 struct IR_IO_APIC_route_entry *ir_entry =
1391                         (struct IR_IO_APIC_route_entry *) entry;
1392                 int index;
1393
1394                 if (!iommu)
1395                         panic("No mapping iommu for ioapic %d\n", apic_id);
1396
1397                 index = alloc_irte(iommu, irq, 1);
1398                 if (index < 0)
1399                         panic("Failed to allocate IRTE for ioapic %d\n", apic_id);
1400
1401                 memset(&irte, 0, sizeof(irte));
1402
1403                 irte.present = 1;
1404                 irte.dst_mode = apic->irq_dest_mode;
1405                 /*
1406                  * Trigger mode in the IRTE will always be edge, and the
1407                  * actual level or edge trigger will be setup in the IO-APIC
1408                  * RTE. This will help simplify level triggered irq migration.
1409                  * For more details, see the comments above explainig IO-APIC
1410                  * irq migration in the presence of interrupt-remapping.
1411                  */
1412                 irte.trigger_mode = 0;
1413                 irte.dlvry_mode = apic->irq_delivery_mode;
1414                 irte.vector = vector;
1415                 irte.dest_id = IRTE_DEST(destination);
1416
1417                 modify_irte(irq, &irte);
1418
1419                 ir_entry->index2 = (index >> 15) & 0x1;
1420                 ir_entry->zero = 0;
1421                 ir_entry->format = 1;
1422                 ir_entry->index = (index & 0x7fff);
1423                 /*
1424                  * IO-APIC RTE will be configured with virtual vector.
1425                  * irq handler will do the explicit EOI to the io-apic.
1426                  */
1427                 ir_entry->vector = pin;
1428         } else {
1429                 entry->delivery_mode = apic->irq_delivery_mode;
1430                 entry->dest_mode = apic->irq_dest_mode;
1431                 entry->dest = destination;
1432                 entry->vector = vector;
1433         }
1434
1435         entry->mask = 0;                                /* enable IRQ */
1436         entry->trigger = trigger;
1437         entry->polarity = polarity;
1438
1439         /* Mask level triggered irqs.
1440          * Use IRQ_DELAYED_DISABLE for edge triggered irqs.
1441          */
1442         if (trigger)
1443                 entry->mask = 1;
1444         return 0;
1445 }
1446
1447 static void setup_IO_APIC_irq(int apic_id, int pin, unsigned int irq, struct irq_desc *desc,
1448                               int trigger, int polarity)
1449 {
1450         struct irq_cfg *cfg;
1451         struct IO_APIC_route_entry entry;
1452         unsigned int dest;
1453
1454         if (!IO_APIC_IRQ(irq))
1455                 return;
1456
1457         cfg = desc->chip_data;
1458
1459         if (assign_irq_vector(irq, cfg, apic->target_cpus()))
1460                 return;
1461
1462         dest = apic->cpu_mask_to_apicid_and(cfg->domain, apic->target_cpus());
1463
1464         apic_printk(APIC_VERBOSE,KERN_DEBUG
1465                     "IOAPIC[%d]: Set routing entry (%d-%d -> 0x%x -> "
1466                     "IRQ %d Mode:%i Active:%i)\n",
1467                     apic_id, mp_ioapics[apic_id].apicid, pin, cfg->vector,
1468                     irq, trigger, polarity);
1469
1470
1471         if (setup_ioapic_entry(mp_ioapics[apic_id].apicid, irq, &entry,
1472                                dest, trigger, polarity, cfg->vector, pin)) {
1473                 printk("Failed to setup ioapic entry for ioapic  %d, pin %d\n",
1474                        mp_ioapics[apic_id].apicid, pin);
1475                 __clear_irq_vector(irq, cfg);
1476                 return;
1477         }
1478
1479         ioapic_register_intr(irq, desc, trigger);
1480         if (irq < NR_IRQS_LEGACY)
1481                 disable_8259A_irq(irq);
1482
1483         ioapic_write_entry(apic_id, pin, entry);
1484 }
1485
1486 static struct {
1487         DECLARE_BITMAP(pin_programmed, MP_MAX_IOAPIC_PIN + 1);
1488 } mp_ioapic_routing[MAX_IO_APICS];
1489
1490 static void __init setup_IO_APIC_irqs(void)
1491 {
1492         int apic_id = 0, pin, idx, irq;
1493         int notcon = 0;
1494         struct irq_desc *desc;
1495         struct irq_cfg *cfg;
1496         int node = cpu_to_node(boot_cpu_id);
1497
1498         apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
1499
1500 #ifdef CONFIG_ACPI
1501         if (!acpi_disabled && acpi_ioapic) {
1502                 apic_id = mp_find_ioapic(0);
1503                 if (apic_id < 0)
1504                         apic_id = 0;
1505         }
1506 #endif
1507
1508         for (pin = 0; pin < nr_ioapic_registers[apic_id]; pin++) {
1509                 idx = find_irq_entry(apic_id, pin, mp_INT);
1510                 if (idx == -1) {
1511                         if (!notcon) {
1512                                 notcon = 1;
1513                                 apic_printk(APIC_VERBOSE,
1514                                         KERN_DEBUG " %d-%d",
1515                                         mp_ioapics[apic_id].apicid, pin);
1516                         } else
1517                                 apic_printk(APIC_VERBOSE, " %d-%d",
1518                                         mp_ioapics[apic_id].apicid, pin);
1519                         continue;
1520                 }
1521                 if (notcon) {
1522                         apic_printk(APIC_VERBOSE,
1523                                 " (apicid-pin) not connected\n");
1524                         notcon = 0;
1525                 }
1526
1527                 irq = pin_2_irq(idx, apic_id, pin);
1528
1529                 /*
1530                  * Skip the timer IRQ if there's a quirk handler
1531                  * installed and if it returns 1:
1532                  */
1533                 if (apic->multi_timer_check &&
1534                                 apic->multi_timer_check(apic_id, irq))
1535                         continue;
1536
1537                 desc = irq_to_desc_alloc_node(irq, node);
1538                 if (!desc) {
1539                         printk(KERN_INFO "can not get irq_desc for %d\n", irq);
1540                         continue;
1541                 }
1542                 cfg = desc->chip_data;
1543                 add_pin_to_irq_node(cfg, node, apic_id, pin);
1544                 /*
1545                  * don't mark it in pin_programmed, so later acpi could
1546                  * set it correctly when irq < 16
1547                  */
1548                 setup_IO_APIC_irq(apic_id, pin, irq, desc,
1549                                 irq_trigger(idx), irq_polarity(idx));
1550         }
1551
1552         if (notcon)
1553                 apic_printk(APIC_VERBOSE,
1554                         " (apicid-pin) not connected\n");
1555 }
1556
1557 /*
1558  * Set up the timer pin, possibly with the 8259A-master behind.
1559  */
1560 static void __init setup_timer_IRQ0_pin(unsigned int apic_id, unsigned int pin,
1561                                         int vector)
1562 {
1563         struct IO_APIC_route_entry entry;
1564
1565         if (intr_remapping_enabled)
1566                 return;
1567
1568         memset(&entry, 0, sizeof(entry));
1569
1570         /*
1571          * We use logical delivery to get the timer IRQ
1572          * to the first CPU.
1573          */
1574         entry.dest_mode = apic->irq_dest_mode;
1575         entry.mask = 0;                 /* don't mask IRQ for edge */
1576         entry.dest = apic->cpu_mask_to_apicid(apic->target_cpus());
1577         entry.delivery_mode = apic->irq_delivery_mode;
1578         entry.polarity = 0;
1579         entry.trigger = 0;
1580         entry.vector = vector;
1581
1582         /*
1583          * The timer IRQ doesn't have to know that behind the
1584          * scene we may have a 8259A-master in AEOI mode ...
1585          */
1586         set_irq_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq, "edge");
1587
1588         /*
1589          * Add it to the IO-APIC irq-routing table:
1590          */
1591         ioapic_write_entry(apic_id, pin, entry);
1592 }
1593
1594
1595 __apicdebuginit(void) print_IO_APIC(void)
1596 {
1597         int apic, i;
1598         union IO_APIC_reg_00 reg_00;
1599         union IO_APIC_reg_01 reg_01;
1600         union IO_APIC_reg_02 reg_02;
1601         union IO_APIC_reg_03 reg_03;
1602         unsigned long flags;
1603         struct irq_cfg *cfg;
1604         struct irq_desc *desc;
1605         unsigned int irq;
1606
1607         if (apic_verbosity == APIC_QUIET)
1608                 return;
1609
1610         printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
1611         for (i = 0; i < nr_ioapics; i++)
1612                 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
1613                        mp_ioapics[i].apicid, nr_ioapic_registers[i]);
1614
1615         /*
1616          * We are a bit conservative about what we expect.  We have to
1617          * know about every hardware change ASAP.
1618          */
1619         printk(KERN_INFO "testing the IO APIC.......................\n");
1620
1621         for (apic = 0; apic < nr_ioapics; apic++) {
1622
1623         spin_lock_irqsave(&ioapic_lock, flags);
1624         reg_00.raw = io_apic_read(apic, 0);
1625         reg_01.raw = io_apic_read(apic, 1);
1626         if (reg_01.bits.version >= 0x10)
1627                 reg_02.raw = io_apic_read(apic, 2);
1628         if (reg_01.bits.version >= 0x20)
1629                 reg_03.raw = io_apic_read(apic, 3);
1630         spin_unlock_irqrestore(&ioapic_lock, flags);
1631
1632         printk("\n");
1633         printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].apicid);
1634         printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
1635         printk(KERN_DEBUG ".......    : physical APIC id: %02X\n", reg_00.bits.ID);
1636         printk(KERN_DEBUG ".......    : Delivery Type: %X\n", reg_00.bits.delivery_type);
1637         printk(KERN_DEBUG ".......    : LTS          : %X\n", reg_00.bits.LTS);
1638
1639         printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)&reg_01);
1640         printk(KERN_DEBUG ".......     : max redirection entries: %04X\n", reg_01.bits.entries);
1641
1642         printk(KERN_DEBUG ".......     : PRQ implemented: %X\n", reg_01.bits.PRQ);
1643         printk(KERN_DEBUG ".......     : IO APIC version: %04X\n", reg_01.bits.version);
1644
1645         /*
1646          * Some Intel chipsets with IO APIC VERSION of 0x1? don't have reg_02,
1647          * but the value of reg_02 is read as the previous read register
1648          * value, so ignore it if reg_02 == reg_01.
1649          */
1650         if (reg_01.bits.version >= 0x10 && reg_02.raw != reg_01.raw) {
1651                 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
1652                 printk(KERN_DEBUG ".......     : arbitration: %02X\n", reg_02.bits.arbitration);
1653         }
1654
1655         /*
1656          * Some Intel chipsets with IO APIC VERSION of 0x2? don't have reg_02
1657          * or reg_03, but the value of reg_0[23] is read as the previous read
1658          * register value, so ignore it if reg_03 == reg_0[12].
1659          */
1660         if (reg_01.bits.version >= 0x20 && reg_03.raw != reg_02.raw &&
1661             reg_03.raw != reg_01.raw) {
1662                 printk(KERN_DEBUG ".... register #03: %08X\n", reg_03.raw);
1663                 printk(KERN_DEBUG ".......     : Boot DT    : %X\n", reg_03.bits.boot_DT);
1664         }
1665
1666         printk(KERN_DEBUG ".... IRQ redirection table:\n");
1667
1668         printk(KERN_DEBUG " NR Dst Mask Trig IRR Pol"
1669                           " Stat Dmod Deli Vect:   \n");
1670
1671         for (i = 0; i <= reg_01.bits.entries; i++) {
1672                 struct IO_APIC_route_entry entry;
1673
1674                 entry = ioapic_read_entry(apic, i);
1675
1676                 printk(KERN_DEBUG " %02x %03X ",
1677                         i,
1678                         entry.dest
1679                 );
1680
1681                 printk("%1d    %1d    %1d   %1d   %1d    %1d    %1d    %02X\n",
1682                         entry.mask,
1683                         entry.trigger,
1684                         entry.irr,
1685                         entry.polarity,
1686                         entry.delivery_status,
1687                         entry.dest_mode,
1688                         entry.delivery_mode,
1689                         entry.vector
1690                 );
1691         }
1692         }
1693         printk(KERN_DEBUG "IRQ to pin mappings:\n");
1694         for_each_irq_desc(irq, desc) {
1695                 struct irq_pin_list *entry;
1696
1697                 cfg = desc->chip_data;
1698                 entry = cfg->irq_2_pin;
1699                 if (!entry)
1700                         continue;
1701                 printk(KERN_DEBUG "IRQ%d ", irq);
1702                 for (;;) {
1703                         printk("-> %d:%d", entry->apic, entry->pin);
1704                         if (!entry->next)
1705                                 break;
1706                         entry = entry->next;
1707                 }
1708                 printk("\n");
1709         }
1710
1711         printk(KERN_INFO ".................................... done.\n");
1712
1713         return;
1714 }
1715
1716 __apicdebuginit(void) print_APIC_bitfield(int base)
1717 {
1718         unsigned int v;
1719         int i, j;
1720
1721         if (apic_verbosity == APIC_QUIET)
1722                 return;
1723
1724         printk(KERN_DEBUG "0123456789abcdef0123456789abcdef\n" KERN_DEBUG);
1725         for (i = 0; i < 8; i++) {
1726                 v = apic_read(base + i*0x10);
1727                 for (j = 0; j < 32; j++) {
1728                         if (v & (1<<j))
1729                                 printk("1");
1730                         else
1731                                 printk("0");
1732                 }
1733                 printk("\n");
1734         }
1735 }
1736
1737 __apicdebuginit(void) print_local_APIC(void *dummy)
1738 {
1739         unsigned int i, v, ver, maxlvt;
1740         u64 icr;
1741
1742         if (apic_verbosity == APIC_QUIET)
1743                 return;
1744
1745         printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1746                 smp_processor_id(), hard_smp_processor_id());
1747         v = apic_read(APIC_ID);
1748         printk(KERN_INFO "... APIC ID:      %08x (%01x)\n", v, read_apic_id());
1749         v = apic_read(APIC_LVR);
1750         printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1751         ver = GET_APIC_VERSION(v);
1752         maxlvt = lapic_get_maxlvt();
1753
1754         v = apic_read(APIC_TASKPRI);
1755         printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1756
1757         if (APIC_INTEGRATED(ver)) {                     /* !82489DX */
1758                 if (!APIC_XAPIC(ver)) {
1759                         v = apic_read(APIC_ARBPRI);
1760                         printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
1761                                v & APIC_ARBPRI_MASK);
1762                 }
1763                 v = apic_read(APIC_PROCPRI);
1764                 printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1765         }
1766
1767         /*
1768          * Remote read supported only in the 82489DX and local APIC for
1769          * Pentium processors.
1770          */
1771         if (!APIC_INTEGRATED(ver) || maxlvt == 3) {
1772                 v = apic_read(APIC_RRR);
1773                 printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1774         }
1775
1776         v = apic_read(APIC_LDR);
1777         printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1778         if (!x2apic_enabled()) {
1779                 v = apic_read(APIC_DFR);
1780                 printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1781         }
1782         v = apic_read(APIC_SPIV);
1783         printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1784
1785         printk(KERN_DEBUG "... APIC ISR field:\n");
1786         print_APIC_bitfield(APIC_ISR);
1787         printk(KERN_DEBUG "... APIC TMR field:\n");
1788         print_APIC_bitfield(APIC_TMR);
1789         printk(KERN_DEBUG "... APIC IRR field:\n");
1790         print_APIC_bitfield(APIC_IRR);
1791
1792         if (APIC_INTEGRATED(ver)) {             /* !82489DX */
1793                 if (maxlvt > 3)         /* Due to the Pentium erratum 3AP. */
1794                         apic_write(APIC_ESR, 0);
1795
1796                 v = apic_read(APIC_ESR);
1797                 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1798         }
1799
1800         icr = apic_icr_read();
1801         printk(KERN_DEBUG "... APIC ICR: %08x\n", (u32)icr);
1802         printk(KERN_DEBUG "... APIC ICR2: %08x\n", (u32)(icr >> 32));
1803
1804         v = apic_read(APIC_LVTT);
1805         printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1806
1807         if (maxlvt > 3) {                       /* PC is LVT#4. */
1808                 v = apic_read(APIC_LVTPC);
1809                 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1810         }
1811         v = apic_read(APIC_LVT0);
1812         printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1813         v = apic_read(APIC_LVT1);
1814         printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1815
1816         if (maxlvt > 2) {                       /* ERR is LVT#3. */
1817                 v = apic_read(APIC_LVTERR);
1818                 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1819         }
1820
1821         v = apic_read(APIC_TMICT);
1822         printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1823         v = apic_read(APIC_TMCCT);
1824         printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1825         v = apic_read(APIC_TDCR);
1826         printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1827
1828         if (boot_cpu_has(X86_FEATURE_EXTAPIC)) {
1829                 v = apic_read(APIC_EFEAT);
1830                 maxlvt = (v >> 16) & 0xff;
1831                 printk(KERN_DEBUG "... APIC EFEAT: %08x\n", v);
1832                 v = apic_read(APIC_ECTRL);
1833                 printk(KERN_DEBUG "... APIC ECTRL: %08x\n", v);
1834                 for (i = 0; i < maxlvt; i++) {
1835                         v = apic_read(APIC_EILVTn(i));
1836                         printk(KERN_DEBUG "... APIC EILVT%d: %08x\n", i, v);
1837                 }
1838         }
1839         printk("\n");
1840 }
1841
1842 __apicdebuginit(void) print_all_local_APICs(void)
1843 {
1844         int cpu;
1845
1846         preempt_disable();
1847         for_each_online_cpu(cpu)
1848                 smp_call_function_single(cpu, print_local_APIC, NULL, 1);
1849         preempt_enable();
1850 }
1851
1852 __apicdebuginit(void) print_PIC(void)
1853 {
1854         unsigned int v;
1855         unsigned long flags;
1856
1857         if (apic_verbosity == APIC_QUIET)
1858                 return;
1859
1860         printk(KERN_DEBUG "\nprinting PIC contents\n");
1861
1862         spin_lock_irqsave(&i8259A_lock, flags);
1863
1864         v = inb(0xa1) << 8 | inb(0x21);
1865         printk(KERN_DEBUG "... PIC  IMR: %04x\n", v);
1866
1867         v = inb(0xa0) << 8 | inb(0x20);
1868         printk(KERN_DEBUG "... PIC  IRR: %04x\n", v);
1869
1870         outb(0x0b,0xa0);
1871         outb(0x0b,0x20);
1872         v = inb(0xa0) << 8 | inb(0x20);
1873         outb(0x0a,0xa0);
1874         outb(0x0a,0x20);
1875
1876         spin_unlock_irqrestore(&i8259A_lock, flags);
1877
1878         printk(KERN_DEBUG "... PIC  ISR: %04x\n", v);
1879
1880         v = inb(0x4d1) << 8 | inb(0x4d0);
1881         printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1882 }
1883
1884 __apicdebuginit(int) print_all_ICs(void)
1885 {
1886         print_PIC();
1887
1888         /* don't print out if apic is not there */
1889         if (!cpu_has_apic || disable_apic)
1890                 return 0;
1891
1892         print_all_local_APICs();
1893         print_IO_APIC();
1894
1895         return 0;
1896 }
1897
1898 fs_initcall(print_all_ICs);
1899
1900
1901 /* Where if anywhere is the i8259 connect in external int mode */
1902 static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
1903
1904 void __init enable_IO_APIC(void)
1905 {
1906         union IO_APIC_reg_01 reg_01;
1907         int i8259_apic, i8259_pin;
1908         int apic;
1909         unsigned long flags;
1910
1911         /*
1912          * The number of IO-APIC IRQ registers (== #pins):
1913          */
1914         for (apic = 0; apic < nr_ioapics; apic++) {
1915                 spin_lock_irqsave(&ioapic_lock, flags);
1916                 reg_01.raw = io_apic_read(apic, 1);
1917                 spin_unlock_irqrestore(&ioapic_lock, flags);
1918                 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
1919         }
1920         for(apic = 0; apic < nr_ioapics; apic++) {
1921                 int pin;
1922                 /* See if any of the pins is in ExtINT mode */
1923                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1924                         struct IO_APIC_route_entry entry;
1925                         entry = ioapic_read_entry(apic, pin);
1926
1927                         /* If the interrupt line is enabled and in ExtInt mode
1928                          * I have found the pin where the i8259 is connected.
1929                          */
1930                         if ((entry.mask == 0) && (entry.delivery_mode == dest_ExtINT)) {
1931                                 ioapic_i8259.apic = apic;
1932                                 ioapic_i8259.pin  = pin;
1933                                 goto found_i8259;
1934                         }
1935                 }
1936         }
1937  found_i8259:
1938         /* Look to see what if the MP table has reported the ExtINT */
1939         /* If we could not find the appropriate pin by looking at the ioapic
1940          * the i8259 probably is not connected the ioapic but give the
1941          * mptable a chance anyway.
1942          */
1943         i8259_pin  = find_isa_irq_pin(0, mp_ExtINT);
1944         i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
1945         /* Trust the MP table if nothing is setup in the hardware */
1946         if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
1947                 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
1948                 ioapic_i8259.pin  = i8259_pin;
1949                 ioapic_i8259.apic = i8259_apic;
1950         }
1951         /* Complain if the MP table and the hardware disagree */
1952         if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
1953                 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
1954         {
1955                 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
1956         }
1957
1958         /*
1959          * Do not trust the IO-APIC being empty at bootup
1960          */
1961         clear_IO_APIC();
1962 }
1963
1964 /*
1965  * Not an __init, needed by the reboot code
1966  */
1967 void disable_IO_APIC(void)
1968 {
1969         /*
1970          * Clear the IO-APIC before rebooting:
1971          */
1972         clear_IO_APIC();
1973
1974         /*
1975          * If the i8259 is routed through an IOAPIC
1976          * Put that IOAPIC in virtual wire mode
1977          * so legacy interrupts can be delivered.
1978          *
1979          * With interrupt-remapping, for now we will use virtual wire A mode,
1980          * as virtual wire B is little complex (need to configure both
1981          * IOAPIC RTE aswell as interrupt-remapping table entry).
1982          * As this gets called during crash dump, keep this simple for now.
1983          */
1984         if (ioapic_i8259.pin != -1 && !intr_remapping_enabled) {
1985                 struct IO_APIC_route_entry entry;
1986
1987                 memset(&entry, 0, sizeof(entry));
1988                 entry.mask            = 0; /* Enabled */
1989                 entry.trigger         = 0; /* Edge */
1990                 entry.irr             = 0;
1991                 entry.polarity        = 0; /* High */
1992                 entry.delivery_status = 0;
1993                 entry.dest_mode       = 0; /* Physical */
1994                 entry.delivery_mode   = dest_ExtINT; /* ExtInt */
1995                 entry.vector          = 0;
1996                 entry.dest            = read_apic_id();
1997
1998                 /*
1999                  * Add it to the IO-APIC irq-routing table:
2000                  */
2001                 ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
2002         }
2003
2004         /*
2005          * Use virtual wire A mode when interrupt remapping is enabled.
2006          */
2007         if (cpu_has_apic)
2008                 disconnect_bsp_APIC(!intr_remapping_enabled &&
2009                                 ioapic_i8259.pin != -1);
2010 }
2011
2012 #ifdef CONFIG_X86_32
2013 /*
2014  * function to set the IO-APIC physical IDs based on the
2015  * values stored in the MPC table.
2016  *
2017  * by Matt Domsch <Matt_Domsch@dell.com>  Tue Dec 21 12:25:05 CST 1999
2018  */
2019
2020 static void __init setup_ioapic_ids_from_mpc(void)
2021 {
2022         union IO_APIC_reg_00 reg_00;
2023         physid_mask_t phys_id_present_map;
2024         int apic_id;
2025         int i;
2026         unsigned char old_id;
2027         unsigned long flags;
2028
2029         if (x86_quirks->setup_ioapic_ids && x86_quirks->setup_ioapic_ids())
2030                 return;
2031
2032         /*
2033          * Don't check I/O APIC IDs for xAPIC systems.  They have
2034          * no meaning without the serial APIC bus.
2035          */
2036         if (!(boot_cpu_data.x86_vendor == X86_VENDOR_INTEL)
2037                 || APIC_XAPIC(apic_version[boot_cpu_physical_apicid]))
2038                 return;
2039         /*
2040          * This is broken; anything with a real cpu count has to
2041          * circumvent this idiocy regardless.
2042          */
2043         phys_id_present_map = apic->ioapic_phys_id_map(phys_cpu_present_map);
2044
2045         /*
2046          * Set the IOAPIC ID to the value stored in the MPC table.
2047          */
2048         for (apic_id = 0; apic_id < nr_ioapics; apic_id++) {
2049
2050                 /* Read the register 0 value */
2051                 spin_lock_irqsave(&ioapic_lock, flags);
2052                 reg_00.raw = io_apic_read(apic_id, 0);
2053                 spin_unlock_irqrestore(&ioapic_lock, flags);
2054
2055                 old_id = mp_ioapics[apic_id].apicid;
2056
2057                 if (mp_ioapics[apic_id].apicid >= get_physical_broadcast()) {
2058                         printk(KERN_ERR "BIOS bug, IO-APIC#%d ID is %d in the MPC table!...\n",
2059                                 apic_id, mp_ioapics[apic_id].apicid);
2060                         printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
2061                                 reg_00.bits.ID);
2062                         mp_ioapics[apic_id].apicid = reg_00.bits.ID;
2063                 }
2064
2065                 /*
2066                  * Sanity check, is the ID really free? Every APIC in a
2067                  * system must have a unique ID or we get lots of nice
2068                  * 'stuck on smp_invalidate_needed IPI wait' messages.
2069                  */
2070                 if (apic->check_apicid_used(phys_id_present_map,
2071                                         mp_ioapics[apic_id].apicid)) {
2072                         printk(KERN_ERR "BIOS bug, IO-APIC#%d ID %d is already used!...\n",
2073                                 apic_id, mp_ioapics[apic_id].apicid);
2074                         for (i = 0; i < get_physical_broadcast(); i++)
2075                                 if (!physid_isset(i, phys_id_present_map))
2076                                         break;
2077                         if (i >= get_physical_broadcast())
2078                                 panic("Max APIC ID exceeded!\n");
2079                         printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
2080                                 i);
2081                         physid_set(i, phys_id_present_map);
2082                         mp_ioapics[apic_id].apicid = i;
2083                 } else {
2084                         physid_mask_t tmp;
2085                         tmp = apic->apicid_to_cpu_present(mp_ioapics[apic_id].apicid);
2086                         apic_printk(APIC_VERBOSE, "Setting %d in the "
2087                                         "phys_id_present_map\n",
2088                                         mp_ioapics[apic_id].apicid);
2089                         physids_or(phys_id_present_map, phys_id_present_map, tmp);
2090                 }
2091
2092
2093                 /*
2094                  * We need to adjust the IRQ routing table
2095                  * if the ID changed.
2096                  */
2097                 if (old_id != mp_ioapics[apic_id].apicid)
2098                         for (i = 0; i < mp_irq_entries; i++)
2099                                 if (mp_irqs[i].dstapic == old_id)
2100                                         mp_irqs[i].dstapic
2101                                                 = mp_ioapics[apic_id].apicid;
2102
2103                 /*
2104                  * Read the right value from the MPC table and
2105                  * write it into the ID register.
2106                  */
2107                 apic_printk(APIC_VERBOSE, KERN_INFO
2108                         "...changing IO-APIC physical APIC ID to %d ...",
2109                         mp_ioapics[apic_id].apicid);
2110
2111                 reg_00.bits.ID = mp_ioapics[apic_id].apicid;
2112                 spin_lock_irqsave(&ioapic_lock, flags);
2113                 io_apic_write(apic_id, 0, reg_00.raw);
2114                 spin_unlock_irqrestore(&ioapic_lock, flags);
2115
2116                 /*
2117                  * Sanity check
2118                  */
2119                 spin_lock_irqsave(&ioapic_lock, flags);
2120                 reg_00.raw = io_apic_read(apic_id, 0);
2121                 spin_unlock_irqrestore(&ioapic_lock, flags);
2122                 if (reg_00.bits.ID != mp_ioapics[apic_id].apicid)
2123                         printk("could not set ID!\n");
2124                 else
2125                         apic_printk(APIC_VERBOSE, " ok.\n");
2126         }
2127 }
2128 #endif
2129
2130 int no_timer_check __initdata;
2131
2132 static int __init notimercheck(char *s)
2133 {
2134         no_timer_check = 1;
2135         return 1;
2136 }
2137 __setup("no_timer_check", notimercheck);
2138
2139 /*
2140  * There is a nasty bug in some older SMP boards, their mptable lies
2141  * about the timer IRQ. We do the following to work around the situation:
2142  *
2143  *      - timer IRQ defaults to IO-APIC IRQ
2144  *      - if this function detects that timer IRQs are defunct, then we fall
2145  *        back to ISA timer IRQs
2146  */
2147 static int __init timer_irq_works(void)
2148 {
2149         unsigned long t1 = jiffies;
2150         unsigned long flags;
2151
2152         if (no_timer_check)
2153                 return 1;
2154
2155         local_save_flags(flags);
2156         local_irq_enable();
2157         /* Let ten ticks pass... */
2158         mdelay((10 * 1000) / HZ);
2159         local_irq_restore(flags);
2160
2161         /*
2162          * Expect a few ticks at least, to be sure some possible
2163          * glue logic does not lock up after one or two first
2164          * ticks in a non-ExtINT mode.  Also the local APIC
2165          * might have cached one ExtINT interrupt.  Finally, at
2166          * least one tick may be lost due to delays.
2167          */
2168
2169         /* jiffies wrap? */
2170         if (time_after(jiffies, t1 + 4))
2171                 return 1;
2172         return 0;
2173 }
2174
2175 /*
2176  * In the SMP+IOAPIC case it might happen that there are an unspecified
2177  * number of pending IRQ events unhandled. These cases are very rare,
2178  * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
2179  * better to do it this way as thus we do not have to be aware of
2180  * 'pending' interrupts in the IRQ path, except at this point.
2181  */
2182 /*
2183  * Edge triggered needs to resend any interrupt
2184  * that was delayed but this is now handled in the device
2185  * independent code.
2186  */
2187
2188 /*
2189  * Starting up a edge-triggered IO-APIC interrupt is
2190  * nasty - we need to make sure that we get the edge.
2191  * If it is already asserted for some reason, we need
2192  * return 1 to indicate that is was pending.
2193  *
2194  * This is not complete - we should be able to fake
2195  * an edge even if it isn't on the 8259A...
2196  */
2197
2198 static unsigned int startup_ioapic_irq(unsigned int irq)
2199 {
2200         int was_pending = 0;
2201         unsigned long flags;
2202         struct irq_cfg *cfg;
2203
2204         spin_lock_irqsave(&ioapic_lock, flags);
2205         if (irq < NR_IRQS_LEGACY) {
2206                 disable_8259A_irq(irq);
2207                 if (i8259A_irq_pending(irq))
2208                         was_pending = 1;
2209         }
2210         cfg = irq_cfg(irq);
2211         __unmask_IO_APIC_irq(cfg);
2212         spin_unlock_irqrestore(&ioapic_lock, flags);
2213
2214         return was_pending;
2215 }
2216
2217 #ifdef CONFIG_X86_64
2218 static int ioapic_retrigger_irq(unsigned int irq)
2219 {
2220
2221         struct irq_cfg *cfg = irq_cfg(irq);
2222         unsigned long flags;
2223
2224         spin_lock_irqsave(&vector_lock, flags);
2225         apic->send_IPI_mask(cpumask_of(cpumask_first(cfg->domain)), cfg->vector);
2226         spin_unlock_irqrestore(&vector_lock, flags);
2227
2228         return 1;
2229 }
2230 #else
2231 static int ioapic_retrigger_irq(unsigned int irq)
2232 {
2233         apic->send_IPI_self(irq_cfg(irq)->vector);
2234
2235         return 1;
2236 }
2237 #endif
2238
2239 /*
2240  * Level and edge triggered IO-APIC interrupts need different handling,
2241  * so we use two separate IRQ descriptors. Edge triggered IRQs can be
2242  * handled with the level-triggered descriptor, but that one has slightly
2243  * more overhead. Level-triggered interrupts cannot be handled with the
2244  * edge-triggered handler, without risking IRQ storms and other ugly
2245  * races.
2246  */
2247
2248 #ifdef CONFIG_SMP
2249 static void send_cleanup_vector(struct irq_cfg *cfg)
2250 {
2251         cpumask_var_t cleanup_mask;
2252
2253         if (unlikely(!alloc_cpumask_var(&cleanup_mask, GFP_ATOMIC))) {
2254                 unsigned int i;
2255                 cfg->move_cleanup_count = 0;
2256                 for_each_cpu_and(i, cfg->old_domain, cpu_online_mask)
2257                         cfg->move_cleanup_count++;
2258                 for_each_cpu_and(i, cfg->old_domain, cpu_online_mask)
2259                         apic->send_IPI_mask(cpumask_of(i), IRQ_MOVE_CLEANUP_VECTOR);
2260         } else {
2261                 cpumask_and(cleanup_mask, cfg->old_domain, cpu_online_mask);
2262                 cfg->move_cleanup_count = cpumask_weight(cleanup_mask);
2263                 apic->send_IPI_mask(cleanup_mask, IRQ_MOVE_CLEANUP_VECTOR);
2264                 free_cpumask_var(cleanup_mask);
2265         }
2266         cfg->move_in_progress = 0;
2267 }
2268
2269 static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, struct irq_cfg *cfg)
2270 {
2271         int apic, pin;
2272         struct irq_pin_list *entry;
2273         u8 vector = cfg->vector;
2274
2275         entry = cfg->irq_2_pin;
2276         for (;;) {
2277                 unsigned int reg;
2278
2279                 if (!entry)
2280                         break;
2281
2282                 apic = entry->apic;
2283                 pin = entry->pin;
2284                 /*
2285                  * With interrupt-remapping, destination information comes
2286                  * from interrupt-remapping table entry.
2287                  */
2288                 if (!irq_remapped(irq))
2289                         io_apic_write(apic, 0x11 + pin*2, dest);
2290                 reg = io_apic_read(apic, 0x10 + pin*2);
2291                 reg &= ~IO_APIC_REDIR_VECTOR_MASK;
2292                 reg |= vector;
2293                 io_apic_modify(apic, 0x10 + pin*2, reg);
2294                 if (!entry->next)
2295                         break;
2296                 entry = entry->next;
2297         }
2298 }
2299
2300 static int
2301 assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask);
2302
2303 /*
2304  * Either sets desc->affinity to a valid value, and returns
2305  * ->cpu_mask_to_apicid of that, or returns BAD_APICID and
2306  * leaves desc->affinity untouched.
2307  */
2308 static unsigned int
2309 set_desc_affinity(struct irq_desc *desc, const struct cpumask *mask)
2310 {
2311         struct irq_cfg *cfg;
2312         unsigned int irq;
2313
2314         if (!cpumask_intersects(mask, cpu_online_mask))
2315                 return BAD_APICID;
2316
2317         irq = desc->irq;
2318         cfg = desc->chip_data;
2319         if (assign_irq_vector(irq, cfg, mask))
2320                 return BAD_APICID;
2321
2322         cpumask_copy(desc->affinity, mask);
2323
2324         return apic->cpu_mask_to_apicid_and(desc->affinity, cfg->domain);
2325 }
2326
2327 static int
2328 set_ioapic_affinity_irq_desc(struct irq_desc *desc, const struct cpumask *mask)
2329 {
2330         struct irq_cfg *cfg;
2331         unsigned long flags;
2332         unsigned int dest;
2333         unsigned int irq;
2334         int ret = -1;
2335
2336         irq = desc->irq;
2337         cfg = desc->chip_data;
2338
2339         spin_lock_irqsave(&ioapic_lock, flags);
2340         dest = set_desc_affinity(desc, mask);
2341         if (dest != BAD_APICID) {
2342                 /* Only the high 8 bits are valid. */
2343                 dest = SET_APIC_LOGICAL_ID(dest);
2344                 __target_IO_APIC_irq(irq, dest, cfg);
2345                 ret = 0;
2346         }
2347         spin_unlock_irqrestore(&ioapic_lock, flags);
2348
2349         return ret;
2350 }
2351
2352 static int
2353 set_ioapic_affinity_irq(unsigned int irq, const struct cpumask *mask)
2354 {
2355         struct irq_desc *desc;
2356
2357         desc = irq_to_desc(irq);
2358
2359         return set_ioapic_affinity_irq_desc(desc, mask);
2360 }
2361
2362 #ifdef CONFIG_INTR_REMAP
2363
2364 /*
2365  * Migrate the IO-APIC irq in the presence of intr-remapping.
2366  *
2367  * For both level and edge triggered, irq migration is a simple atomic
2368  * update(of vector and cpu destination) of IRTE and flush the hardware cache.
2369  *
2370  * For level triggered, we eliminate the io-apic RTE modification (with the
2371  * updated vector information), by using a virtual vector (io-apic pin number).
2372  * Real vector that is used for interrupting cpu will be coming from
2373  * the interrupt-remapping table entry.
2374  */
2375 static int
2376 migrate_ioapic_irq_desc(struct irq_desc *desc, const struct cpumask *mask)
2377 {
2378         struct irq_cfg *cfg;
2379         struct irte irte;
2380         unsigned int dest;
2381         unsigned int irq;
2382         int ret = -1;
2383
2384         if (!cpumask_intersects(mask, cpu_online_mask))
2385                 return ret;
2386
2387         irq = desc->irq;
2388         if (get_irte(irq, &irte))
2389                 return ret;
2390
2391         cfg = desc->chip_data;
2392         if (assign_irq_vector(irq, cfg, mask))
2393                 return ret;
2394
2395         dest = apic->cpu_mask_to_apicid_and(cfg->domain, mask);
2396
2397         irte.vector = cfg->vector;
2398         irte.dest_id = IRTE_DEST(dest);
2399
2400         /*
2401          * Modified the IRTE and flushes the Interrupt entry cache.
2402          */
2403         modify_irte(irq, &irte);
2404
2405         if (cfg->move_in_progress)
2406                 send_cleanup_vector(cfg);
2407
2408         cpumask_copy(desc->affinity, mask);
2409
2410         return 0;
2411 }
2412
2413 /*
2414  * Migrates the IRQ destination in the process context.
2415  */
2416 static int set_ir_ioapic_affinity_irq_desc(struct irq_desc *desc,
2417                                             const struct cpumask *mask)
2418 {
2419         return migrate_ioapic_irq_desc(desc, mask);
2420 }
2421 static int set_ir_ioapic_affinity_irq(unsigned int irq,
2422                                        const struct cpumask *mask)
2423 {
2424         struct irq_desc *desc = irq_to_desc(irq);
2425
2426         return set_ir_ioapic_affinity_irq_desc(desc, mask);
2427 }
2428 #else
2429 static inline int set_ir_ioapic_affinity_irq_desc(struct irq_desc *desc,
2430                                                    const struct cpumask *mask)
2431 {
2432         return 0;
2433 }
2434 #endif
2435
2436 asmlinkage void smp_irq_move_cleanup_interrupt(void)
2437 {
2438         unsigned vector, me;
2439
2440         ack_APIC_irq();
2441         exit_idle();
2442         irq_enter();
2443
2444         me = smp_processor_id();
2445         for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
2446                 unsigned int irq;
2447                 unsigned int irr;
2448                 struct irq_desc *desc;
2449                 struct irq_cfg *cfg;
2450                 irq = __get_cpu_var(vector_irq)[vector];
2451
2452                 if (irq == -1)
2453                         continue;
2454
2455                 desc = irq_to_desc(irq);
2456                 if (!desc)
2457                         continue;
2458
2459                 cfg = irq_cfg(irq);
2460                 spin_lock(&desc->lock);
2461                 if (!cfg->move_cleanup_count)
2462                         goto unlock;
2463
2464                 if (vector == cfg->vector && cpumask_test_cpu(me, cfg->domain))
2465                         goto unlock;
2466
2467                 irr = apic_read(APIC_IRR + (vector / 32 * 0x10));
2468                 /*
2469                  * Check if the vector that needs to be cleanedup is
2470                  * registered at the cpu's IRR. If so, then this is not
2471                  * the best time to clean it up. Lets clean it up in the
2472                  * next attempt by sending another IRQ_MOVE_CLEANUP_VECTOR
2473                  * to myself.
2474                  */
2475                 if (irr  & (1 << (vector % 32))) {
2476                         apic->send_IPI_self(IRQ_MOVE_CLEANUP_VECTOR);
2477                         goto unlock;
2478                 }
2479                 __get_cpu_var(vector_irq)[vector] = -1;
2480                 cfg->move_cleanup_count--;
2481 unlock:
2482                 spin_unlock(&desc->lock);
2483         }
2484
2485         irq_exit();
2486 }
2487
2488 static void irq_complete_move(struct irq_desc **descp)
2489 {
2490         struct irq_desc *desc = *descp;
2491         struct irq_cfg *cfg = desc->chip_data;
2492         unsigned vector, me;
2493
2494         if (likely(!cfg->move_in_progress))
2495                 return;
2496
2497         vector = ~get_irq_regs()->orig_ax;
2498         me = smp_processor_id();
2499
2500         if (vector == cfg->vector && cpumask_test_cpu(me, cfg->domain))
2501                 send_cleanup_vector(cfg);
2502 }
2503 #else
2504 static inline void irq_complete_move(struct irq_desc **descp) {}
2505 #endif
2506
2507 static void ack_apic_edge(unsigned int irq)
2508 {
2509         struct irq_desc *desc = irq_to_desc(irq);
2510
2511         irq_complete_move(&desc);
2512         move_native_irq(irq);
2513         ack_APIC_irq();
2514 }
2515
2516 atomic_t irq_mis_count;
2517
2518 static void ack_apic_level(unsigned int irq)
2519 {
2520         struct irq_desc *desc = irq_to_desc(irq);
2521
2522 #ifdef CONFIG_X86_32
2523         unsigned long v;
2524         int i;
2525 #endif
2526         struct irq_cfg *cfg;
2527         int do_unmask_irq = 0;
2528
2529         irq_complete_move(&desc);
2530 #ifdef CONFIG_GENERIC_PENDING_IRQ
2531         /* If we are moving the irq we need to mask it */
2532         if (unlikely(desc->status & IRQ_MOVE_PENDING)) {
2533                 do_unmask_irq = 1;
2534                 mask_IO_APIC_irq_desc(desc);
2535         }
2536 #endif
2537
2538 #ifdef CONFIG_X86_32
2539         /*
2540         * It appears there is an erratum which affects at least version 0x11
2541         * of I/O APIC (that's the 82093AA and cores integrated into various
2542         * chipsets).  Under certain conditions a level-triggered interrupt is
2543         * erroneously delivered as edge-triggered one but the respective IRR
2544         * bit gets set nevertheless.  As a result the I/O unit expects an EOI
2545         * message but it will never arrive and further interrupts are blocked
2546         * from the source.  The exact reason is so far unknown, but the
2547         * phenomenon was observed when two consecutive interrupt requests
2548         * from a given source get delivered to the same CPU and the source is
2549         * temporarily disabled in between.
2550         *
2551         * A workaround is to simulate an EOI message manually.  We achieve it
2552         * by setting the trigger mode to edge and then to level when the edge
2553         * trigger mode gets detected in the TMR of a local APIC for a
2554         * level-triggered interrupt.  We mask the source for the time of the
2555         * operation to prevent an edge-triggered interrupt escaping meanwhile.
2556         * The idea is from Manfred Spraul.  --macro
2557         */
2558         cfg = desc->chip_data;
2559         i = cfg->vector;
2560
2561         v = apic_read(APIC_TMR + ((i & ~0x1f) >> 1));
2562 #endif
2563
2564         /*
2565          * We must acknowledge the irq before we move it or the acknowledge will
2566          * not propagate properly.
2567          */
2568         ack_APIC_irq();
2569
2570         /* Now we can move and renable the irq */
2571         if (unlikely(do_unmask_irq)) {
2572                 /* Only migrate the irq if the ack has been received.
2573                  *
2574                  * On rare occasions the broadcast level triggered ack gets
2575                  * delayed going to ioapics, and if we reprogram the
2576                  * vector while Remote IRR is still set the irq will never
2577                  * fire again.
2578                  *
2579                  * To prevent this scenario we read the Remote IRR bit
2580                  * of the ioapic.  This has two effects.
2581                  * - On any sane system the read of the ioapic will
2582                  *   flush writes (and acks) going to the ioapic from
2583                  *   this cpu.
2584                  * - We get to see if the ACK has actually been delivered.
2585                  *
2586                  * Based on failed experiments of reprogramming the
2587                  * ioapic entry from outside of irq context starting
2588                  * with masking the ioapic entry and then polling until
2589                  * Remote IRR was clear before reprogramming the
2590                  * ioapic I don't trust the Remote IRR bit to be
2591                  * completey accurate.
2592                  *
2593                  * However there appears to be no other way to plug
2594                  * this race, so if the Remote IRR bit is not
2595                  * accurate and is causing problems then it is a hardware bug
2596                  * and you can go talk to the chipset vendor about it.
2597                  */
2598                 cfg = desc->chip_data;
2599                 if (!io_apic_level_ack_pending(cfg))
2600                         move_masked_irq(irq);
2601                 unmask_IO_APIC_irq_desc(desc);
2602         }
2603
2604 #ifdef CONFIG_X86_32
2605         if (!(v & (1 << (i & 0x1f)))) {
2606                 atomic_inc(&irq_mis_count);
2607                 spin_lock(&ioapic_lock);
2608                 __mask_and_edge_IO_APIC_irq(cfg);
2609                 __unmask_and_level_IO_APIC_irq(cfg);
2610                 spin_unlock(&ioapic_lock);
2611         }
2612 #endif
2613 }
2614
2615 #ifdef CONFIG_INTR_REMAP
2616 static void __eoi_ioapic_irq(unsigned int irq, struct irq_cfg *cfg)
2617 {
2618         int apic, pin;
2619         struct irq_pin_list *entry;
2620
2621         entry = cfg->irq_2_pin;
2622         for (;;) {
2623
2624                 if (!entry)
2625                         break;
2626
2627                 apic = entry->apic;
2628                 pin = entry->pin;
2629                 io_apic_eoi(apic, pin);
2630                 entry = entry->next;
2631         }
2632 }
2633
2634 static void
2635 eoi_ioapic_irq(struct irq_desc *desc)
2636 {
2637         struct irq_cfg *cfg;
2638         unsigned long flags;
2639         unsigned int irq;
2640
2641         irq = desc->irq;
2642         cfg = desc->chip_data;
2643
2644         spin_lock_irqsave(&ioapic_lock, flags);
2645         __eoi_ioapic_irq(irq, cfg);
2646         spin_unlock_irqrestore(&ioapic_lock, flags);
2647 }
2648
2649 static void ir_ack_apic_edge(unsigned int irq)
2650 {
2651         ack_APIC_irq();
2652 }
2653
2654 static void ir_ack_apic_level(unsigned int irq)
2655 {
2656         struct irq_desc *desc = irq_to_desc(irq);
2657
2658         ack_APIC_irq();
2659         eoi_ioapic_irq(desc);
2660 }
2661 #endif /* CONFIG_INTR_REMAP */
2662
2663 static struct irq_chip ioapic_chip __read_mostly = {
2664         .name           = "IO-APIC",
2665         .startup        = startup_ioapic_irq,
2666         .mask           = mask_IO_APIC_irq,
2667         .unmask         = unmask_IO_APIC_irq,
2668         .ack            = ack_apic_edge,
2669         .eoi            = ack_apic_level,
2670 #ifdef CONFIG_SMP
2671         .set_affinity   = set_ioapic_affinity_irq,
2672 #endif
2673         .retrigger      = ioapic_retrigger_irq,
2674 };
2675
2676 static struct irq_chip ir_ioapic_chip __read_mostly = {
2677         .name           = "IR-IO-APIC",
2678         .startup        = startup_ioapic_irq,
2679         .mask           = mask_IO_APIC_irq,
2680         .unmask         = unmask_IO_APIC_irq,
2681 #ifdef CONFIG_INTR_REMAP
2682         .ack            = ir_ack_apic_edge,
2683         .eoi            = ir_ack_apic_level,
2684 #ifdef CONFIG_SMP
2685         .set_affinity   = set_ir_ioapic_affinity_irq,
2686 #endif
2687 #endif
2688         .retrigger      = ioapic_retrigger_irq,
2689 };
2690
2691 static inline void init_IO_APIC_traps(void)
2692 {
2693         int irq;
2694         struct irq_desc *desc;
2695         struct irq_cfg *cfg;
2696
2697         /*
2698          * NOTE! The local APIC isn't very good at handling
2699          * multiple interrupts at the same interrupt level.
2700          * As the interrupt level is determined by taking the
2701          * vector number and shifting that right by 4, we
2702          * want to spread these out a bit so that they don't
2703          * all fall in the same interrupt level.
2704          *
2705          * Also, we've got to be careful not to trash gate
2706          * 0x80, because int 0x80 is hm, kind of importantish. ;)
2707          */
2708         for_each_irq_desc(irq, desc) {
2709                 cfg = desc->chip_data;
2710                 if (IO_APIC_IRQ(irq) && cfg && !cfg->vector) {
2711                         /*
2712                          * Hmm.. We don't have an entry for this,
2713                          * so default to an old-fashioned 8259
2714                          * interrupt if we can..
2715                          */
2716                         if (irq < NR_IRQS_LEGACY)
2717                                 make_8259A_irq(irq);
2718                         else
2719                                 /* Strange. Oh, well.. */
2720                                 desc->chip = &no_irq_chip;
2721                 }
2722         }
2723 }
2724
2725 /*
2726  * The local APIC irq-chip implementation:
2727  */
2728
2729 static void mask_lapic_irq(unsigned int irq)
2730 {
2731         unsigned long v;
2732
2733         v = apic_read(APIC_LVT0);
2734         apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
2735 }
2736
2737 static void unmask_lapic_irq(unsigned int irq)
2738 {
2739         unsigned long v;
2740
2741         v = apic_read(APIC_LVT0);
2742         apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
2743 }
2744
2745 static void ack_lapic_irq(unsigned int irq)
2746 {
2747         ack_APIC_irq();
2748 }
2749
2750 static struct irq_chip lapic_chip __read_mostly = {
2751         .name           = "local-APIC",
2752         .mask           = mask_lapic_irq,
2753         .unmask         = unmask_lapic_irq,
2754         .ack            = ack_lapic_irq,
2755 };
2756
2757 static void lapic_register_intr(int irq, struct irq_desc *desc)
2758 {
2759         desc->status &= ~IRQ_LEVEL;
2760         set_irq_chip_and_handler_name(irq, &lapic_chip, handle_edge_irq,
2761                                       "edge");
2762 }
2763
2764 static void __init setup_nmi(void)
2765 {
2766         /*
2767          * Dirty trick to enable the NMI watchdog ...
2768          * We put the 8259A master into AEOI mode and
2769          * unmask on all local APICs LVT0 as NMI.
2770          *
2771          * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
2772          * is from Maciej W. Rozycki - so we do not have to EOI from
2773          * the NMI handler or the timer interrupt.
2774          */
2775         apic_printk(APIC_VERBOSE, KERN_INFO "activating NMI Watchdog ...");
2776
2777         enable_NMI_through_LVT0();
2778
2779         apic_printk(APIC_VERBOSE, " done.\n");
2780 }
2781
2782 /*
2783  * This looks a bit hackish but it's about the only one way of sending
2784  * a few INTA cycles to 8259As and any associated glue logic.  ICR does
2785  * not support the ExtINT mode, unfortunately.  We need to send these
2786  * cycles as some i82489DX-based boards have glue logic that keeps the
2787  * 8259A interrupt line asserted until INTA.  --macro
2788  */
2789 static inline void __init unlock_ExtINT_logic(void)
2790 {
2791         int apic, pin, i;
2792         struct IO_APIC_route_entry entry0, entry1;
2793         unsigned char save_control, save_freq_select;
2794
2795         pin  = find_isa_irq_pin(8, mp_INT);
2796         if (pin == -1) {
2797                 WARN_ON_ONCE(1);
2798                 return;
2799         }
2800         apic = find_isa_irq_apic(8, mp_INT);
2801         if (apic == -1) {
2802                 WARN_ON_ONCE(1);
2803                 return;
2804         }
2805
2806         entry0 = ioapic_read_entry(apic, pin);
2807         clear_IO_APIC_pin(apic, pin);
2808
2809         memset(&entry1, 0, sizeof(entry1));
2810
2811         entry1.dest_mode = 0;                   /* physical delivery */
2812         entry1.mask = 0;                        /* unmask IRQ now */
2813         entry1.dest = hard_smp_processor_id();
2814         entry1.delivery_mode = dest_ExtINT;
2815         entry1.polarity = entry0.polarity;
2816         entry1.trigger = 0;
2817         entry1.vector = 0;
2818
2819         ioapic_write_entry(apic, pin, entry1);
2820
2821         save_control = CMOS_READ(RTC_CONTROL);
2822         save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
2823         CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
2824                    RTC_FREQ_SELECT);
2825         CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
2826
2827         i = 100;
2828         while (i-- > 0) {
2829                 mdelay(10);
2830                 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
2831                         i -= 10;
2832         }
2833
2834         CMOS_WRITE(save_control, RTC_CONTROL);
2835         CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
2836         clear_IO_APIC_pin(apic, pin);
2837
2838         ioapic_write_entry(apic, pin, entry0);
2839 }
2840
2841 static int disable_timer_pin_1 __initdata;
2842 /* Actually the next is obsolete, but keep it for paranoid reasons -AK */
2843 static int __init disable_timer_pin_setup(char *arg)
2844 {
2845         disable_timer_pin_1 = 1;
2846         return 0;
2847 }
2848 early_param("disable_timer_pin_1", disable_timer_pin_setup);
2849
2850 int timer_through_8259 __initdata;
2851
2852 /*
2853  * This code may look a bit paranoid, but it's supposed to cooperate with
2854  * a wide range of boards and BIOS bugs.  Fortunately only the timer IRQ
2855  * is so screwy.  Thanks to Brian Perkins for testing/hacking this beast
2856  * fanatically on his truly buggy board.
2857  *
2858  * FIXME: really need to revamp this for all platforms.
2859  */
2860 static inline void __init check_timer(void)
2861 {
2862         struct irq_desc *desc = irq_to_desc(0);
2863         struct irq_cfg *cfg = desc->chip_data;
2864         int node = cpu_to_node(boot_cpu_id);
2865         int apic1, pin1, apic2, pin2;
2866         unsigned long flags;
2867         int no_pin1 = 0;
2868
2869         local_irq_save(flags);
2870
2871         /*
2872          * get/set the timer IRQ vector:
2873          */
2874         disable_8259A_irq(0);
2875         assign_irq_vector(0, cfg, apic->target_cpus());
2876
2877         /*
2878          * As IRQ0 is to be enabled in the 8259A, the virtual
2879          * wire has to be disabled in the local APIC.  Also
2880          * timer interrupts need to be acknowledged manually in
2881          * the 8259A for the i82489DX when using the NMI
2882          * watchdog as that APIC treats NMIs as level-triggered.
2883          * The AEOI mode will finish them in the 8259A
2884          * automatically.
2885          */
2886         apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
2887         init_8259A(1);
2888 #ifdef CONFIG_X86_32
2889         {
2890                 unsigned int ver;
2891
2892                 ver = apic_read(APIC_LVR);
2893                 ver = GET_APIC_VERSION(ver);
2894                 timer_ack = (nmi_watchdog == NMI_IO_APIC && !APIC_INTEGRATED(ver));
2895         }
2896 #endif
2897
2898         pin1  = find_isa_irq_pin(0, mp_INT);
2899         apic1 = find_isa_irq_apic(0, mp_INT);
2900         pin2  = ioapic_i8259.pin;
2901         apic2 = ioapic_i8259.apic;
2902
2903         apic_printk(APIC_QUIET, KERN_INFO "..TIMER: vector=0x%02X "
2904                     "apic1=%d pin1=%d apic2=%d pin2=%d\n",
2905                     cfg->vector, apic1, pin1, apic2, pin2);
2906
2907         /*
2908          * Some BIOS writers are clueless and report the ExtINTA
2909          * I/O APIC input from the cascaded 8259A as the timer
2910          * interrupt input.  So just in case, if only one pin
2911          * was found above, try it both directly and through the
2912          * 8259A.
2913          */
2914         if (pin1 == -1) {
2915                 if (intr_remapping_enabled)
2916                         panic("BIOS bug: timer not connected to IO-APIC");
2917                 pin1 = pin2;
2918                 apic1 = apic2;
2919                 no_pin1 = 1;
2920         } else if (pin2 == -1) {
2921                 pin2 = pin1;
2922                 apic2 = apic1;
2923         }
2924
2925         if (pin1 != -1) {
2926                 /*
2927                  * Ok, does IRQ0 through the IOAPIC work?
2928                  */
2929                 if (no_pin1) {
2930                         add_pin_to_irq_node(cfg, node, apic1, pin1);
2931                         setup_timer_IRQ0_pin(apic1, pin1, cfg->vector);
2932                 } else {
2933                         /* for edge trigger, setup_IO_APIC_irq already
2934                          * leave it unmasked.
2935                          * so only need to unmask if it is level-trigger
2936                          * do we really have level trigger timer?
2937                          */
2938                         int idx;
2939                         idx = find_irq_entry(apic1, pin1, mp_INT);
2940                         if (idx != -1 && irq_trigger(idx))
2941                                 unmask_IO_APIC_irq_desc(desc);
2942                 }
2943                 if (timer_irq_works()) {
2944                         if (nmi_watchdog == NMI_IO_APIC) {
2945                                 setup_nmi();
2946                                 enable_8259A_irq(0);
2947                         }
2948                         if (disable_timer_pin_1 > 0)
2949                                 clear_IO_APIC_pin(0, pin1);
2950                         goto out;
2951                 }
2952                 if (intr_remapping_enabled)
2953                         panic("timer doesn't work through Interrupt-remapped IO-APIC");
2954                 local_irq_disable();
2955                 clear_IO_APIC_pin(apic1, pin1);
2956                 if (!no_pin1)
2957                         apic_printk(APIC_QUIET, KERN_ERR "..MP-BIOS bug: "
2958                                     "8254 timer not connected to IO-APIC\n");
2959
2960                 apic_printk(APIC_QUIET, KERN_INFO "...trying to set up timer "
2961                             "(IRQ0) through the 8259A ...\n");
2962                 apic_printk(APIC_QUIET, KERN_INFO
2963                             "..... (found apic %d pin %d) ...\n", apic2, pin2);
2964                 /*
2965                  * legacy devices should be connected to IO APIC #0
2966                  */
2967                 replace_pin_at_irq_node(cfg, node, apic1, pin1, apic2, pin2);
2968                 setup_timer_IRQ0_pin(apic2, pin2, cfg->vector);
2969                 enable_8259A_irq(0);
2970                 if (timer_irq_works()) {
2971                         apic_printk(APIC_QUIET, KERN_INFO "....... works.\n");
2972                         timer_through_8259 = 1;
2973                         if (nmi_watchdog == NMI_IO_APIC) {
2974                                 disable_8259A_irq(0);
2975                                 setup_nmi();
2976                                 enable_8259A_irq(0);
2977                         }
2978                         goto out;
2979                 }
2980                 /*
2981                  * Cleanup, just in case ...
2982                  */
2983                 local_irq_disable();
2984                 disable_8259A_irq(0);
2985                 clear_IO_APIC_pin(apic2, pin2);
2986                 apic_printk(APIC_QUIET, KERN_INFO "....... failed.\n");
2987         }
2988
2989         if (nmi_watchdog == NMI_IO_APIC) {
2990                 apic_printk(APIC_QUIET, KERN_WARNING "timer doesn't work "
2991                             "through the IO-APIC - disabling NMI Watchdog!\n");
2992                 nmi_watchdog = NMI_NONE;
2993         }
2994 #ifdef CONFIG_X86_32
2995         timer_ack = 0;
2996 #endif
2997
2998         apic_printk(APIC_QUIET, KERN_INFO
2999                     "...trying to set up timer as Virtual Wire IRQ...\n");
3000
3001         lapic_register_intr(0, desc);
3002         apic_write(APIC_LVT0, APIC_DM_FIXED | cfg->vector);     /* Fixed mode */
3003         enable_8259A_irq(0);
3004
3005         if (timer_irq_works()) {
3006                 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
3007                 goto out;
3008         }
3009         local_irq_disable();
3010         disable_8259A_irq(0);
3011         apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | cfg->vector);
3012         apic_printk(APIC_QUIET, KERN_INFO "..... failed.\n");
3013
3014         apic_printk(APIC_QUIET, KERN_INFO
3015                     "...trying to set up timer as ExtINT IRQ...\n");
3016
3017         init_8259A(0);
3018         make_8259A_irq(0);
3019         apic_write(APIC_LVT0, APIC_DM_EXTINT);
3020
3021         unlock_ExtINT_logic();
3022
3023         if (timer_irq_works()) {
3024                 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
3025                 goto out;
3026         }
3027         local_irq_disable();
3028         apic_printk(APIC_QUIET, KERN_INFO "..... failed :(.\n");
3029         panic("IO-APIC + timer doesn't work!  Boot with apic=debug and send a "
3030                 "report.  Then try booting with the 'noapic' option.\n");
3031 out:
3032         local_irq_restore(flags);
3033 }
3034
3035 /*
3036  * Traditionally ISA IRQ2 is the cascade IRQ, and is not available
3037  * to devices.  However there may be an I/O APIC pin available for
3038  * this interrupt regardless.  The pin may be left unconnected, but
3039  * typically it will be reused as an ExtINT cascade interrupt for
3040  * the master 8259A.  In the MPS case such a pin will normally be
3041  * reported as an ExtINT interrupt in the MP table.  With ACPI
3042  * there is no provision for ExtINT interrupts, and in the absence
3043  * of an override it would be treated as an ordinary ISA I/O APIC
3044  * interrupt, that is edge-triggered and unmasked by default.  We
3045  * used to do this, but it caused problems on some systems because
3046  * of the NMI watchdog and sometimes IRQ0 of the 8254 timer using
3047  * the same ExtINT cascade interrupt to drive the local APIC of the
3048  * bootstrap processor.  Therefore we refrain from routing IRQ2 to
3049  * the I/O APIC in all cases now.  No actual device should request
3050  * it anyway.  --macro
3051  */
3052 #define PIC_IRQS        (1 << PIC_CASCADE_IR)
3053
3054 void __init setup_IO_APIC(void)
3055 {
3056
3057         /*
3058          * calling enable_IO_APIC() is moved to setup_local_APIC for BP
3059          */
3060
3061         io_apic_irqs = ~PIC_IRQS;
3062
3063         apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
3064         /*
3065          * Set up IO-APIC IRQ routing.
3066          */
3067 #ifdef CONFIG_X86_32
3068         if (!acpi_ioapic)
3069                 setup_ioapic_ids_from_mpc();
3070 #endif
3071         sync_Arb_IDs();
3072         setup_IO_APIC_irqs();
3073         init_IO_APIC_traps();
3074         check_timer();
3075 }
3076
3077 /*
3078  *      Called after all the initialization is done. If we didnt find any
3079  *      APIC bugs then we can allow the modify fast path
3080  */
3081
3082 static int __init io_apic_bug_finalize(void)
3083 {
3084         if (sis_apic_bug == -1)
3085                 sis_apic_bug = 0;
3086         return 0;
3087 }
3088
3089 late_initcall(io_apic_bug_finalize);
3090
3091 struct sysfs_ioapic_data {
3092         struct sys_device dev;
3093         struct IO_APIC_route_entry entry[0];
3094 };
3095 static struct sysfs_ioapic_data * mp_ioapic_data[MAX_IO_APICS];
3096
3097 static int ioapic_suspend(struct sys_device *dev, pm_message_t state)
3098 {
3099         struct IO_APIC_route_entry *entry;
3100         struct sysfs_ioapic_data *data;
3101         int i;
3102
3103         data = container_of(dev, struct sysfs_ioapic_data, dev);
3104         entry = data->entry;
3105         for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ )
3106                 *entry = ioapic_read_entry(dev->id, i);
3107
3108         return 0;
3109 }
3110
3111 static int ioapic_resume(struct sys_device *dev)
3112 {
3113         struct IO_APIC_route_entry *entry;
3114         struct sysfs_ioapic_data *data;
3115         unsigned long flags;
3116         union IO_APIC_reg_00 reg_00;
3117         int i;
3118
3119         data = container_of(dev, struct sysfs_ioapic_data, dev);
3120         entry = data->entry;
3121
3122         spin_lock_irqsave(&ioapic_lock, flags);
3123         reg_00.raw = io_apic_read(dev->id, 0);
3124         if (reg_00.bits.ID != mp_ioapics[dev->id].apicid) {
3125                 reg_00.bits.ID = mp_ioapics[dev->id].apicid;
3126                 io_apic_write(dev->id, 0, reg_00.raw);
3127         }
3128         spin_unlock_irqrestore(&ioapic_lock, flags);
3129         for (i = 0; i < nr_ioapic_registers[dev->id]; i++)
3130                 ioapic_write_entry(dev->id, i, entry[i]);
3131
3132         return 0;
3133 }
3134
3135 static struct sysdev_class ioapic_sysdev_class = {
3136         .name = "ioapic",
3137         .suspend = ioapic_suspend,
3138         .resume = ioapic_resume,
3139 };
3140
3141 static int __init ioapic_init_sysfs(void)
3142 {
3143         struct sys_device * dev;
3144         int i, size, error;
3145
3146         error = sysdev_class_register(&ioapic_sysdev_class);
3147         if (error)
3148                 return error;
3149
3150         for (i = 0; i < nr_ioapics; i++ ) {
3151                 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
3152                         * sizeof(struct IO_APIC_route_entry);
3153                 mp_ioapic_data[i] = kzalloc(size, GFP_KERNEL);
3154                 if (!mp_ioapic_data[i]) {
3155                         printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
3156                         continue;
3157                 }
3158                 dev = &mp_ioapic_data[i]->dev;
3159                 dev->id = i;
3160                 dev->cls = &ioapic_sysdev_class;
3161                 error = sysdev_register(dev);
3162                 if (error) {
3163                         kfree(mp_ioapic_data[i]);
3164                         mp_ioapic_data[i] = NULL;
3165                         printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
3166                         continue;
3167                 }
3168         }
3169
3170         return 0;
3171 }
3172
3173 device_initcall(ioapic_init_sysfs);
3174
3175 static int nr_irqs_gsi = NR_IRQS_LEGACY;
3176 /*
3177  * Dynamic irq allocate and deallocation
3178  */
3179 unsigned int create_irq_nr(unsigned int irq_want, int node)
3180 {
3181         /* Allocate an unused irq */
3182         unsigned int irq;
3183         unsigned int new;
3184         unsigned long flags;
3185         struct irq_cfg *cfg_new = NULL;
3186         struct irq_desc *desc_new = NULL;
3187
3188         irq = 0;
3189         if (irq_want < nr_irqs_gsi)
3190                 irq_want = nr_irqs_gsi;
3191
3192         spin_lock_irqsave(&vector_lock, flags);
3193         for (new = irq_want; new < nr_irqs; new++) {
3194                 desc_new = irq_to_desc_alloc_node(new, node);
3195                 if (!desc_new) {
3196                         printk(KERN_INFO "can not get irq_desc for %d\n", new);
3197                         continue;
3198                 }
3199                 cfg_new = desc_new->chip_data;
3200
3201                 if (cfg_new->vector != 0)
3202                         continue;
3203
3204                 desc_new = move_irq_desc(desc_new, node);
3205
3206                 if (__assign_irq_vector(new, cfg_new, apic->target_cpus()) == 0)
3207                         irq = new;
3208                 break;
3209         }
3210         spin_unlock_irqrestore(&vector_lock, flags);
3211
3212         if (irq > 0) {
3213                 dynamic_irq_init(irq);
3214                 /* restore it, in case dynamic_irq_init clear it */
3215                 if (desc_new)
3216                         desc_new->chip_data = cfg_new;
3217         }
3218         return irq;
3219 }
3220
3221 int create_irq(void)
3222 {
3223         int node = cpu_to_node(boot_cpu_id);
3224         unsigned int irq_want;
3225         int irq;
3226
3227         irq_want = nr_irqs_gsi;
3228         irq = create_irq_nr(irq_want, node);
3229
3230         if (irq == 0)
3231                 irq = -1;
3232
3233         return irq;
3234 }
3235
3236 void destroy_irq(unsigned int irq)
3237 {
3238         unsigned long flags;
3239         struct irq_cfg *cfg;
3240         struct irq_desc *desc;
3241
3242         /* store it, in case dynamic_irq_cleanup clear it */
3243         desc = irq_to_desc(irq);
3244         cfg = desc->chip_data;
3245         dynamic_irq_cleanup(irq);
3246         /* connect back irq_cfg */
3247         if (desc)
3248                 desc->chip_data = cfg;
3249
3250         free_irte(irq);
3251         spin_lock_irqsave(&vector_lock, flags);
3252         __clear_irq_vector(irq, cfg);
3253         spin_unlock_irqrestore(&vector_lock, flags);
3254 }
3255
3256 /*
3257  * MSI message composition
3258  */
3259 #ifdef CONFIG_PCI_MSI
3260 static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq, struct msi_msg *msg)
3261 {
3262         struct irq_cfg *cfg;
3263         int err;
3264         unsigned dest;
3265
3266         if (disable_apic)
3267                 return -ENXIO;
3268
3269         cfg = irq_cfg(irq);
3270         err = assign_irq_vector(irq, cfg, apic->target_cpus());
3271         if (err)
3272                 return err;
3273
3274         dest = apic->cpu_mask_to_apicid_and(cfg->domain, apic->target_cpus());
3275
3276         if (irq_remapped(irq)) {
3277                 struct irte irte;
3278                 int ir_index;
3279                 u16 sub_handle;
3280
3281                 ir_index = map_irq_to_irte_handle(irq, &sub_handle);
3282                 BUG_ON(ir_index == -1);
3283
3284                 memset (&irte, 0, sizeof(irte));
3285
3286                 irte.present = 1;
3287                 irte.dst_mode = apic->irq_dest_mode;
3288                 irte.trigger_mode = 0; /* edge */
3289                 irte.dlvry_mode = apic->irq_delivery_mode;
3290                 irte.vector = cfg->vector;
3291                 irte.dest_id = IRTE_DEST(dest);
3292
3293                 modify_irte(irq, &irte);
3294
3295                 msg->address_hi = MSI_ADDR_BASE_HI;
3296                 msg->data = sub_handle;
3297                 msg->address_lo = MSI_ADDR_BASE_LO | MSI_ADDR_IR_EXT_INT |
3298                                   MSI_ADDR_IR_SHV |
3299                                   MSI_ADDR_IR_INDEX1(ir_index) |
3300                                   MSI_ADDR_IR_INDEX2(ir_index);
3301         } else {
3302                 if (x2apic_enabled())
3303                         msg->address_hi = MSI_ADDR_BASE_HI |
3304                                           MSI_ADDR_EXT_DEST_ID(dest);
3305                 else
3306                         msg->address_hi = MSI_ADDR_BASE_HI;
3307
3308                 msg->address_lo =
3309                         MSI_ADDR_BASE_LO |
3310                         ((apic->irq_dest_mode == 0) ?
3311                                 MSI_ADDR_DEST_MODE_PHYSICAL:
3312                                 MSI_ADDR_DEST_MODE_LOGICAL) |
3313                         ((apic->irq_delivery_mode != dest_LowestPrio) ?
3314                                 MSI_ADDR_REDIRECTION_CPU:
3315                                 MSI_ADDR_REDIRECTION_LOWPRI) |
3316                         MSI_ADDR_DEST_ID(dest);
3317
3318                 msg->data =
3319                         MSI_DATA_TRIGGER_EDGE |
3320                         MSI_DATA_LEVEL_ASSERT |
3321                         ((apic->irq_delivery_mode != dest_LowestPrio) ?
3322                                 MSI_DATA_DELIVERY_FIXED:
3323                                 MSI_DATA_DELIVERY_LOWPRI) |
3324                         MSI_DATA_VECTOR(cfg->vector);
3325         }
3326         return err;
3327 }
3328
3329 #ifdef CONFIG_SMP
3330 static int set_msi_irq_affinity(unsigned int irq, const struct cpumask *mask)
3331 {
3332         struct irq_desc *desc = irq_to_desc(irq);
3333         struct irq_cfg *cfg;
3334         struct msi_msg msg;
3335         unsigned int dest;
3336
3337         dest = set_desc_affinity(desc, mask);
3338         if (dest == BAD_APICID)
3339                 return -1;
3340
3341         cfg = desc->chip_data;
3342
3343         read_msi_msg_desc(desc, &msg);
3344
3345         msg.data &= ~MSI_DATA_VECTOR_MASK;
3346         msg.data |= MSI_DATA_VECTOR(cfg->vector);
3347         msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3348         msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3349
3350         write_msi_msg_desc(desc, &msg);
3351
3352         return 0;
3353 }
3354 #ifdef CONFIG_INTR_REMAP
3355 /*
3356  * Migrate the MSI irq to another cpumask. This migration is
3357  * done in the process context using interrupt-remapping hardware.
3358  */
3359 static int
3360 ir_set_msi_irq_affinity(unsigned int irq, const struct cpumask *mask)
3361 {
3362         struct irq_desc *desc = irq_to_desc(irq);
3363         struct irq_cfg *cfg = desc->chip_data;
3364         unsigned int dest;
3365         struct irte irte;
3366
3367         if (get_irte(irq, &irte))
3368                 return -1;
3369
3370         dest = set_desc_affinity(desc, mask);
3371         if (dest == BAD_APICID)
3372                 return -1;
3373
3374         irte.vector = cfg->vector;
3375         irte.dest_id = IRTE_DEST(dest);
3376
3377         /*
3378          * atomically update the IRTE with the new destination and vector.
3379          */
3380         modify_irte(irq, &irte);
3381
3382         /*
3383          * After this point, all the interrupts will start arriving
3384          * at the new destination. So, time to cleanup the previous
3385          * vector allocation.
3386          */
3387         if (cfg->move_in_progress)
3388                 send_cleanup_vector(cfg);
3389
3390         return 0;
3391 }
3392
3393 #endif
3394 #endif /* CONFIG_SMP */
3395
3396 /*
3397  * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
3398  * which implement the MSI or MSI-X Capability Structure.
3399  */
3400 static struct irq_chip msi_chip = {
3401         .name           = "PCI-MSI",
3402         .unmask         = unmask_msi_irq,
3403         .mask           = mask_msi_irq,
3404         .ack            = ack_apic_edge,
3405 #ifdef CONFIG_SMP
3406         .set_affinity   = set_msi_irq_affinity,
3407 #endif
3408         .retrigger      = ioapic_retrigger_irq,
3409 };
3410
3411 static struct irq_chip msi_ir_chip = {
3412         .name           = "IR-PCI-MSI",
3413         .unmask         = unmask_msi_irq,
3414         .mask           = mask_msi_irq,
3415 #ifdef CONFIG_INTR_REMAP
3416         .ack            = ir_ack_apic_edge,
3417 #ifdef CONFIG_SMP
3418         .set_affinity   = ir_set_msi_irq_affinity,
3419 #endif
3420 #endif
3421         .retrigger      = ioapic_retrigger_irq,
3422 };
3423
3424 /*
3425  * Map the PCI dev to the corresponding remapping hardware unit
3426  * and allocate 'nvec' consecutive interrupt-remapping table entries
3427  * in it.
3428  */
3429 static int msi_alloc_irte(struct pci_dev *dev, int irq, int nvec)
3430 {
3431         struct intel_iommu *iommu;
3432         int index;
3433
3434         iommu = map_dev_to_ir(dev);
3435         if (!iommu) {
3436                 printk(KERN_ERR
3437                        "Unable to map PCI %s to iommu\n", pci_name(dev));
3438                 return -ENOENT;
3439         }
3440
3441         index = alloc_irte(iommu, irq, nvec);
3442         if (index < 0) {
3443                 printk(KERN_ERR
3444                        "Unable to allocate %d IRTE for PCI %s\n", nvec,
3445                        pci_name(dev));
3446                 return -ENOSPC;
3447         }
3448         return index;
3449 }
3450
3451 static int setup_msi_irq(struct pci_dev *dev, struct msi_desc *msidesc, int irq)
3452 {
3453         int ret;
3454         struct msi_msg msg;
3455
3456         ret = msi_compose_msg(dev, irq, &msg);
3457         if (ret < 0)
3458                 return ret;
3459
3460         set_irq_msi(irq, msidesc);
3461         write_msi_msg(irq, &msg);
3462
3463         if (irq_remapped(irq)) {
3464                 struct irq_desc *desc = irq_to_desc(irq);
3465                 /*
3466                  * irq migration in process context
3467                  */
3468                 desc->status |= IRQ_MOVE_PCNTXT;
3469                 set_irq_chip_and_handler_name(irq, &msi_ir_chip, handle_edge_irq, "edge");
3470         } else
3471                 set_irq_chip_and_handler_name(irq, &msi_chip, handle_edge_irq, "edge");
3472
3473         dev_printk(KERN_DEBUG, &dev->dev, "irq %d for MSI/MSI-X\n", irq);
3474
3475         return 0;
3476 }
3477
3478 int arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
3479 {
3480         unsigned int irq;
3481         int ret, sub_handle;
3482         struct msi_desc *msidesc;
3483         unsigned int irq_want;
3484         struct intel_iommu *iommu = NULL;
3485         int index = 0;
3486         int node;
3487
3488         /* x86 doesn't support multiple MSI yet */
3489         if (type == PCI_CAP_ID_MSI && nvec > 1)
3490                 return 1;
3491
3492         node = dev_to_node(&dev->dev);
3493         irq_want = nr_irqs_gsi;
3494         sub_handle = 0;
3495         list_for_each_entry(msidesc, &dev->msi_list, list) {
3496                 irq = create_irq_nr(irq_want, node);
3497                 if (irq == 0)
3498                         return -1;
3499                 irq_want = irq + 1;
3500                 if (!intr_remapping_enabled)
3501                         goto no_ir;
3502
3503                 if (!sub_handle) {
3504                         /*
3505                          * allocate the consecutive block of IRTE's
3506                          * for 'nvec'
3507                          */
3508                         index = msi_alloc_irte(dev, irq, nvec);
3509                         if (index < 0) {
3510                                 ret = index;
3511                                 goto error;
3512                         }
3513                 } else {
3514                         iommu = map_dev_to_ir(dev);
3515                         if (!iommu) {
3516                                 ret = -ENOENT;
3517                                 goto error;
3518                         }
3519                         /*
3520                          * setup the mapping between the irq and the IRTE
3521                          * base index, the sub_handle pointing to the
3522                          * appropriate interrupt remap table entry.
3523                          */
3524                         set_irte_irq(irq, iommu, index, sub_handle);
3525                 }
3526 no_ir:
3527                 ret = setup_msi_irq(dev, msidesc, irq);
3528                 if (ret < 0)
3529                         goto error;
3530                 sub_handle++;
3531         }
3532         return 0;
3533
3534 error:
3535         destroy_irq(irq);
3536         return ret;
3537 }
3538
3539 void arch_teardown_msi_irq(unsigned int irq)
3540 {
3541         destroy_irq(irq);
3542 }
3543
3544 #if defined (CONFIG_DMAR) || defined (CONFIG_INTR_REMAP)
3545 #ifdef CONFIG_SMP
3546 static int dmar_msi_set_affinity(unsigned int irq, const struct cpumask *mask)
3547 {
3548         struct irq_desc *desc = irq_to_desc(irq);
3549         struct irq_cfg *cfg;
3550         struct msi_msg msg;
3551         unsigned int dest;
3552
3553         dest = set_desc_affinity(desc, mask);
3554         if (dest == BAD_APICID)
3555                 return -1;
3556
3557         cfg = desc->chip_data;
3558
3559         dmar_msi_read(irq, &msg);
3560
3561         msg.data &= ~MSI_DATA_VECTOR_MASK;
3562         msg.data |= MSI_DATA_VECTOR(cfg->vector);
3563         msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3564         msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3565
3566         dmar_msi_write(irq, &msg);
3567
3568         return 0;
3569 }
3570
3571 #endif /* CONFIG_SMP */
3572
3573 static struct irq_chip dmar_msi_type = {
3574         .name = "DMAR_MSI",
3575         .unmask = dmar_msi_unmask,
3576         .mask = dmar_msi_mask,
3577         .ack = ack_apic_edge,
3578 #ifdef CONFIG_SMP
3579         .set_affinity = dmar_msi_set_affinity,
3580 #endif
3581         .retrigger = ioapic_retrigger_irq,
3582 };
3583
3584 int arch_setup_dmar_msi(unsigned int irq)
3585 {
3586         int ret;
3587         struct msi_msg msg;
3588
3589         ret = msi_compose_msg(NULL, irq, &msg);
3590         if (ret < 0)
3591                 return ret;
3592         dmar_msi_write(irq, &msg);
3593         set_irq_chip_and_handler_name(irq, &dmar_msi_type, handle_edge_irq,
3594                 "edge");
3595         return 0;
3596 }
3597 #endif
3598
3599 #ifdef CONFIG_HPET_TIMER
3600
3601 #ifdef CONFIG_SMP
3602 static int hpet_msi_set_affinity(unsigned int irq, const struct cpumask *mask)
3603 {
3604         struct irq_desc *desc = irq_to_desc(irq);
3605         struct irq_cfg *cfg;
3606         struct msi_msg msg;
3607         unsigned int dest;
3608
3609         dest = set_desc_affinity(desc, mask);
3610         if (dest == BAD_APICID)
3611                 return -1;
3612
3613         cfg = desc->chip_data;
3614
3615         hpet_msi_read(irq, &msg);
3616
3617         msg.data &= ~MSI_DATA_VECTOR_MASK;
3618         msg.data |= MSI_DATA_VECTOR(cfg->vector);
3619         msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3620         msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3621
3622         hpet_msi_write(irq, &msg);
3623
3624         return 0;
3625 }
3626
3627 #endif /* CONFIG_SMP */
3628
3629 static struct irq_chip hpet_msi_type = {
3630         .name = "HPET_MSI",
3631         .unmask = hpet_msi_unmask,
3632         .mask = hpet_msi_mask,
3633         .ack = ack_apic_edge,
3634 #ifdef CONFIG_SMP
3635         .set_affinity = hpet_msi_set_affinity,
3636 #endif
3637         .retrigger = ioapic_retrigger_irq,
3638 };
3639
3640 int arch_setup_hpet_msi(unsigned int irq)
3641 {
3642         int ret;
3643         struct msi_msg msg;
3644         struct irq_desc *desc = irq_to_desc(irq);
3645
3646         ret = msi_compose_msg(NULL, irq, &msg);
3647         if (ret < 0)
3648                 return ret;
3649
3650         hpet_msi_write(irq, &msg);
3651         desc->status |= IRQ_MOVE_PCNTXT;
3652         set_irq_chip_and_handler_name(irq, &hpet_msi_type, handle_edge_irq,
3653                 "edge");
3654
3655         return 0;
3656 }
3657 #endif
3658
3659 #endif /* CONFIG_PCI_MSI */
3660 /*
3661  * Hypertransport interrupt support
3662  */
3663 #ifdef CONFIG_HT_IRQ
3664
3665 #ifdef CONFIG_SMP
3666
3667 static void target_ht_irq(unsigned int irq, unsigned int dest, u8 vector)
3668 {
3669         struct ht_irq_msg msg;
3670         fetch_ht_irq_msg(irq, &msg);
3671
3672         msg.address_lo &= ~(HT_IRQ_LOW_VECTOR_MASK | HT_IRQ_LOW_DEST_ID_MASK);
3673         msg.address_hi &= ~(HT_IRQ_HIGH_DEST_ID_MASK);
3674
3675         msg.address_lo |= HT_IRQ_LOW_VECTOR(vector) | HT_IRQ_LOW_DEST_ID(dest);
3676         msg.address_hi |= HT_IRQ_HIGH_DEST_ID(dest);
3677
3678         write_ht_irq_msg(irq, &msg);
3679 }
3680
3681 static int set_ht_irq_affinity(unsigned int irq, const struct cpumask *mask)
3682 {
3683         struct irq_desc *desc = irq_to_desc(irq);
3684         struct irq_cfg *cfg;
3685         unsigned int dest;
3686
3687         dest = set_desc_affinity(desc, mask);
3688         if (dest == BAD_APICID)
3689                 return -1;
3690
3691         cfg = desc->chip_data;
3692
3693         target_ht_irq(irq, dest, cfg->vector);
3694
3695         return 0;
3696 }
3697
3698 #endif
3699
3700 static struct irq_chip ht_irq_chip = {
3701         .name           = "PCI-HT",
3702         .mask           = mask_ht_irq,
3703         .unmask         = unmask_ht_irq,
3704         .ack            = ack_apic_edge,
3705 #ifdef CONFIG_SMP
3706         .set_affinity   = set_ht_irq_affinity,
3707 #endif
3708         .retrigger      = ioapic_retrigger_irq,
3709 };
3710
3711 int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev)
3712 {
3713         struct irq_cfg *cfg;
3714         int err;
3715
3716         if (disable_apic)
3717                 return -ENXIO;
3718
3719         cfg = irq_cfg(irq);
3720         err = assign_irq_vector(irq, cfg, apic->target_cpus());
3721         if (!err) {
3722                 struct ht_irq_msg msg;
3723                 unsigned dest;
3724
3725                 dest = apic->cpu_mask_to_apicid_and(cfg->domain,
3726                                                     apic->target_cpus());
3727
3728                 msg.address_hi = HT_IRQ_HIGH_DEST_ID(dest);
3729
3730                 msg.address_lo =
3731                         HT_IRQ_LOW_BASE |
3732                         HT_IRQ_LOW_DEST_ID(dest) |
3733                         HT_IRQ_LOW_VECTOR(cfg->vector) |
3734                         ((apic->irq_dest_mode == 0) ?
3735                                 HT_IRQ_LOW_DM_PHYSICAL :
3736                                 HT_IRQ_LOW_DM_LOGICAL) |
3737                         HT_IRQ_LOW_RQEOI_EDGE |
3738                         ((apic->irq_delivery_mode != dest_LowestPrio) ?
3739                                 HT_IRQ_LOW_MT_FIXED :
3740                                 HT_IRQ_LOW_MT_ARBITRATED) |
3741                         HT_IRQ_LOW_IRQ_MASKED;
3742
3743                 write_ht_irq_msg(irq, &msg);
3744
3745                 set_irq_chip_and_handler_name(irq, &ht_irq_chip,
3746                                               handle_edge_irq, "edge");
3747
3748                 dev_printk(KERN_DEBUG, &dev->dev, "irq %d for HT\n", irq);
3749         }
3750         return err;
3751 }
3752 #endif /* CONFIG_HT_IRQ */
3753
3754 #ifdef CONFIG_X86_UV
3755 /*
3756  * Re-target the irq to the specified CPU and enable the specified MMR located
3757  * on the specified blade to allow the sending of MSIs to the specified CPU.
3758  */
3759 int arch_enable_uv_irq(char *irq_name, unsigned int irq, int cpu, int mmr_blade,
3760                        unsigned long mmr_offset)
3761 {
3762         const struct cpumask *eligible_cpu = cpumask_of(cpu);
3763         struct irq_cfg *cfg;
3764         int mmr_pnode;
3765         unsigned long mmr_value;
3766         struct uv_IO_APIC_route_entry *entry;
3767         unsigned long flags;
3768         int err;
3769
3770         BUILD_BUG_ON(sizeof(struct uv_IO_APIC_route_entry) != sizeof(unsigned long));
3771
3772         cfg = irq_cfg(irq);
3773
3774         err = assign_irq_vector(irq, cfg, eligible_cpu);
3775         if (err != 0)
3776                 return err;
3777
3778         spin_lock_irqsave(&vector_lock, flags);
3779         set_irq_chip_and_handler_name(irq, &uv_irq_chip, handle_percpu_irq,
3780                                       irq_name);
3781         spin_unlock_irqrestore(&vector_lock, flags);
3782
3783         mmr_value = 0;
3784         entry = (struct uv_IO_APIC_route_entry *)&mmr_value;
3785         entry->vector           = cfg->vector;
3786         entry->delivery_mode    = apic->irq_delivery_mode;
3787         entry->dest_mode        = apic->irq_dest_mode;
3788         entry->polarity         = 0;
3789         entry->trigger          = 0;
3790         entry->mask             = 0;
3791         entry->dest             = apic->cpu_mask_to_apicid(eligible_cpu);
3792
3793         mmr_pnode = uv_blade_to_pnode(mmr_blade);
3794         uv_write_global_mmr64(mmr_pnode, mmr_offset, mmr_value);
3795
3796         return irq;
3797 }
3798
3799 /*
3800  * Disable the specified MMR located on the specified blade so that MSIs are
3801  * longer allowed to be sent.
3802  */
3803 void arch_disable_uv_irq(int mmr_blade, unsigned long mmr_offset)
3804 {
3805         unsigned long mmr_value;
3806         struct uv_IO_APIC_route_entry *entry;
3807         int mmr_pnode;
3808
3809         BUILD_BUG_ON(sizeof(struct uv_IO_APIC_route_entry) != sizeof(unsigned long));
3810
3811         mmr_value = 0;
3812         entry = (struct uv_IO_APIC_route_entry *)&mmr_value;
3813         entry->mask = 1;
3814
3815         mmr_pnode = uv_blade_to_pnode(mmr_blade);
3816         uv_write_global_mmr64(mmr_pnode, mmr_offset, mmr_value);
3817 }
3818 #endif /* CONFIG_X86_64 */
3819
3820 int __init io_apic_get_redir_entries (int ioapic)
3821 {
3822         union IO_APIC_reg_01    reg_01;
3823         unsigned long flags;
3824
3825         spin_lock_irqsave(&ioapic_lock, flags);
3826         reg_01.raw = io_apic_read(ioapic, 1);
3827         spin_unlock_irqrestore(&ioapic_lock, flags);
3828
3829         return reg_01.bits.entries;
3830 }
3831
3832 void __init probe_nr_irqs_gsi(void)
3833 {
3834         int nr = 0;
3835
3836         nr = acpi_probe_gsi();
3837         if (nr > nr_irqs_gsi) {
3838                 nr_irqs_gsi = nr;
3839         } else {
3840                 /* for acpi=off or acpi is not compiled in */
3841                 int idx;
3842
3843                 nr = 0;
3844                 for (idx = 0; idx < nr_ioapics; idx++)
3845                         nr += io_apic_get_redir_entries(idx) + 1;
3846
3847                 if (nr > nr_irqs_gsi)
3848                         nr_irqs_gsi = nr;
3849         }
3850
3851         printk(KERN_DEBUG "nr_irqs_gsi: %d\n", nr_irqs_gsi);
3852 }
3853
3854 #ifdef CONFIG_SPARSE_IRQ
3855 int __init arch_probe_nr_irqs(void)
3856 {
3857         int nr;
3858
3859         if (nr_irqs > (NR_VECTORS * nr_cpu_ids))
3860                 nr_irqs = NR_VECTORS * nr_cpu_ids;
3861
3862         nr = nr_irqs_gsi + 8 * nr_cpu_ids;
3863 #if defined(CONFIG_PCI_MSI) || defined(CONFIG_HT_IRQ)
3864         /*
3865          * for MSI and HT dyn irq
3866          */
3867         nr += nr_irqs_gsi * 16;
3868 #endif
3869         if (nr < nr_irqs)
3870                 nr_irqs = nr;
3871
3872         return 0;
3873 }
3874 #endif
3875
3876 static int __io_apic_set_pci_routing(struct device *dev, int irq,
3877                                 struct io_apic_irq_attr *irq_attr)
3878 {
3879         struct irq_desc *desc;
3880         struct irq_cfg *cfg;
3881         int node;
3882         int ioapic, pin;
3883         int trigger, polarity;
3884
3885         ioapic = irq_attr->ioapic;
3886         if (!IO_APIC_IRQ(irq)) {
3887                 apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
3888                         ioapic);
3889                 return -EINVAL;
3890         }
3891
3892         if (dev)
3893                 node = dev_to_node(dev);
3894         else
3895                 node = cpu_to_node(boot_cpu_id);
3896
3897         desc = irq_to_desc_alloc_node(irq, node);
3898         if (!desc) {
3899                 printk(KERN_INFO "can not get irq_desc %d\n", irq);
3900                 return 0;
3901         }
3902
3903         pin = irq_attr->ioapic_pin;
3904         trigger = irq_attr->trigger;
3905         polarity = irq_attr->polarity;
3906
3907         /*
3908          * IRQs < 16 are already in the irq_2_pin[] map
3909          */
3910         if (irq >= NR_IRQS_LEGACY) {
3911                 cfg = desc->chip_data;
3912                 add_pin_to_irq_node(cfg, node, ioapic, pin);
3913         }
3914
3915         setup_IO_APIC_irq(ioapic, pin, irq, desc, trigger, polarity);
3916
3917         return 0;
3918 }
3919
3920 int io_apic_set_pci_routing(struct device *dev, int irq,
3921                                 struct io_apic_irq_attr *irq_attr)
3922 {
3923         int ioapic, pin;
3924         /*
3925          * Avoid pin reprogramming.  PRTs typically include entries
3926          * with redundant pin->gsi mappings (but unique PCI devices);
3927          * we only program the IOAPIC on the first.
3928          */
3929         ioapic = irq_attr->ioapic;
3930         pin = irq_attr->ioapic_pin;
3931         if (test_bit(pin, mp_ioapic_routing[ioapic].pin_programmed)) {
3932                 pr_debug("Pin %d-%d already programmed\n",
3933                          mp_ioapics[ioapic].apicid, pin);
3934                 return 0;
3935         }
3936         set_bit(pin, mp_ioapic_routing[ioapic].pin_programmed);
3937
3938         return __io_apic_set_pci_routing(dev, irq, irq_attr);
3939 }
3940
3941 /* --------------------------------------------------------------------------
3942                           ACPI-based IOAPIC Configuration
3943    -------------------------------------------------------------------------- */
3944
3945 #ifdef CONFIG_ACPI
3946
3947 #ifdef CONFIG_X86_32
3948 int __init io_apic_get_unique_id(int ioapic, int apic_id)
3949 {
3950         union IO_APIC_reg_00 reg_00;
3951         static physid_mask_t apic_id_map = PHYSID_MASK_NONE;
3952         physid_mask_t tmp;
3953         unsigned long flags;
3954         int i = 0;
3955
3956         /*
3957          * The P4 platform supports up to 256 APIC IDs on two separate APIC
3958          * buses (one for LAPICs, one for IOAPICs), where predecessors only
3959          * supports up to 16 on one shared APIC bus.
3960          *
3961          * TBD: Expand LAPIC/IOAPIC support on P4-class systems to take full
3962          *      advantage of new APIC bus architecture.
3963          */
3964
3965         if (physids_empty(apic_id_map))
3966                 apic_id_map = apic->ioapic_phys_id_map(phys_cpu_present_map);
3967
3968         spin_lock_irqsave(&ioapic_lock, flags);
3969         reg_00.raw = io_apic_read(ioapic, 0);
3970         spin_unlock_irqrestore(&ioapic_lock, flags);
3971
3972         if (apic_id >= get_physical_broadcast()) {
3973                 printk(KERN_WARNING "IOAPIC[%d]: Invalid apic_id %d, trying "
3974                         "%d\n", ioapic, apic_id, reg_00.bits.ID);
3975                 apic_id = reg_00.bits.ID;
3976         }
3977
3978         /*
3979          * Every APIC in a system must have a unique ID or we get lots of nice
3980          * 'stuck on smp_invalidate_needed IPI wait' messages.
3981          */
3982         if (apic->check_apicid_used(apic_id_map, apic_id)) {
3983
3984                 for (i = 0; i < get_physical_broadcast(); i++) {
3985                         if (!apic->check_apicid_used(apic_id_map, i))
3986                                 break;
3987                 }
3988
3989                 if (i == get_physical_broadcast())
3990                         panic("Max apic_id exceeded!\n");
3991
3992                 printk(KERN_WARNING "IOAPIC[%d]: apic_id %d already used, "
3993                         "trying %d\n", ioapic, apic_id, i);
3994
3995                 apic_id = i;
3996         }
3997
3998         tmp = apic->apicid_to_cpu_present(apic_id);
3999         physids_or(apic_id_map, apic_id_map, tmp);
4000
4001         if (reg_00.bits.ID != apic_id) {
4002                 reg_00.bits.ID = apic_id;
4003
4004                 spin_lock_irqsave(&ioapic_lock, flags);
4005                 io_apic_write(ioapic, 0, reg_00.raw);
4006                 reg_00.raw = io_apic_read(ioapic, 0);
4007                 spin_unlock_irqrestore(&ioapic_lock, flags);
4008
4009                 /* Sanity check */
4010                 if (reg_00.bits.ID != apic_id) {
4011                         printk("IOAPIC[%d]: Unable to change apic_id!\n", ioapic);
4012                         return -1;
4013                 }
4014         }
4015
4016         apic_printk(APIC_VERBOSE, KERN_INFO
4017                         "IOAPIC[%d]: Assigned apic_id %d\n", ioapic, apic_id);
4018
4019         return apic_id;
4020 }
4021 #endif
4022
4023 int __init io_apic_get_version(int ioapic)
4024 {
4025         union IO_APIC_reg_01    reg_01;
4026         unsigned long flags;
4027
4028         spin_lock_irqsave(&ioapic_lock, flags);
4029         reg_01.raw = io_apic_read(ioapic, 1);
4030         spin_unlock_irqrestore(&ioapic_lock, flags);
4031
4032         return reg_01.bits.version;
4033 }
4034
4035 int acpi_get_override_irq(int bus_irq, int *trigger, int *polarity)
4036 {
4037         int i;
4038
4039         if (skip_ioapic_setup)
4040                 return -1;
4041
4042         for (i = 0; i < mp_irq_entries; i++)
4043                 if (mp_irqs[i].irqtype == mp_INT &&
4044                     mp_irqs[i].srcbusirq == bus_irq)
4045                         break;
4046         if (i >= mp_irq_entries)
4047                 return -1;
4048
4049         *trigger = irq_trigger(i);
4050         *polarity = irq_polarity(i);
4051         return 0;
4052 }
4053
4054 #endif /* CONFIG_ACPI */
4055
4056 /*
4057  * This function currently is only a helper for the i386 smp boot process where
4058  * we need to reprogram the ioredtbls to cater for the cpus which have come online
4059  * so mask in all cases should simply be apic->target_cpus()
4060  */
4061 #ifdef CONFIG_SMP
4062 void __init setup_ioapic_dest(void)
4063 {
4064         int pin, ioapic = 0, irq, irq_entry;
4065         struct irq_desc *desc;
4066         const struct cpumask *mask;
4067
4068         if (skip_ioapic_setup == 1)
4069                 return;
4070
4071 #ifdef CONFIG_ACPI
4072         if (!acpi_disabled && acpi_ioapic) {
4073                 ioapic = mp_find_ioapic(0);
4074                 if (ioapic < 0)
4075                         ioapic = 0;
4076         }
4077 #endif
4078
4079         for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
4080                 irq_entry = find_irq_entry(ioapic, pin, mp_INT);
4081                 if (irq_entry == -1)
4082                         continue;
4083                 irq = pin_2_irq(irq_entry, ioapic, pin);
4084
4085                 desc = irq_to_desc(irq);
4086
4087                 /*
4088                  * Honour affinities which have been set in early boot
4089                  */
4090                 if (desc->status &
4091                     (IRQ_NO_BALANCING | IRQ_AFFINITY_SET))
4092                         mask = desc->affinity;
4093                 else
4094                         mask = apic->target_cpus();
4095
4096                 if (intr_remapping_enabled)
4097                         set_ir_ioapic_affinity_irq_desc(desc, mask);
4098                 else
4099                         set_ioapic_affinity_irq_desc(desc, mask);
4100         }
4101
4102 }
4103 #endif
4104
4105 #define IOAPIC_RESOURCE_NAME_SIZE 11
4106
4107 static struct resource *ioapic_resources;
4108
4109 static struct resource * __init ioapic_setup_resources(void)
4110 {
4111         unsigned long n;
4112         struct resource *res;
4113         char *mem;
4114         int i;
4115
4116         if (nr_ioapics <= 0)
4117                 return NULL;
4118
4119         n = IOAPIC_RESOURCE_NAME_SIZE + sizeof(struct resource);
4120         n *= nr_ioapics;
4121
4122         mem = alloc_bootmem(n);
4123         res = (void *)mem;
4124
4125         if (mem != NULL) {
4126                 mem += sizeof(struct resource) * nr_ioapics;
4127
4128                 for (i = 0; i < nr_ioapics; i++) {
4129                         res[i].name = mem;
4130                         res[i].flags = IORESOURCE_MEM | IORESOURCE_BUSY;
4131                         sprintf(mem,  "IOAPIC %u", i);
4132                         mem += IOAPIC_RESOURCE_NAME_SIZE;
4133                 }
4134         }
4135
4136         ioapic_resources = res;
4137
4138         return res;
4139 }
4140
4141 void __init ioapic_init_mappings(void)
4142 {
4143         unsigned long ioapic_phys, idx = FIX_IO_APIC_BASE_0;
4144         struct resource *ioapic_res;
4145         int i;
4146
4147         ioapic_res = ioapic_setup_resources();
4148         for (i = 0; i < nr_ioapics; i++) {
4149                 if (smp_found_config) {
4150                         ioapic_phys = mp_ioapics[i].apicaddr;
4151 #ifdef CONFIG_X86_32
4152                         if (!ioapic_phys) {
4153                                 printk(KERN_ERR
4154                                        "WARNING: bogus zero IO-APIC "
4155                                        "address found in MPTABLE, "
4156                                        "disabling IO/APIC support!\n");
4157                                 smp_found_config = 0;
4158                                 skip_ioapic_setup = 1;
4159                                 goto fake_ioapic_page;
4160                         }
4161 #endif
4162                 } else {
4163 #ifdef CONFIG_X86_32
4164 fake_ioapic_page:
4165 #endif
4166                         ioapic_phys = (unsigned long)
4167                                 alloc_bootmem_pages(PAGE_SIZE);
4168                         ioapic_phys = __pa(ioapic_phys);
4169                 }
4170                 set_fixmap_nocache(idx, ioapic_phys);
4171                 apic_printk(APIC_VERBOSE,
4172                             "mapped IOAPIC to %08lx (%08lx)\n",
4173                             __fix_to_virt(idx), ioapic_phys);
4174                 idx++;
4175
4176                 if (ioapic_res != NULL) {
4177                         ioapic_res->start = ioapic_phys;
4178                         ioapic_res->end = ioapic_phys + (4 * 1024) - 1;
4179                         ioapic_res++;
4180                 }
4181         }
4182 }
4183
4184 static int __init ioapic_insert_resources(void)
4185 {
4186         int i;
4187         struct resource *r = ioapic_resources;
4188
4189         if (!r) {
4190                 if (nr_ioapics > 0) {
4191                         printk(KERN_ERR
4192                                 "IO APIC resources couldn't be allocated.\n");
4193                         return -1;
4194                 }
4195                 return 0;
4196         }
4197
4198         for (i = 0; i < nr_ioapics; i++) {
4199                 insert_resource(&iomem_resource, r);
4200                 r++;
4201         }
4202
4203         return 0;
4204 }
4205
4206 /* Insert the IO APIC resources after PCI initialization has occured to handle
4207  * IO APICS that are mapped in on a BAR in PCI space. */
4208 late_initcall(ioapic_insert_resources);