Merge branch 'x86/cpufeature' into irq/numa
[sfrench/cifs-2.6.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/uv/uv_hub.h>
63 #include <asm/uv/uv_irq.h>
64
65 #include <asm/apic.h>
66
67 #define __apicdebuginit(type) static type __init
68
69 /*
70  *      Is the SiS APIC rmw bug present ?
71  *      -1 = don't know, 0 = no, 1 = yes
72  */
73 int sis_apic_bug = -1;
74
75 static DEFINE_SPINLOCK(ioapic_lock);
76 static DEFINE_SPINLOCK(vector_lock);
77
78 /*
79  * # of IRQ routing registers
80  */
81 int nr_ioapic_registers[MAX_IO_APICS];
82
83 /* I/O APIC entries */
84 struct mpc_ioapic mp_ioapics[MAX_IO_APICS];
85 int nr_ioapics;
86
87 /* MP IRQ source entries */
88 struct mpc_intsrc mp_irqs[MAX_IRQ_SOURCES];
89
90 /* # of MP IRQ source entries */
91 int mp_irq_entries;
92
93 #if defined (CONFIG_MCA) || defined (CONFIG_EISA)
94 int mp_bus_id_to_type[MAX_MP_BUSSES];
95 #endif
96
97 DECLARE_BITMAP(mp_bus_not_pci, MAX_MP_BUSSES);
98
99 int skip_ioapic_setup;
100
101 void arch_disable_smp_support(void)
102 {
103 #ifdef CONFIG_PCI
104         noioapicquirk = 1;
105         noioapicreroute = -1;
106 #endif
107         skip_ioapic_setup = 1;
108 }
109
110 static int __init parse_noapic(char *str)
111 {
112         /* disable IO-APIC */
113         arch_disable_smp_support();
114         return 0;
115 }
116 early_param("noapic", parse_noapic);
117
118 struct irq_pin_list;
119
120 /*
121  * This is performance-critical, we want to do it O(1)
122  *
123  * the indexing order of this array favors 1:1 mappings
124  * between pins and IRQs.
125  */
126
127 struct irq_pin_list {
128         int apic, pin;
129         struct irq_pin_list *next;
130 };
131
132 static struct irq_pin_list *get_one_free_irq_2_pin(int node)
133 {
134         struct irq_pin_list *pin;
135
136         pin = kzalloc_node(sizeof(*pin), GFP_ATOMIC, node);
137
138         return pin;
139 }
140
141 struct irq_cfg {
142         struct irq_pin_list *irq_2_pin;
143         cpumask_var_t domain;
144         cpumask_var_t old_domain;
145         unsigned move_cleanup_count;
146         u8 vector;
147         u8 move_in_progress : 1;
148 };
149
150 /* irq_cfg is indexed by the sum of all RTEs in all I/O APICs. */
151 #ifdef CONFIG_SPARSE_IRQ
152 static struct irq_cfg irq_cfgx[] = {
153 #else
154 static struct irq_cfg irq_cfgx[NR_IRQS] = {
155 #endif
156         [0]  = { .vector = IRQ0_VECTOR,  },
157         [1]  = { .vector = IRQ1_VECTOR,  },
158         [2]  = { .vector = IRQ2_VECTOR,  },
159         [3]  = { .vector = IRQ3_VECTOR,  },
160         [4]  = { .vector = IRQ4_VECTOR,  },
161         [5]  = { .vector = IRQ5_VECTOR,  },
162         [6]  = { .vector = IRQ6_VECTOR,  },
163         [7]  = { .vector = IRQ7_VECTOR,  },
164         [8]  = { .vector = IRQ8_VECTOR,  },
165         [9]  = { .vector = IRQ9_VECTOR,  },
166         [10] = { .vector = IRQ10_VECTOR, },
167         [11] = { .vector = IRQ11_VECTOR, },
168         [12] = { .vector = IRQ12_VECTOR, },
169         [13] = { .vector = IRQ13_VECTOR, },
170         [14] = { .vector = IRQ14_VECTOR, },
171         [15] = { .vector = IRQ15_VECTOR, },
172 };
173
174 int __init arch_early_irq_init(void)
175 {
176         struct irq_cfg *cfg;
177         struct irq_desc *desc;
178         int count;
179         int i;
180
181         cfg = irq_cfgx;
182         count = ARRAY_SIZE(irq_cfgx);
183
184         for (i = 0; i < count; i++) {
185                 desc = irq_to_desc(i);
186                 desc->chip_data = &cfg[i];
187                 alloc_bootmem_cpumask_var(&cfg[i].domain);
188                 alloc_bootmem_cpumask_var(&cfg[i].old_domain);
189                 if (i < NR_IRQS_LEGACY)
190                         cpumask_setall(cfg[i].domain);
191         }
192
193         return 0;
194 }
195
196 #ifdef CONFIG_SPARSE_IRQ
197 static struct irq_cfg *irq_cfg(unsigned int irq)
198 {
199         struct irq_cfg *cfg = NULL;
200         struct irq_desc *desc;
201
202         desc = irq_to_desc(irq);
203         if (desc)
204                 cfg = desc->chip_data;
205
206         return cfg;
207 }
208
209 static struct irq_cfg *get_one_free_irq_cfg(int node)
210 {
211         struct irq_cfg *cfg;
212
213         cfg = kzalloc_node(sizeof(*cfg), GFP_ATOMIC, node);
214         if (cfg) {
215                 if (!alloc_cpumask_var_node(&cfg->domain, GFP_ATOMIC, node)) {
216                         kfree(cfg);
217                         cfg = NULL;
218                 } else if (!alloc_cpumask_var_node(&cfg->old_domain,
219                                                           GFP_ATOMIC, node)) {
220                         free_cpumask_var(cfg->domain);
221                         kfree(cfg);
222                         cfg = NULL;
223                 } else {
224                         cpumask_clear(cfg->domain);
225                         cpumask_clear(cfg->old_domain);
226                 }
227         }
228
229         return cfg;
230 }
231
232 int arch_init_chip_data(struct irq_desc *desc, int node)
233 {
234         struct irq_cfg *cfg;
235
236         cfg = desc->chip_data;
237         if (!cfg) {
238                 desc->chip_data = get_one_free_irq_cfg(node);
239                 if (!desc->chip_data) {
240                         printk(KERN_ERR "can not alloc irq_cfg\n");
241                         BUG_ON(1);
242                 }
243         }
244
245         return 0;
246 }
247
248 /* for move_irq_desc */
249 static void
250 init_copy_irq_2_pin(struct irq_cfg *old_cfg, struct irq_cfg *cfg, int node)
251 {
252         struct irq_pin_list *old_entry, *head, *tail, *entry;
253
254         cfg->irq_2_pin = NULL;
255         old_entry = old_cfg->irq_2_pin;
256         if (!old_entry)
257                 return;
258
259         entry = get_one_free_irq_2_pin(node);
260         if (!entry)
261                 return;
262
263         entry->apic     = old_entry->apic;
264         entry->pin      = old_entry->pin;
265         head            = entry;
266         tail            = entry;
267         old_entry       = old_entry->next;
268         while (old_entry) {
269                 entry = get_one_free_irq_2_pin(node);
270                 if (!entry) {
271                         entry = head;
272                         while (entry) {
273                                 head = entry->next;
274                                 kfree(entry);
275                                 entry = head;
276                         }
277                         /* still use the old one */
278                         return;
279                 }
280                 entry->apic     = old_entry->apic;
281                 entry->pin      = old_entry->pin;
282                 tail->next      = entry;
283                 tail            = entry;
284                 old_entry       = old_entry->next;
285         }
286
287         tail->next = NULL;
288         cfg->irq_2_pin = head;
289 }
290
291 static void free_irq_2_pin(struct irq_cfg *old_cfg, struct irq_cfg *cfg)
292 {
293         struct irq_pin_list *entry, *next;
294
295         if (old_cfg->irq_2_pin == cfg->irq_2_pin)
296                 return;
297
298         entry = old_cfg->irq_2_pin;
299
300         while (entry) {
301                 next = entry->next;
302                 kfree(entry);
303                 entry = next;
304         }
305         old_cfg->irq_2_pin = NULL;
306 }
307
308 void arch_init_copy_chip_data(struct irq_desc *old_desc,
309                                  struct irq_desc *desc, int node)
310 {
311         struct irq_cfg *cfg;
312         struct irq_cfg *old_cfg;
313
314         cfg = get_one_free_irq_cfg(node);
315
316         if (!cfg)
317                 return;
318
319         desc->chip_data = cfg;
320
321         old_cfg = old_desc->chip_data;
322
323         memcpy(cfg, old_cfg, sizeof(struct irq_cfg));
324
325         init_copy_irq_2_pin(old_cfg, cfg, node);
326 }
327
328 static void free_irq_cfg(struct irq_cfg *old_cfg)
329 {
330         kfree(old_cfg);
331 }
332
333 void arch_free_chip_data(struct irq_desc *old_desc, struct irq_desc *desc)
334 {
335         struct irq_cfg *old_cfg, *cfg;
336
337         old_cfg = old_desc->chip_data;
338         cfg = desc->chip_data;
339
340         if (old_cfg == cfg)
341                 return;
342
343         if (old_cfg) {
344                 free_irq_2_pin(old_cfg, cfg);
345                 free_irq_cfg(old_cfg);
346                 old_desc->chip_data = NULL;
347         }
348 }
349 /* end for move_irq_desc */
350
351 #else
352 static struct irq_cfg *irq_cfg(unsigned int irq)
353 {
354         return irq < nr_irqs ? irq_cfgx + irq : NULL;
355 }
356
357 #endif
358
359 struct io_apic {
360         unsigned int index;
361         unsigned int unused[3];
362         unsigned int data;
363         unsigned int unused2[11];
364         unsigned int eoi;
365 };
366
367 static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
368 {
369         return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
370                 + (mp_ioapics[idx].apicaddr & ~PAGE_MASK);
371 }
372
373 static inline void io_apic_eoi(unsigned int apic, unsigned int vector)
374 {
375         struct io_apic __iomem *io_apic = io_apic_base(apic);
376         writel(vector, &io_apic->eoi);
377 }
378
379 static inline unsigned int io_apic_read(unsigned int apic, unsigned int reg)
380 {
381         struct io_apic __iomem *io_apic = io_apic_base(apic);
382         writel(reg, &io_apic->index);
383         return readl(&io_apic->data);
384 }
385
386 static inline void io_apic_write(unsigned int apic, unsigned int reg, unsigned int value)
387 {
388         struct io_apic __iomem *io_apic = io_apic_base(apic);
389         writel(reg, &io_apic->index);
390         writel(value, &io_apic->data);
391 }
392
393 /*
394  * Re-write a value: to be used for read-modify-write
395  * cycles where the read already set up the index register.
396  *
397  * Older SiS APIC requires we rewrite the index register
398  */
399 static inline void io_apic_modify(unsigned int apic, unsigned int reg, unsigned int value)
400 {
401         struct io_apic __iomem *io_apic = io_apic_base(apic);
402
403         if (sis_apic_bug)
404                 writel(reg, &io_apic->index);
405         writel(value, &io_apic->data);
406 }
407
408 static bool io_apic_level_ack_pending(struct irq_cfg *cfg)
409 {
410         struct irq_pin_list *entry;
411         unsigned long flags;
412
413         spin_lock_irqsave(&ioapic_lock, flags);
414         entry = cfg->irq_2_pin;
415         for (;;) {
416                 unsigned int reg;
417                 int pin;
418
419                 if (!entry)
420                         break;
421                 pin = entry->pin;
422                 reg = io_apic_read(entry->apic, 0x10 + pin*2);
423                 /* Is the remote IRR bit set? */
424                 if (reg & IO_APIC_REDIR_REMOTE_IRR) {
425                         spin_unlock_irqrestore(&ioapic_lock, flags);
426                         return true;
427                 }
428                 if (!entry->next)
429                         break;
430                 entry = entry->next;
431         }
432         spin_unlock_irqrestore(&ioapic_lock, flags);
433
434         return false;
435 }
436
437 union entry_union {
438         struct { u32 w1, w2; };
439         struct IO_APIC_route_entry entry;
440 };
441
442 static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
443 {
444         union entry_union eu;
445         unsigned long flags;
446         spin_lock_irqsave(&ioapic_lock, flags);
447         eu.w1 = io_apic_read(apic, 0x10 + 2 * pin);
448         eu.w2 = io_apic_read(apic, 0x11 + 2 * pin);
449         spin_unlock_irqrestore(&ioapic_lock, flags);
450         return eu.entry;
451 }
452
453 /*
454  * When we write a new IO APIC routing entry, we need to write the high
455  * word first! If the mask bit in the low word is clear, we will enable
456  * the interrupt, and we need to make sure the entry is fully populated
457  * before that happens.
458  */
459 static void
460 __ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
461 {
462         union entry_union eu;
463         eu.entry = e;
464         io_apic_write(apic, 0x11 + 2*pin, eu.w2);
465         io_apic_write(apic, 0x10 + 2*pin, eu.w1);
466 }
467
468 void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
469 {
470         unsigned long flags;
471         spin_lock_irqsave(&ioapic_lock, flags);
472         __ioapic_write_entry(apic, pin, e);
473         spin_unlock_irqrestore(&ioapic_lock, flags);
474 }
475
476 /*
477  * When we mask an IO APIC routing entry, we need to write the low
478  * word first, in order to set the mask bit before we change the
479  * high bits!
480  */
481 static void ioapic_mask_entry(int apic, int pin)
482 {
483         unsigned long flags;
484         union entry_union eu = { .entry.mask = 1 };
485
486         spin_lock_irqsave(&ioapic_lock, flags);
487         io_apic_write(apic, 0x10 + 2*pin, eu.w1);
488         io_apic_write(apic, 0x11 + 2*pin, eu.w2);
489         spin_unlock_irqrestore(&ioapic_lock, flags);
490 }
491
492 /*
493  * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
494  * shared ISA-space IRQs, so we have to support them. We are super
495  * fast in the common case, and fast for shared ISA-space IRQs.
496  */
497 static void add_pin_to_irq_node(struct irq_cfg *cfg, int node, int apic, int pin)
498 {
499         struct irq_pin_list *entry;
500
501         entry = cfg->irq_2_pin;
502         if (!entry) {
503                 entry = get_one_free_irq_2_pin(node);
504                 if (!entry) {
505                         printk(KERN_ERR "can not alloc irq_2_pin to add %d - %d\n",
506                                         apic, pin);
507                         return;
508                 }
509                 cfg->irq_2_pin = entry;
510                 entry->apic = apic;
511                 entry->pin = pin;
512                 return;
513         }
514
515         while (entry->next) {
516                 /* not again, please */
517                 if (entry->apic == apic && entry->pin == pin)
518                         return;
519
520                 entry = entry->next;
521         }
522
523         entry->next = get_one_free_irq_2_pin(node);
524         entry = entry->next;
525         entry->apic = apic;
526         entry->pin = pin;
527 }
528
529 /*
530  * Reroute an IRQ to a different pin.
531  */
532 static void __init replace_pin_at_irq_node(struct irq_cfg *cfg, int node,
533                                       int oldapic, int oldpin,
534                                       int newapic, int newpin)
535 {
536         struct irq_pin_list *entry = cfg->irq_2_pin;
537         int replaced = 0;
538
539         while (entry) {
540                 if (entry->apic == oldapic && entry->pin == oldpin) {
541                         entry->apic = newapic;
542                         entry->pin = newpin;
543                         replaced = 1;
544                         /* every one is different, right? */
545                         break;
546                 }
547                 entry = entry->next;
548         }
549
550         /* why? call replace before add? */
551         if (!replaced)
552                 add_pin_to_irq_node(cfg, node, newapic, newpin);
553 }
554
555 static inline void io_apic_modify_irq(struct irq_cfg *cfg,
556                                 int mask_and, int mask_or,
557                                 void (*final)(struct irq_pin_list *entry))
558 {
559         int pin;
560         struct irq_pin_list *entry;
561
562         for (entry = cfg->irq_2_pin; entry != NULL; entry = entry->next) {
563                 unsigned int reg;
564                 pin = entry->pin;
565                 reg = io_apic_read(entry->apic, 0x10 + pin * 2);
566                 reg &= mask_and;
567                 reg |= mask_or;
568                 io_apic_modify(entry->apic, 0x10 + pin * 2, reg);
569                 if (final)
570                         final(entry);
571         }
572 }
573
574 static void __unmask_IO_APIC_irq(struct irq_cfg *cfg)
575 {
576         io_apic_modify_irq(cfg, ~IO_APIC_REDIR_MASKED, 0, NULL);
577 }
578
579 #ifdef CONFIG_X86_64
580 static void io_apic_sync(struct irq_pin_list *entry)
581 {
582         /*
583          * Synchronize the IO-APIC and the CPU by doing
584          * a dummy read from the IO-APIC
585          */
586         struct io_apic __iomem *io_apic;
587         io_apic = io_apic_base(entry->apic);
588         readl(&io_apic->data);
589 }
590
591 static void __mask_IO_APIC_irq(struct irq_cfg *cfg)
592 {
593         io_apic_modify_irq(cfg, ~0, IO_APIC_REDIR_MASKED, &io_apic_sync);
594 }
595 #else /* CONFIG_X86_32 */
596 static void __mask_IO_APIC_irq(struct irq_cfg *cfg)
597 {
598         io_apic_modify_irq(cfg, ~0, IO_APIC_REDIR_MASKED, NULL);
599 }
600
601 static void __mask_and_edge_IO_APIC_irq(struct irq_cfg *cfg)
602 {
603         io_apic_modify_irq(cfg, ~IO_APIC_REDIR_LEVEL_TRIGGER,
604                         IO_APIC_REDIR_MASKED, NULL);
605 }
606
607 static void __unmask_and_level_IO_APIC_irq(struct irq_cfg *cfg)
608 {
609         io_apic_modify_irq(cfg, ~IO_APIC_REDIR_MASKED,
610                         IO_APIC_REDIR_LEVEL_TRIGGER, NULL);
611 }
612 #endif /* CONFIG_X86_32 */
613
614 static void mask_IO_APIC_irq_desc(struct irq_desc *desc)
615 {
616         struct irq_cfg *cfg = desc->chip_data;
617         unsigned long flags;
618
619         BUG_ON(!cfg);
620
621         spin_lock_irqsave(&ioapic_lock, flags);
622         __mask_IO_APIC_irq(cfg);
623         spin_unlock_irqrestore(&ioapic_lock, flags);
624 }
625
626 static void unmask_IO_APIC_irq_desc(struct irq_desc *desc)
627 {
628         struct irq_cfg *cfg = desc->chip_data;
629         unsigned long flags;
630
631         spin_lock_irqsave(&ioapic_lock, flags);
632         __unmask_IO_APIC_irq(cfg);
633         spin_unlock_irqrestore(&ioapic_lock, flags);
634 }
635
636 static void mask_IO_APIC_irq(unsigned int irq)
637 {
638         struct irq_desc *desc = irq_to_desc(irq);
639
640         mask_IO_APIC_irq_desc(desc);
641 }
642 static void unmask_IO_APIC_irq(unsigned int irq)
643 {
644         struct irq_desc *desc = irq_to_desc(irq);
645
646         unmask_IO_APIC_irq_desc(desc);
647 }
648
649 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
650 {
651         struct IO_APIC_route_entry entry;
652
653         /* Check delivery_mode to be sure we're not clearing an SMI pin */
654         entry = ioapic_read_entry(apic, pin);
655         if (entry.delivery_mode == dest_SMI)
656                 return;
657         /*
658          * Disable it in the IO-APIC irq-routing table:
659          */
660         ioapic_mask_entry(apic, pin);
661 }
662
663 static void clear_IO_APIC (void)
664 {
665         int apic, pin;
666
667         for (apic = 0; apic < nr_ioapics; apic++)
668                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
669                         clear_IO_APIC_pin(apic, pin);
670 }
671
672 #ifdef CONFIG_X86_32
673 /*
674  * support for broken MP BIOSs, enables hand-redirection of PIRQ0-7 to
675  * specific CPU-side IRQs.
676  */
677
678 #define MAX_PIRQS 8
679 static int pirq_entries[MAX_PIRQS] = {
680         [0 ... MAX_PIRQS - 1] = -1
681 };
682
683 static int __init ioapic_pirq_setup(char *str)
684 {
685         int i, max;
686         int ints[MAX_PIRQS+1];
687
688         get_options(str, ARRAY_SIZE(ints), ints);
689
690         apic_printk(APIC_VERBOSE, KERN_INFO
691                         "PIRQ redirection, working around broken MP-BIOS.\n");
692         max = MAX_PIRQS;
693         if (ints[0] < MAX_PIRQS)
694                 max = ints[0];
695
696         for (i = 0; i < max; i++) {
697                 apic_printk(APIC_VERBOSE, KERN_DEBUG
698                                 "... PIRQ%d -> IRQ %d\n", i, ints[i+1]);
699                 /*
700                  * PIRQs are mapped upside down, usually.
701                  */
702                 pirq_entries[MAX_PIRQS-i-1] = ints[i+1];
703         }
704         return 1;
705 }
706
707 __setup("pirq=", ioapic_pirq_setup);
708 #endif /* CONFIG_X86_32 */
709
710 struct IO_APIC_route_entry **alloc_ioapic_entries(void)
711 {
712         int apic;
713         struct IO_APIC_route_entry **ioapic_entries;
714
715         ioapic_entries = kzalloc(sizeof(*ioapic_entries) * nr_ioapics,
716                                 GFP_ATOMIC);
717         if (!ioapic_entries)
718                 return 0;
719
720         for (apic = 0; apic < nr_ioapics; apic++) {
721                 ioapic_entries[apic] =
722                         kzalloc(sizeof(struct IO_APIC_route_entry) *
723                                 nr_ioapic_registers[apic], GFP_ATOMIC);
724                 if (!ioapic_entries[apic])
725                         goto nomem;
726         }
727
728         return ioapic_entries;
729
730 nomem:
731         while (--apic >= 0)
732                 kfree(ioapic_entries[apic]);
733         kfree(ioapic_entries);
734
735         return 0;
736 }
737
738 /*
739  * Saves all the IO-APIC RTE's
740  */
741 int save_IO_APIC_setup(struct IO_APIC_route_entry **ioapic_entries)
742 {
743         int apic, pin;
744
745         if (!ioapic_entries)
746                 return -ENOMEM;
747
748         for (apic = 0; apic < nr_ioapics; apic++) {
749                 if (!ioapic_entries[apic])
750                         return -ENOMEM;
751
752                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
753                         ioapic_entries[apic][pin] =
754                                 ioapic_read_entry(apic, pin);
755         }
756
757         return 0;
758 }
759
760 /*
761  * Mask all IO APIC entries.
762  */
763 void mask_IO_APIC_setup(struct IO_APIC_route_entry **ioapic_entries)
764 {
765         int apic, pin;
766
767         if (!ioapic_entries)
768                 return;
769
770         for (apic = 0; apic < nr_ioapics; apic++) {
771                 if (!ioapic_entries[apic])
772                         break;
773
774                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
775                         struct IO_APIC_route_entry entry;
776
777                         entry = ioapic_entries[apic][pin];
778                         if (!entry.mask) {
779                                 entry.mask = 1;
780                                 ioapic_write_entry(apic, pin, entry);
781                         }
782                 }
783         }
784 }
785
786 /*
787  * Restore IO APIC entries which was saved in ioapic_entries.
788  */
789 int restore_IO_APIC_setup(struct IO_APIC_route_entry **ioapic_entries)
790 {
791         int apic, pin;
792
793         if (!ioapic_entries)
794                 return -ENOMEM;
795
796         for (apic = 0; apic < nr_ioapics; apic++) {
797                 if (!ioapic_entries[apic])
798                         return -ENOMEM;
799
800                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
801                         ioapic_write_entry(apic, pin,
802                                         ioapic_entries[apic][pin]);
803         }
804         return 0;
805 }
806
807 void free_ioapic_entries(struct IO_APIC_route_entry **ioapic_entries)
808 {
809         int apic;
810
811         for (apic = 0; apic < nr_ioapics; apic++)
812                 kfree(ioapic_entries[apic]);
813
814         kfree(ioapic_entries);
815 }
816
817 /*
818  * Find the IRQ entry number of a certain pin.
819  */
820 static int find_irq_entry(int apic, int pin, int type)
821 {
822         int i;
823
824         for (i = 0; i < mp_irq_entries; i++)
825                 if (mp_irqs[i].irqtype == type &&
826                     (mp_irqs[i].dstapic == mp_ioapics[apic].apicid ||
827                      mp_irqs[i].dstapic == MP_APIC_ALL) &&
828                     mp_irqs[i].dstirq == pin)
829                         return i;
830
831         return -1;
832 }
833
834 /*
835  * Find the pin to which IRQ[irq] (ISA) is connected
836  */
837 static int __init find_isa_irq_pin(int irq, int type)
838 {
839         int i;
840
841         for (i = 0; i < mp_irq_entries; i++) {
842                 int lbus = mp_irqs[i].srcbus;
843
844                 if (test_bit(lbus, mp_bus_not_pci) &&
845                     (mp_irqs[i].irqtype == type) &&
846                     (mp_irqs[i].srcbusirq == irq))
847
848                         return mp_irqs[i].dstirq;
849         }
850         return -1;
851 }
852
853 static int __init find_isa_irq_apic(int irq, int type)
854 {
855         int i;
856
857         for (i = 0; i < mp_irq_entries; i++) {
858                 int lbus = mp_irqs[i].srcbus;
859
860                 if (test_bit(lbus, mp_bus_not_pci) &&
861                     (mp_irqs[i].irqtype == type) &&
862                     (mp_irqs[i].srcbusirq == irq))
863                         break;
864         }
865         if (i < mp_irq_entries) {
866                 int apic;
867                 for(apic = 0; apic < nr_ioapics; apic++) {
868                         if (mp_ioapics[apic].apicid == mp_irqs[i].dstapic)
869                                 return apic;
870                 }
871         }
872
873         return -1;
874 }
875
876 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
877 /*
878  * EISA Edge/Level control register, ELCR
879  */
880 static int EISA_ELCR(unsigned int irq)
881 {
882         if (irq < NR_IRQS_LEGACY) {
883                 unsigned int port = 0x4d0 + (irq >> 3);
884                 return (inb(port) >> (irq & 7)) & 1;
885         }
886         apic_printk(APIC_VERBOSE, KERN_INFO
887                         "Broken MPtable reports ISA irq %d\n", irq);
888         return 0;
889 }
890
891 #endif
892
893 /* ISA interrupts are always polarity zero edge triggered,
894  * when listed as conforming in the MP table. */
895
896 #define default_ISA_trigger(idx)        (0)
897 #define default_ISA_polarity(idx)       (0)
898
899 /* EISA interrupts are always polarity zero and can be edge or level
900  * trigger depending on the ELCR value.  If an interrupt is listed as
901  * EISA conforming in the MP table, that means its trigger type must
902  * be read in from the ELCR */
903
904 #define default_EISA_trigger(idx)       (EISA_ELCR(mp_irqs[idx].srcbusirq))
905 #define default_EISA_polarity(idx)      default_ISA_polarity(idx)
906
907 /* PCI interrupts are always polarity one level triggered,
908  * when listed as conforming in the MP table. */
909
910 #define default_PCI_trigger(idx)        (1)
911 #define default_PCI_polarity(idx)       (1)
912
913 /* MCA interrupts are always polarity zero level triggered,
914  * when listed as conforming in the MP table. */
915
916 #define default_MCA_trigger(idx)        (1)
917 #define default_MCA_polarity(idx)       default_ISA_polarity(idx)
918
919 static int MPBIOS_polarity(int idx)
920 {
921         int bus = mp_irqs[idx].srcbus;
922         int polarity;
923
924         /*
925          * Determine IRQ line polarity (high active or low active):
926          */
927         switch (mp_irqs[idx].irqflag & 3)
928         {
929                 case 0: /* conforms, ie. bus-type dependent polarity */
930                         if (test_bit(bus, mp_bus_not_pci))
931                                 polarity = default_ISA_polarity(idx);
932                         else
933                                 polarity = default_PCI_polarity(idx);
934                         break;
935                 case 1: /* high active */
936                 {
937                         polarity = 0;
938                         break;
939                 }
940                 case 2: /* reserved */
941                 {
942                         printk(KERN_WARNING "broken BIOS!!\n");
943                         polarity = 1;
944                         break;
945                 }
946                 case 3: /* low active */
947                 {
948                         polarity = 1;
949                         break;
950                 }
951                 default: /* invalid */
952                 {
953                         printk(KERN_WARNING "broken BIOS!!\n");
954                         polarity = 1;
955                         break;
956                 }
957         }
958         return polarity;
959 }
960
961 static int MPBIOS_trigger(int idx)
962 {
963         int bus = mp_irqs[idx].srcbus;
964         int trigger;
965
966         /*
967          * Determine IRQ trigger mode (edge or level sensitive):
968          */
969         switch ((mp_irqs[idx].irqflag>>2) & 3)
970         {
971                 case 0: /* conforms, ie. bus-type dependent */
972                         if (test_bit(bus, mp_bus_not_pci))
973                                 trigger = default_ISA_trigger(idx);
974                         else
975                                 trigger = default_PCI_trigger(idx);
976 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
977                         switch (mp_bus_id_to_type[bus]) {
978                                 case MP_BUS_ISA: /* ISA pin */
979                                 {
980                                         /* set before the switch */
981                                         break;
982                                 }
983                                 case MP_BUS_EISA: /* EISA pin */
984                                 {
985                                         trigger = default_EISA_trigger(idx);
986                                         break;
987                                 }
988                                 case MP_BUS_PCI: /* PCI pin */
989                                 {
990                                         /* set before the switch */
991                                         break;
992                                 }
993                                 case MP_BUS_MCA: /* MCA pin */
994                                 {
995                                         trigger = default_MCA_trigger(idx);
996                                         break;
997                                 }
998                                 default:
999                                 {
1000                                         printk(KERN_WARNING "broken BIOS!!\n");
1001                                         trigger = 1;
1002                                         break;
1003                                 }
1004                         }
1005 #endif
1006                         break;
1007                 case 1: /* edge */
1008                 {
1009                         trigger = 0;
1010                         break;
1011                 }
1012                 case 2: /* reserved */
1013                 {
1014                         printk(KERN_WARNING "broken BIOS!!\n");
1015                         trigger = 1;
1016                         break;
1017                 }
1018                 case 3: /* level */
1019                 {
1020                         trigger = 1;
1021                         break;
1022                 }
1023                 default: /* invalid */
1024                 {
1025                         printk(KERN_WARNING "broken BIOS!!\n");
1026                         trigger = 0;
1027                         break;
1028                 }
1029         }
1030         return trigger;
1031 }
1032
1033 static inline int irq_polarity(int idx)
1034 {
1035         return MPBIOS_polarity(idx);
1036 }
1037
1038 static inline int irq_trigger(int idx)
1039 {
1040         return MPBIOS_trigger(idx);
1041 }
1042
1043 int (*ioapic_renumber_irq)(int ioapic, int irq);
1044 static int pin_2_irq(int idx, int apic, int pin)
1045 {
1046         int irq, i;
1047         int bus = mp_irqs[idx].srcbus;
1048
1049         /*
1050          * Debugging check, we are in big trouble if this message pops up!
1051          */
1052         if (mp_irqs[idx].dstirq != pin)
1053                 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
1054
1055         if (test_bit(bus, mp_bus_not_pci)) {
1056                 irq = mp_irqs[idx].srcbusirq;
1057         } else {
1058                 /*
1059                  * PCI IRQs are mapped in order
1060                  */
1061                 i = irq = 0;
1062                 while (i < apic)
1063                         irq += nr_ioapic_registers[i++];
1064                 irq += pin;
1065                 /*
1066                  * For MPS mode, so far only needed by ES7000 platform
1067                  */
1068                 if (ioapic_renumber_irq)
1069                         irq = ioapic_renumber_irq(apic, irq);
1070         }
1071
1072 #ifdef CONFIG_X86_32
1073         /*
1074          * PCI IRQ command line redirection. Yes, limits are hardcoded.
1075          */
1076         if ((pin >= 16) && (pin <= 23)) {
1077                 if (pirq_entries[pin-16] != -1) {
1078                         if (!pirq_entries[pin-16]) {
1079                                 apic_printk(APIC_VERBOSE, KERN_DEBUG
1080                                                 "disabling PIRQ%d\n", pin-16);
1081                         } else {
1082                                 irq = pirq_entries[pin-16];
1083                                 apic_printk(APIC_VERBOSE, KERN_DEBUG
1084                                                 "using PIRQ%d -> IRQ %d\n",
1085                                                 pin-16, irq);
1086                         }
1087                 }
1088         }
1089 #endif
1090
1091         return irq;
1092 }
1093
1094 /*
1095  * Find a specific PCI IRQ entry.
1096  * Not an __init, possibly needed by modules
1097  */
1098 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin,
1099                                 struct io_apic_irq_attr *irq_attr)
1100 {
1101         int apic, i, best_guess = -1;
1102
1103         apic_printk(APIC_DEBUG,
1104                     "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
1105                     bus, slot, pin);
1106         if (test_bit(bus, mp_bus_not_pci)) {
1107                 apic_printk(APIC_VERBOSE,
1108                             "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
1109                 return -1;
1110         }
1111         for (i = 0; i < mp_irq_entries; i++) {
1112                 int lbus = mp_irqs[i].srcbus;
1113
1114                 for (apic = 0; apic < nr_ioapics; apic++)
1115                         if (mp_ioapics[apic].apicid == mp_irqs[i].dstapic ||
1116                             mp_irqs[i].dstapic == MP_APIC_ALL)
1117                                 break;
1118
1119                 if (!test_bit(lbus, mp_bus_not_pci) &&
1120                     !mp_irqs[i].irqtype &&
1121                     (bus == lbus) &&
1122                     (slot == ((mp_irqs[i].srcbusirq >> 2) & 0x1f))) {
1123                         int irq = pin_2_irq(i, apic, mp_irqs[i].dstirq);
1124
1125                         if (!(apic || IO_APIC_IRQ(irq)))
1126                                 continue;
1127
1128                         if (pin == (mp_irqs[i].srcbusirq & 3)) {
1129                                 set_io_apic_irq_attr(irq_attr, apic,
1130                                                      mp_irqs[i].dstirq,
1131                                                      irq_trigger(i),
1132                                                      irq_polarity(i));
1133                                 return irq;
1134                         }
1135                         /*
1136                          * Use the first all-but-pin matching entry as a
1137                          * best-guess fuzzy result for broken mptables.
1138                          */
1139                         if (best_guess < 0) {
1140                                 set_io_apic_irq_attr(irq_attr, apic,
1141                                                      mp_irqs[i].dstirq,
1142                                                      irq_trigger(i),
1143                                                      irq_polarity(i));
1144                                 best_guess = irq;
1145                         }
1146                 }
1147         }
1148         return best_guess;
1149 }
1150 EXPORT_SYMBOL(IO_APIC_get_PCI_irq_vector);
1151
1152 void lock_vector_lock(void)
1153 {
1154         /* Used to the online set of cpus does not change
1155          * during assign_irq_vector.
1156          */
1157         spin_lock(&vector_lock);
1158 }
1159
1160 void unlock_vector_lock(void)
1161 {
1162         spin_unlock(&vector_lock);
1163 }
1164
1165 static int
1166 __assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask)
1167 {
1168         /*
1169          * NOTE! The local APIC isn't very good at handling
1170          * multiple interrupts at the same interrupt level.
1171          * As the interrupt level is determined by taking the
1172          * vector number and shifting that right by 4, we
1173          * want to spread these out a bit so that they don't
1174          * all fall in the same interrupt level.
1175          *
1176          * Also, we've got to be careful not to trash gate
1177          * 0x80, because int 0x80 is hm, kind of importantish. ;)
1178          */
1179         static int current_vector = FIRST_DEVICE_VECTOR, current_offset = 0;
1180         unsigned int old_vector;
1181         int cpu, err;
1182         cpumask_var_t tmp_mask;
1183
1184         if ((cfg->move_in_progress) || cfg->move_cleanup_count)
1185                 return -EBUSY;
1186
1187         if (!alloc_cpumask_var(&tmp_mask, GFP_ATOMIC))
1188                 return -ENOMEM;
1189
1190         old_vector = cfg->vector;
1191         if (old_vector) {
1192                 cpumask_and(tmp_mask, mask, cpu_online_mask);
1193                 cpumask_and(tmp_mask, cfg->domain, tmp_mask);
1194                 if (!cpumask_empty(tmp_mask)) {
1195                         free_cpumask_var(tmp_mask);
1196                         return 0;
1197                 }
1198         }
1199
1200         /* Only try and allocate irqs on cpus that are present */
1201         err = -ENOSPC;
1202         for_each_cpu_and(cpu, mask, cpu_online_mask) {
1203                 int new_cpu;
1204                 int vector, offset;
1205
1206                 apic->vector_allocation_domain(cpu, tmp_mask);
1207
1208                 vector = current_vector;
1209                 offset = current_offset;
1210 next:
1211                 vector += 8;
1212                 if (vector >= first_system_vector) {
1213                         /* If out of vectors on large boxen, must share them. */
1214                         offset = (offset + 1) % 8;
1215                         vector = FIRST_DEVICE_VECTOR + offset;
1216                 }
1217                 if (unlikely(current_vector == vector))
1218                         continue;
1219
1220                 if (test_bit(vector, used_vectors))
1221                         goto next;
1222
1223                 for_each_cpu_and(new_cpu, tmp_mask, cpu_online_mask)
1224                         if (per_cpu(vector_irq, new_cpu)[vector] != -1)
1225                                 goto next;
1226                 /* Found one! */
1227                 current_vector = vector;
1228                 current_offset = offset;
1229                 if (old_vector) {
1230                         cfg->move_in_progress = 1;
1231                         cpumask_copy(cfg->old_domain, cfg->domain);
1232                 }
1233                 for_each_cpu_and(new_cpu, tmp_mask, cpu_online_mask)
1234                         per_cpu(vector_irq, new_cpu)[vector] = irq;
1235                 cfg->vector = vector;
1236                 cpumask_copy(cfg->domain, tmp_mask);
1237                 err = 0;
1238                 break;
1239         }
1240         free_cpumask_var(tmp_mask);
1241         return err;
1242 }
1243
1244 static int
1245 assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask)
1246 {
1247         int err;
1248         unsigned long flags;
1249
1250         spin_lock_irqsave(&vector_lock, flags);
1251         err = __assign_irq_vector(irq, cfg, mask);
1252         spin_unlock_irqrestore(&vector_lock, flags);
1253         return err;
1254 }
1255
1256 static void __clear_irq_vector(int irq, struct irq_cfg *cfg)
1257 {
1258         int cpu, vector;
1259
1260         BUG_ON(!cfg->vector);
1261
1262         vector = cfg->vector;
1263         for_each_cpu_and(cpu, cfg->domain, cpu_online_mask)
1264                 per_cpu(vector_irq, cpu)[vector] = -1;
1265
1266         cfg->vector = 0;
1267         cpumask_clear(cfg->domain);
1268
1269         if (likely(!cfg->move_in_progress))
1270                 return;
1271         for_each_cpu_and(cpu, cfg->old_domain, cpu_online_mask) {
1272                 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS;
1273                                                                 vector++) {
1274                         if (per_cpu(vector_irq, cpu)[vector] != irq)
1275                                 continue;
1276                         per_cpu(vector_irq, cpu)[vector] = -1;
1277                         break;
1278                 }
1279         }
1280         cfg->move_in_progress = 0;
1281 }
1282
1283 void __setup_vector_irq(int cpu)
1284 {
1285         /* Initialize vector_irq on a new cpu */
1286         /* This function must be called with vector_lock held */
1287         int irq, vector;
1288         struct irq_cfg *cfg;
1289         struct irq_desc *desc;
1290
1291         /* Mark the inuse vectors */
1292         for_each_irq_desc(irq, desc) {
1293                 cfg = desc->chip_data;
1294                 if (!cpumask_test_cpu(cpu, cfg->domain))
1295                         continue;
1296                 vector = cfg->vector;
1297                 per_cpu(vector_irq, cpu)[vector] = irq;
1298         }
1299         /* Mark the free vectors */
1300         for (vector = 0; vector < NR_VECTORS; ++vector) {
1301                 irq = per_cpu(vector_irq, cpu)[vector];
1302                 if (irq < 0)
1303                         continue;
1304
1305                 cfg = irq_cfg(irq);
1306                 if (!cpumask_test_cpu(cpu, cfg->domain))
1307                         per_cpu(vector_irq, cpu)[vector] = -1;
1308         }
1309 }
1310
1311 static struct irq_chip ioapic_chip;
1312 static struct irq_chip ir_ioapic_chip;
1313
1314 #define IOAPIC_AUTO     -1
1315 #define IOAPIC_EDGE     0
1316 #define IOAPIC_LEVEL    1
1317
1318 #ifdef CONFIG_X86_32
1319 static inline int IO_APIC_irq_trigger(int irq)
1320 {
1321         int apic, idx, pin;
1322
1323         for (apic = 0; apic < nr_ioapics; apic++) {
1324                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1325                         idx = find_irq_entry(apic, pin, mp_INT);
1326                         if ((idx != -1) && (irq == pin_2_irq(idx, apic, pin)))
1327                                 return irq_trigger(idx);
1328                 }
1329         }
1330         /*
1331          * nonexistent IRQs are edge default
1332          */
1333         return 0;
1334 }
1335 #else
1336 static inline int IO_APIC_irq_trigger(int irq)
1337 {
1338         return 1;
1339 }
1340 #endif
1341
1342 static void ioapic_register_intr(int irq, struct irq_desc *desc, unsigned long trigger)
1343 {
1344
1345         if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1346             trigger == IOAPIC_LEVEL)
1347                 desc->status |= IRQ_LEVEL;
1348         else
1349                 desc->status &= ~IRQ_LEVEL;
1350
1351         if (irq_remapped(irq)) {
1352                 desc->status |= IRQ_MOVE_PCNTXT;
1353                 if (trigger)
1354                         set_irq_chip_and_handler_name(irq, &ir_ioapic_chip,
1355                                                       handle_fasteoi_irq,
1356                                                      "fasteoi");
1357                 else
1358                         set_irq_chip_and_handler_name(irq, &ir_ioapic_chip,
1359                                                       handle_edge_irq, "edge");
1360                 return;
1361         }
1362
1363         if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1364             trigger == IOAPIC_LEVEL)
1365                 set_irq_chip_and_handler_name(irq, &ioapic_chip,
1366                                               handle_fasteoi_irq,
1367                                               "fasteoi");
1368         else
1369                 set_irq_chip_and_handler_name(irq, &ioapic_chip,
1370                                               handle_edge_irq, "edge");
1371 }
1372
1373 int setup_ioapic_entry(int apic_id, int irq,
1374                        struct IO_APIC_route_entry *entry,
1375                        unsigned int destination, int trigger,
1376                        int polarity, int vector, int pin)
1377 {
1378         /*
1379          * add it to the IO-APIC irq-routing table:
1380          */
1381         memset(entry,0,sizeof(*entry));
1382
1383         if (intr_remapping_enabled) {
1384                 struct intel_iommu *iommu = map_ioapic_to_ir(apic_id);
1385                 struct irte irte;
1386                 struct IR_IO_APIC_route_entry *ir_entry =
1387                         (struct IR_IO_APIC_route_entry *) entry;
1388                 int index;
1389
1390                 if (!iommu)
1391                         panic("No mapping iommu for ioapic %d\n", apic_id);
1392
1393                 index = alloc_irte(iommu, irq, 1);
1394                 if (index < 0)
1395                         panic("Failed to allocate IRTE for ioapic %d\n", apic_id);
1396
1397                 memset(&irte, 0, sizeof(irte));
1398
1399                 irte.present = 1;
1400                 irte.dst_mode = apic->irq_dest_mode;
1401                 /*
1402                  * Trigger mode in the IRTE will always be edge, and the
1403                  * actual level or edge trigger will be setup in the IO-APIC
1404                  * RTE. This will help simplify level triggered irq migration.
1405                  * For more details, see the comments above explainig IO-APIC
1406                  * irq migration in the presence of interrupt-remapping.
1407                  */
1408                 irte.trigger_mode = 0;
1409                 irte.dlvry_mode = apic->irq_delivery_mode;
1410                 irte.vector = vector;
1411                 irte.dest_id = IRTE_DEST(destination);
1412
1413                 modify_irte(irq, &irte);
1414
1415                 ir_entry->index2 = (index >> 15) & 0x1;
1416                 ir_entry->zero = 0;
1417                 ir_entry->format = 1;
1418                 ir_entry->index = (index & 0x7fff);
1419                 /*
1420                  * IO-APIC RTE will be configured with virtual vector.
1421                  * irq handler will do the explicit EOI to the io-apic.
1422                  */
1423                 ir_entry->vector = pin;
1424         } else {
1425                 entry->delivery_mode = apic->irq_delivery_mode;
1426                 entry->dest_mode = apic->irq_dest_mode;
1427                 entry->dest = destination;
1428                 entry->vector = vector;
1429         }
1430
1431         entry->mask = 0;                                /* enable IRQ */
1432         entry->trigger = trigger;
1433         entry->polarity = polarity;
1434
1435         /* Mask level triggered irqs.
1436          * Use IRQ_DELAYED_DISABLE for edge triggered irqs.
1437          */
1438         if (trigger)
1439                 entry->mask = 1;
1440         return 0;
1441 }
1442
1443 static void setup_IO_APIC_irq(int apic_id, int pin, unsigned int irq, struct irq_desc *desc,
1444                               int trigger, int polarity)
1445 {
1446         struct irq_cfg *cfg;
1447         struct IO_APIC_route_entry entry;
1448         unsigned int dest;
1449
1450         if (!IO_APIC_IRQ(irq))
1451                 return;
1452
1453         cfg = desc->chip_data;
1454
1455         if (assign_irq_vector(irq, cfg, apic->target_cpus()))
1456                 return;
1457
1458         dest = apic->cpu_mask_to_apicid_and(cfg->domain, apic->target_cpus());
1459
1460         apic_printk(APIC_VERBOSE,KERN_DEBUG
1461                     "IOAPIC[%d]: Set routing entry (%d-%d -> 0x%x -> "
1462                     "IRQ %d Mode:%i Active:%i)\n",
1463                     apic_id, mp_ioapics[apic_id].apicid, pin, cfg->vector,
1464                     irq, trigger, polarity);
1465
1466
1467         if (setup_ioapic_entry(mp_ioapics[apic_id].apicid, irq, &entry,
1468                                dest, trigger, polarity, cfg->vector, pin)) {
1469                 printk("Failed to setup ioapic entry for ioapic  %d, pin %d\n",
1470                        mp_ioapics[apic_id].apicid, pin);
1471                 __clear_irq_vector(irq, cfg);
1472                 return;
1473         }
1474
1475         ioapic_register_intr(irq, desc, trigger);
1476         if (irq < NR_IRQS_LEGACY)
1477                 disable_8259A_irq(irq);
1478
1479         ioapic_write_entry(apic_id, pin, entry);
1480 }
1481
1482 static struct {
1483         DECLARE_BITMAP(pin_programmed, MP_MAX_IOAPIC_PIN + 1);
1484 } mp_ioapic_routing[MAX_IO_APICS];
1485
1486 static void __init setup_IO_APIC_irqs(void)
1487 {
1488         int apic_id = 0, pin, idx, irq;
1489         int notcon = 0;
1490         struct irq_desc *desc;
1491         struct irq_cfg *cfg;
1492         int node = cpu_to_node(boot_cpu_id);
1493
1494         apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
1495
1496 #ifdef CONFIG_ACPI
1497         if (!acpi_disabled && acpi_ioapic) {
1498                 apic_id = mp_find_ioapic(0);
1499                 if (apic_id < 0)
1500                         apic_id = 0;
1501         }
1502 #endif
1503
1504         for (pin = 0; pin < nr_ioapic_registers[apic_id]; pin++) {
1505                 idx = find_irq_entry(apic_id, pin, mp_INT);
1506                 if (idx == -1) {
1507                         if (!notcon) {
1508                                 notcon = 1;
1509                                 apic_printk(APIC_VERBOSE,
1510                                         KERN_DEBUG " %d-%d",
1511                                         mp_ioapics[apic_id].apicid, pin);
1512                         } else
1513                                 apic_printk(APIC_VERBOSE, " %d-%d",
1514                                         mp_ioapics[apic_id].apicid, pin);
1515                         continue;
1516                 }
1517                 if (notcon) {
1518                         apic_printk(APIC_VERBOSE,
1519                                 " (apicid-pin) not connected\n");
1520                         notcon = 0;
1521                 }
1522
1523                 irq = pin_2_irq(idx, apic_id, pin);
1524
1525                 /*
1526                  * Skip the timer IRQ if there's a quirk handler
1527                  * installed and if it returns 1:
1528                  */
1529                 if (apic->multi_timer_check &&
1530                                 apic->multi_timer_check(apic_id, irq))
1531                         continue;
1532
1533                 desc = irq_to_desc_alloc_node(irq, node);
1534                 if (!desc) {
1535                         printk(KERN_INFO "can not get irq_desc for %d\n", irq);
1536                         continue;
1537                 }
1538                 cfg = desc->chip_data;
1539                 add_pin_to_irq_node(cfg, node, apic_id, pin);
1540                 /*
1541                  * don't mark it in pin_programmed, so later acpi could
1542                  * set it correctly when irq < 16
1543                  */
1544                 setup_IO_APIC_irq(apic_id, pin, irq, desc,
1545                                 irq_trigger(idx), irq_polarity(idx));
1546         }
1547
1548         if (notcon)
1549                 apic_printk(APIC_VERBOSE,
1550                         " (apicid-pin) not connected\n");
1551 }
1552
1553 /*
1554  * Set up the timer pin, possibly with the 8259A-master behind.
1555  */
1556 static void __init setup_timer_IRQ0_pin(unsigned int apic_id, unsigned int pin,
1557                                         int vector)
1558 {
1559         struct IO_APIC_route_entry entry;
1560
1561         if (intr_remapping_enabled)
1562                 return;
1563
1564         memset(&entry, 0, sizeof(entry));
1565
1566         /*
1567          * We use logical delivery to get the timer IRQ
1568          * to the first CPU.
1569          */
1570         entry.dest_mode = apic->irq_dest_mode;
1571         entry.mask = 0;                 /* don't mask IRQ for edge */
1572         entry.dest = apic->cpu_mask_to_apicid(apic->target_cpus());
1573         entry.delivery_mode = apic->irq_delivery_mode;
1574         entry.polarity = 0;
1575         entry.trigger = 0;
1576         entry.vector = vector;
1577
1578         /*
1579          * The timer IRQ doesn't have to know that behind the
1580          * scene we may have a 8259A-master in AEOI mode ...
1581          */
1582         set_irq_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq, "edge");
1583
1584         /*
1585          * Add it to the IO-APIC irq-routing table:
1586          */
1587         ioapic_write_entry(apic_id, pin, entry);
1588 }
1589
1590
1591 __apicdebuginit(void) print_IO_APIC(void)
1592 {
1593         int apic, i;
1594         union IO_APIC_reg_00 reg_00;
1595         union IO_APIC_reg_01 reg_01;
1596         union IO_APIC_reg_02 reg_02;
1597         union IO_APIC_reg_03 reg_03;
1598         unsigned long flags;
1599         struct irq_cfg *cfg;
1600         struct irq_desc *desc;
1601         unsigned int irq;
1602
1603         if (apic_verbosity == APIC_QUIET)
1604                 return;
1605
1606         printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
1607         for (i = 0; i < nr_ioapics; i++)
1608                 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
1609                        mp_ioapics[i].apicid, nr_ioapic_registers[i]);
1610
1611         /*
1612          * We are a bit conservative about what we expect.  We have to
1613          * know about every hardware change ASAP.
1614          */
1615         printk(KERN_INFO "testing the IO APIC.......................\n");
1616
1617         for (apic = 0; apic < nr_ioapics; apic++) {
1618
1619         spin_lock_irqsave(&ioapic_lock, flags);
1620         reg_00.raw = io_apic_read(apic, 0);
1621         reg_01.raw = io_apic_read(apic, 1);
1622         if (reg_01.bits.version >= 0x10)
1623                 reg_02.raw = io_apic_read(apic, 2);
1624         if (reg_01.bits.version >= 0x20)
1625                 reg_03.raw = io_apic_read(apic, 3);
1626         spin_unlock_irqrestore(&ioapic_lock, flags);
1627
1628         printk("\n");
1629         printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].apicid);
1630         printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
1631         printk(KERN_DEBUG ".......    : physical APIC id: %02X\n", reg_00.bits.ID);
1632         printk(KERN_DEBUG ".......    : Delivery Type: %X\n", reg_00.bits.delivery_type);
1633         printk(KERN_DEBUG ".......    : LTS          : %X\n", reg_00.bits.LTS);
1634
1635         printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)&reg_01);
1636         printk(KERN_DEBUG ".......     : max redirection entries: %04X\n", reg_01.bits.entries);
1637
1638         printk(KERN_DEBUG ".......     : PRQ implemented: %X\n", reg_01.bits.PRQ);
1639         printk(KERN_DEBUG ".......     : IO APIC version: %04X\n", reg_01.bits.version);
1640
1641         /*
1642          * Some Intel chipsets with IO APIC VERSION of 0x1? don't have reg_02,
1643          * but the value of reg_02 is read as the previous read register
1644          * value, so ignore it if reg_02 == reg_01.
1645          */
1646         if (reg_01.bits.version >= 0x10 && reg_02.raw != reg_01.raw) {
1647                 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
1648                 printk(KERN_DEBUG ".......     : arbitration: %02X\n", reg_02.bits.arbitration);
1649         }
1650
1651         /*
1652          * Some Intel chipsets with IO APIC VERSION of 0x2? don't have reg_02
1653          * or reg_03, but the value of reg_0[23] is read as the previous read
1654          * register value, so ignore it if reg_03 == reg_0[12].
1655          */
1656         if (reg_01.bits.version >= 0x20 && reg_03.raw != reg_02.raw &&
1657             reg_03.raw != reg_01.raw) {
1658                 printk(KERN_DEBUG ".... register #03: %08X\n", reg_03.raw);
1659                 printk(KERN_DEBUG ".......     : Boot DT    : %X\n", reg_03.bits.boot_DT);
1660         }
1661
1662         printk(KERN_DEBUG ".... IRQ redirection table:\n");
1663
1664         printk(KERN_DEBUG " NR Dst Mask Trig IRR Pol"
1665                           " Stat Dmod Deli Vect:   \n");
1666
1667         for (i = 0; i <= reg_01.bits.entries; i++) {
1668                 struct IO_APIC_route_entry entry;
1669
1670                 entry = ioapic_read_entry(apic, i);
1671
1672                 printk(KERN_DEBUG " %02x %03X ",
1673                         i,
1674                         entry.dest
1675                 );
1676
1677                 printk("%1d    %1d    %1d   %1d   %1d    %1d    %1d    %02X\n",
1678                         entry.mask,
1679                         entry.trigger,
1680                         entry.irr,
1681                         entry.polarity,
1682                         entry.delivery_status,
1683                         entry.dest_mode,
1684                         entry.delivery_mode,
1685                         entry.vector
1686                 );
1687         }
1688         }
1689         printk(KERN_DEBUG "IRQ to pin mappings:\n");
1690         for_each_irq_desc(irq, desc) {
1691                 struct irq_pin_list *entry;
1692
1693                 cfg = desc->chip_data;
1694                 entry = cfg->irq_2_pin;
1695                 if (!entry)
1696                         continue;
1697                 printk(KERN_DEBUG "IRQ%d ", irq);
1698                 for (;;) {
1699                         printk("-> %d:%d", entry->apic, entry->pin);
1700                         if (!entry->next)
1701                                 break;
1702                         entry = entry->next;
1703                 }
1704                 printk("\n");
1705         }
1706
1707         printk(KERN_INFO ".................................... done.\n");
1708
1709         return;
1710 }
1711
1712 __apicdebuginit(void) print_APIC_bitfield(int base)
1713 {
1714         unsigned int v;
1715         int i, j;
1716
1717         if (apic_verbosity == APIC_QUIET)
1718                 return;
1719
1720         printk(KERN_DEBUG "0123456789abcdef0123456789abcdef\n" KERN_DEBUG);
1721         for (i = 0; i < 8; i++) {
1722                 v = apic_read(base + i*0x10);
1723                 for (j = 0; j < 32; j++) {
1724                         if (v & (1<<j))
1725                                 printk("1");
1726                         else
1727                                 printk("0");
1728                 }
1729                 printk("\n");
1730         }
1731 }
1732
1733 __apicdebuginit(void) print_local_APIC(void *dummy)
1734 {
1735         unsigned int i, v, ver, maxlvt;
1736         u64 icr;
1737
1738         if (apic_verbosity == APIC_QUIET)
1739                 return;
1740
1741         printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1742                 smp_processor_id(), hard_smp_processor_id());
1743         v = apic_read(APIC_ID);
1744         printk(KERN_INFO "... APIC ID:      %08x (%01x)\n", v, read_apic_id());
1745         v = apic_read(APIC_LVR);
1746         printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1747         ver = GET_APIC_VERSION(v);
1748         maxlvt = lapic_get_maxlvt();
1749
1750         v = apic_read(APIC_TASKPRI);
1751         printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1752
1753         if (APIC_INTEGRATED(ver)) {                     /* !82489DX */
1754                 if (!APIC_XAPIC(ver)) {
1755                         v = apic_read(APIC_ARBPRI);
1756                         printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
1757                                v & APIC_ARBPRI_MASK);
1758                 }
1759                 v = apic_read(APIC_PROCPRI);
1760                 printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1761         }
1762
1763         /*
1764          * Remote read supported only in the 82489DX and local APIC for
1765          * Pentium processors.
1766          */
1767         if (!APIC_INTEGRATED(ver) || maxlvt == 3) {
1768                 v = apic_read(APIC_RRR);
1769                 printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1770         }
1771
1772         v = apic_read(APIC_LDR);
1773         printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1774         if (!x2apic_enabled()) {
1775                 v = apic_read(APIC_DFR);
1776                 printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1777         }
1778         v = apic_read(APIC_SPIV);
1779         printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1780
1781         printk(KERN_DEBUG "... APIC ISR field:\n");
1782         print_APIC_bitfield(APIC_ISR);
1783         printk(KERN_DEBUG "... APIC TMR field:\n");
1784         print_APIC_bitfield(APIC_TMR);
1785         printk(KERN_DEBUG "... APIC IRR field:\n");
1786         print_APIC_bitfield(APIC_IRR);
1787
1788         if (APIC_INTEGRATED(ver)) {             /* !82489DX */
1789                 if (maxlvt > 3)         /* Due to the Pentium erratum 3AP. */
1790                         apic_write(APIC_ESR, 0);
1791
1792                 v = apic_read(APIC_ESR);
1793                 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1794         }
1795
1796         icr = apic_icr_read();
1797         printk(KERN_DEBUG "... APIC ICR: %08x\n", (u32)icr);
1798         printk(KERN_DEBUG "... APIC ICR2: %08x\n", (u32)(icr >> 32));
1799
1800         v = apic_read(APIC_LVTT);
1801         printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1802
1803         if (maxlvt > 3) {                       /* PC is LVT#4. */
1804                 v = apic_read(APIC_LVTPC);
1805                 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1806         }
1807         v = apic_read(APIC_LVT0);
1808         printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1809         v = apic_read(APIC_LVT1);
1810         printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1811
1812         if (maxlvt > 2) {                       /* ERR is LVT#3. */
1813                 v = apic_read(APIC_LVTERR);
1814                 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1815         }
1816
1817         v = apic_read(APIC_TMICT);
1818         printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1819         v = apic_read(APIC_TMCCT);
1820         printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1821         v = apic_read(APIC_TDCR);
1822         printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1823
1824         if (boot_cpu_has(X86_FEATURE_EXTAPIC)) {
1825                 v = apic_read(APIC_EFEAT);
1826                 maxlvt = (v >> 16) & 0xff;
1827                 printk(KERN_DEBUG "... APIC EFEAT: %08x\n", v);
1828                 v = apic_read(APIC_ECTRL);
1829                 printk(KERN_DEBUG "... APIC ECTRL: %08x\n", v);
1830                 for (i = 0; i < maxlvt; i++) {
1831                         v = apic_read(APIC_EILVTn(i));
1832                         printk(KERN_DEBUG "... APIC EILVT%d: %08x\n", i, v);
1833                 }
1834         }
1835         printk("\n");
1836 }
1837
1838 __apicdebuginit(void) print_all_local_APICs(void)
1839 {
1840         int cpu;
1841
1842         preempt_disable();
1843         for_each_online_cpu(cpu)
1844                 smp_call_function_single(cpu, print_local_APIC, NULL, 1);
1845         preempt_enable();
1846 }
1847
1848 __apicdebuginit(void) print_PIC(void)
1849 {
1850         unsigned int v;
1851         unsigned long flags;
1852
1853         if (apic_verbosity == APIC_QUIET)
1854                 return;
1855
1856         printk(KERN_DEBUG "\nprinting PIC contents\n");
1857
1858         spin_lock_irqsave(&i8259A_lock, flags);
1859
1860         v = inb(0xa1) << 8 | inb(0x21);
1861         printk(KERN_DEBUG "... PIC  IMR: %04x\n", v);
1862
1863         v = inb(0xa0) << 8 | inb(0x20);
1864         printk(KERN_DEBUG "... PIC  IRR: %04x\n", v);
1865
1866         outb(0x0b,0xa0);
1867         outb(0x0b,0x20);
1868         v = inb(0xa0) << 8 | inb(0x20);
1869         outb(0x0a,0xa0);
1870         outb(0x0a,0x20);
1871
1872         spin_unlock_irqrestore(&i8259A_lock, flags);
1873
1874         printk(KERN_DEBUG "... PIC  ISR: %04x\n", v);
1875
1876         v = inb(0x4d1) << 8 | inb(0x4d0);
1877         printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1878 }
1879
1880 __apicdebuginit(int) print_all_ICs(void)
1881 {
1882         print_PIC();
1883
1884         /* don't print out if apic is not there */
1885         if (!cpu_has_apic || disable_apic)
1886                 return 0;
1887
1888         print_all_local_APICs();
1889         print_IO_APIC();
1890
1891         return 0;
1892 }
1893
1894 fs_initcall(print_all_ICs);
1895
1896
1897 /* Where if anywhere is the i8259 connect in external int mode */
1898 static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
1899
1900 void __init enable_IO_APIC(void)
1901 {
1902         union IO_APIC_reg_01 reg_01;
1903         int i8259_apic, i8259_pin;
1904         int apic;
1905         unsigned long flags;
1906
1907         /*
1908          * The number of IO-APIC IRQ registers (== #pins):
1909          */
1910         for (apic = 0; apic < nr_ioapics; apic++) {
1911                 spin_lock_irqsave(&ioapic_lock, flags);
1912                 reg_01.raw = io_apic_read(apic, 1);
1913                 spin_unlock_irqrestore(&ioapic_lock, flags);
1914                 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
1915         }
1916         for(apic = 0; apic < nr_ioapics; apic++) {
1917                 int pin;
1918                 /* See if any of the pins is in ExtINT mode */
1919                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1920                         struct IO_APIC_route_entry entry;
1921                         entry = ioapic_read_entry(apic, pin);
1922
1923                         /* If the interrupt line is enabled and in ExtInt mode
1924                          * I have found the pin where the i8259 is connected.
1925                          */
1926                         if ((entry.mask == 0) && (entry.delivery_mode == dest_ExtINT)) {
1927                                 ioapic_i8259.apic = apic;
1928                                 ioapic_i8259.pin  = pin;
1929                                 goto found_i8259;
1930                         }
1931                 }
1932         }
1933  found_i8259:
1934         /* Look to see what if the MP table has reported the ExtINT */
1935         /* If we could not find the appropriate pin by looking at the ioapic
1936          * the i8259 probably is not connected the ioapic but give the
1937          * mptable a chance anyway.
1938          */
1939         i8259_pin  = find_isa_irq_pin(0, mp_ExtINT);
1940         i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
1941         /* Trust the MP table if nothing is setup in the hardware */
1942         if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
1943                 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
1944                 ioapic_i8259.pin  = i8259_pin;
1945                 ioapic_i8259.apic = i8259_apic;
1946         }
1947         /* Complain if the MP table and the hardware disagree */
1948         if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
1949                 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
1950         {
1951                 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
1952         }
1953
1954         /*
1955          * Do not trust the IO-APIC being empty at bootup
1956          */
1957         clear_IO_APIC();
1958 }
1959
1960 /*
1961  * Not an __init, needed by the reboot code
1962  */
1963 void disable_IO_APIC(void)
1964 {
1965         /*
1966          * Clear the IO-APIC before rebooting:
1967          */
1968         clear_IO_APIC();
1969
1970         /*
1971          * If the i8259 is routed through an IOAPIC
1972          * Put that IOAPIC in virtual wire mode
1973          * so legacy interrupts can be delivered.
1974          *
1975          * With interrupt-remapping, for now we will use virtual wire A mode,
1976          * as virtual wire B is little complex (need to configure both
1977          * IOAPIC RTE aswell as interrupt-remapping table entry).
1978          * As this gets called during crash dump, keep this simple for now.
1979          */
1980         if (ioapic_i8259.pin != -1 && !intr_remapping_enabled) {
1981                 struct IO_APIC_route_entry entry;
1982
1983                 memset(&entry, 0, sizeof(entry));
1984                 entry.mask            = 0; /* Enabled */
1985                 entry.trigger         = 0; /* Edge */
1986                 entry.irr             = 0;
1987                 entry.polarity        = 0; /* High */
1988                 entry.delivery_status = 0;
1989                 entry.dest_mode       = 0; /* Physical */
1990                 entry.delivery_mode   = dest_ExtINT; /* ExtInt */
1991                 entry.vector          = 0;
1992                 entry.dest            = read_apic_id();
1993
1994                 /*
1995                  * Add it to the IO-APIC irq-routing table:
1996                  */
1997                 ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
1998         }
1999
2000         /*
2001          * Use virtual wire A mode when interrupt remapping is enabled.
2002          */
2003         disconnect_bsp_APIC(!intr_remapping_enabled && ioapic_i8259.pin != -1);
2004 }
2005
2006 #ifdef CONFIG_X86_32
2007 /*
2008  * function to set the IO-APIC physical IDs based on the
2009  * values stored in the MPC table.
2010  *
2011  * by Matt Domsch <Matt_Domsch@dell.com>  Tue Dec 21 12:25:05 CST 1999
2012  */
2013
2014 static void __init setup_ioapic_ids_from_mpc(void)
2015 {
2016         union IO_APIC_reg_00 reg_00;
2017         physid_mask_t phys_id_present_map;
2018         int apic_id;
2019         int i;
2020         unsigned char old_id;
2021         unsigned long flags;
2022
2023         if (x86_quirks->setup_ioapic_ids && x86_quirks->setup_ioapic_ids())
2024                 return;
2025
2026         /*
2027          * Don't check I/O APIC IDs for xAPIC systems.  They have
2028          * no meaning without the serial APIC bus.
2029          */
2030         if (!(boot_cpu_data.x86_vendor == X86_VENDOR_INTEL)
2031                 || APIC_XAPIC(apic_version[boot_cpu_physical_apicid]))
2032                 return;
2033         /*
2034          * This is broken; anything with a real cpu count has to
2035          * circumvent this idiocy regardless.
2036          */
2037         phys_id_present_map = apic->ioapic_phys_id_map(phys_cpu_present_map);
2038
2039         /*
2040          * Set the IOAPIC ID to the value stored in the MPC table.
2041          */
2042         for (apic_id = 0; apic_id < nr_ioapics; apic_id++) {
2043
2044                 /* Read the register 0 value */
2045                 spin_lock_irqsave(&ioapic_lock, flags);
2046                 reg_00.raw = io_apic_read(apic_id, 0);
2047                 spin_unlock_irqrestore(&ioapic_lock, flags);
2048
2049                 old_id = mp_ioapics[apic_id].apicid;
2050
2051                 if (mp_ioapics[apic_id].apicid >= get_physical_broadcast()) {
2052                         printk(KERN_ERR "BIOS bug, IO-APIC#%d ID is %d in the MPC table!...\n",
2053                                 apic_id, mp_ioapics[apic_id].apicid);
2054                         printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
2055                                 reg_00.bits.ID);
2056                         mp_ioapics[apic_id].apicid = reg_00.bits.ID;
2057                 }
2058
2059                 /*
2060                  * Sanity check, is the ID really free? Every APIC in a
2061                  * system must have a unique ID or we get lots of nice
2062                  * 'stuck on smp_invalidate_needed IPI wait' messages.
2063                  */
2064                 if (apic->check_apicid_used(phys_id_present_map,
2065                                         mp_ioapics[apic_id].apicid)) {
2066                         printk(KERN_ERR "BIOS bug, IO-APIC#%d ID %d is already used!...\n",
2067                                 apic_id, mp_ioapics[apic_id].apicid);
2068                         for (i = 0; i < get_physical_broadcast(); i++)
2069                                 if (!physid_isset(i, phys_id_present_map))
2070                                         break;
2071                         if (i >= get_physical_broadcast())
2072                                 panic("Max APIC ID exceeded!\n");
2073                         printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
2074                                 i);
2075                         physid_set(i, phys_id_present_map);
2076                         mp_ioapics[apic_id].apicid = i;
2077                 } else {
2078                         physid_mask_t tmp;
2079                         tmp = apic->apicid_to_cpu_present(mp_ioapics[apic_id].apicid);
2080                         apic_printk(APIC_VERBOSE, "Setting %d in the "
2081                                         "phys_id_present_map\n",
2082                                         mp_ioapics[apic_id].apicid);
2083                         physids_or(phys_id_present_map, phys_id_present_map, tmp);
2084                 }
2085
2086
2087                 /*
2088                  * We need to adjust the IRQ routing table
2089                  * if the ID changed.
2090                  */
2091                 if (old_id != mp_ioapics[apic_id].apicid)
2092                         for (i = 0; i < mp_irq_entries; i++)
2093                                 if (mp_irqs[i].dstapic == old_id)
2094                                         mp_irqs[i].dstapic
2095                                                 = mp_ioapics[apic_id].apicid;
2096
2097                 /*
2098                  * Read the right value from the MPC table and
2099                  * write it into the ID register.
2100                  */
2101                 apic_printk(APIC_VERBOSE, KERN_INFO
2102                         "...changing IO-APIC physical APIC ID to %d ...",
2103                         mp_ioapics[apic_id].apicid);
2104
2105                 reg_00.bits.ID = mp_ioapics[apic_id].apicid;
2106                 spin_lock_irqsave(&ioapic_lock, flags);
2107                 io_apic_write(apic_id, 0, reg_00.raw);
2108                 spin_unlock_irqrestore(&ioapic_lock, flags);
2109
2110                 /*
2111                  * Sanity check
2112                  */
2113                 spin_lock_irqsave(&ioapic_lock, flags);
2114                 reg_00.raw = io_apic_read(apic_id, 0);
2115                 spin_unlock_irqrestore(&ioapic_lock, flags);
2116                 if (reg_00.bits.ID != mp_ioapics[apic_id].apicid)
2117                         printk("could not set ID!\n");
2118                 else
2119                         apic_printk(APIC_VERBOSE, " ok.\n");
2120         }
2121 }
2122 #endif
2123
2124 int no_timer_check __initdata;
2125
2126 static int __init notimercheck(char *s)
2127 {
2128         no_timer_check = 1;
2129         return 1;
2130 }
2131 __setup("no_timer_check", notimercheck);
2132
2133 /*
2134  * There is a nasty bug in some older SMP boards, their mptable lies
2135  * about the timer IRQ. We do the following to work around the situation:
2136  *
2137  *      - timer IRQ defaults to IO-APIC IRQ
2138  *      - if this function detects that timer IRQs are defunct, then we fall
2139  *        back to ISA timer IRQs
2140  */
2141 static int __init timer_irq_works(void)
2142 {
2143         unsigned long t1 = jiffies;
2144         unsigned long flags;
2145
2146         if (no_timer_check)
2147                 return 1;
2148
2149         local_save_flags(flags);
2150         local_irq_enable();
2151         /* Let ten ticks pass... */
2152         mdelay((10 * 1000) / HZ);
2153         local_irq_restore(flags);
2154
2155         /*
2156          * Expect a few ticks at least, to be sure some possible
2157          * glue logic does not lock up after one or two first
2158          * ticks in a non-ExtINT mode.  Also the local APIC
2159          * might have cached one ExtINT interrupt.  Finally, at
2160          * least one tick may be lost due to delays.
2161          */
2162
2163         /* jiffies wrap? */
2164         if (time_after(jiffies, t1 + 4))
2165                 return 1;
2166         return 0;
2167 }
2168
2169 /*
2170  * In the SMP+IOAPIC case it might happen that there are an unspecified
2171  * number of pending IRQ events unhandled. These cases are very rare,
2172  * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
2173  * better to do it this way as thus we do not have to be aware of
2174  * 'pending' interrupts in the IRQ path, except at this point.
2175  */
2176 /*
2177  * Edge triggered needs to resend any interrupt
2178  * that was delayed but this is now handled in the device
2179  * independent code.
2180  */
2181
2182 /*
2183  * Starting up a edge-triggered IO-APIC interrupt is
2184  * nasty - we need to make sure that we get the edge.
2185  * If it is already asserted for some reason, we need
2186  * return 1 to indicate that is was pending.
2187  *
2188  * This is not complete - we should be able to fake
2189  * an edge even if it isn't on the 8259A...
2190  */
2191
2192 static unsigned int startup_ioapic_irq(unsigned int irq)
2193 {
2194         int was_pending = 0;
2195         unsigned long flags;
2196         struct irq_cfg *cfg;
2197
2198         spin_lock_irqsave(&ioapic_lock, flags);
2199         if (irq < NR_IRQS_LEGACY) {
2200                 disable_8259A_irq(irq);
2201                 if (i8259A_irq_pending(irq))
2202                         was_pending = 1;
2203         }
2204         cfg = irq_cfg(irq);
2205         __unmask_IO_APIC_irq(cfg);
2206         spin_unlock_irqrestore(&ioapic_lock, flags);
2207
2208         return was_pending;
2209 }
2210
2211 #ifdef CONFIG_X86_64
2212 static int ioapic_retrigger_irq(unsigned int irq)
2213 {
2214
2215         struct irq_cfg *cfg = irq_cfg(irq);
2216         unsigned long flags;
2217
2218         spin_lock_irqsave(&vector_lock, flags);
2219         apic->send_IPI_mask(cpumask_of(cpumask_first(cfg->domain)), cfg->vector);
2220         spin_unlock_irqrestore(&vector_lock, flags);
2221
2222         return 1;
2223 }
2224 #else
2225 static int ioapic_retrigger_irq(unsigned int irq)
2226 {
2227         apic->send_IPI_self(irq_cfg(irq)->vector);
2228
2229         return 1;
2230 }
2231 #endif
2232
2233 /*
2234  * Level and edge triggered IO-APIC interrupts need different handling,
2235  * so we use two separate IRQ descriptors. Edge triggered IRQs can be
2236  * handled with the level-triggered descriptor, but that one has slightly
2237  * more overhead. Level-triggered interrupts cannot be handled with the
2238  * edge-triggered handler, without risking IRQ storms and other ugly
2239  * races.
2240  */
2241
2242 #ifdef CONFIG_SMP
2243 static void send_cleanup_vector(struct irq_cfg *cfg)
2244 {
2245         cpumask_var_t cleanup_mask;
2246
2247         if (unlikely(!alloc_cpumask_var(&cleanup_mask, GFP_ATOMIC))) {
2248                 unsigned int i;
2249                 cfg->move_cleanup_count = 0;
2250                 for_each_cpu_and(i, cfg->old_domain, cpu_online_mask)
2251                         cfg->move_cleanup_count++;
2252                 for_each_cpu_and(i, cfg->old_domain, cpu_online_mask)
2253                         apic->send_IPI_mask(cpumask_of(i), IRQ_MOVE_CLEANUP_VECTOR);
2254         } else {
2255                 cpumask_and(cleanup_mask, cfg->old_domain, cpu_online_mask);
2256                 cfg->move_cleanup_count = cpumask_weight(cleanup_mask);
2257                 apic->send_IPI_mask(cleanup_mask, IRQ_MOVE_CLEANUP_VECTOR);
2258                 free_cpumask_var(cleanup_mask);
2259         }
2260         cfg->move_in_progress = 0;
2261 }
2262
2263 static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, struct irq_cfg *cfg)
2264 {
2265         int apic, pin;
2266         struct irq_pin_list *entry;
2267         u8 vector = cfg->vector;
2268
2269         entry = cfg->irq_2_pin;
2270         for (;;) {
2271                 unsigned int reg;
2272
2273                 if (!entry)
2274                         break;
2275
2276                 apic = entry->apic;
2277                 pin = entry->pin;
2278                 /*
2279                  * With interrupt-remapping, destination information comes
2280                  * from interrupt-remapping table entry.
2281                  */
2282                 if (!irq_remapped(irq))
2283                         io_apic_write(apic, 0x11 + pin*2, dest);
2284                 reg = io_apic_read(apic, 0x10 + pin*2);
2285                 reg &= ~IO_APIC_REDIR_VECTOR_MASK;
2286                 reg |= vector;
2287                 io_apic_modify(apic, 0x10 + pin*2, reg);
2288                 if (!entry->next)
2289                         break;
2290                 entry = entry->next;
2291         }
2292 }
2293
2294 static int
2295 assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask);
2296
2297 /*
2298  * Either sets desc->affinity to a valid value, and returns
2299  * ->cpu_mask_to_apicid of that, or returns BAD_APICID and
2300  * leaves desc->affinity untouched.
2301  */
2302 static unsigned int
2303 set_desc_affinity(struct irq_desc *desc, const struct cpumask *mask)
2304 {
2305         struct irq_cfg *cfg;
2306         unsigned int irq;
2307
2308         if (!cpumask_intersects(mask, cpu_online_mask))
2309                 return BAD_APICID;
2310
2311         irq = desc->irq;
2312         cfg = desc->chip_data;
2313         if (assign_irq_vector(irq, cfg, mask))
2314                 return BAD_APICID;
2315
2316         cpumask_copy(desc->affinity, mask);
2317
2318         return apic->cpu_mask_to_apicid_and(desc->affinity, cfg->domain);
2319 }
2320
2321 static int
2322 set_ioapic_affinity_irq_desc(struct irq_desc *desc, const struct cpumask *mask)
2323 {
2324         struct irq_cfg *cfg;
2325         unsigned long flags;
2326         unsigned int dest;
2327         unsigned int irq;
2328         int ret = -1;
2329
2330         irq = desc->irq;
2331         cfg = desc->chip_data;
2332
2333         spin_lock_irqsave(&ioapic_lock, flags);
2334         dest = set_desc_affinity(desc, mask);
2335         if (dest != BAD_APICID) {
2336                 /* Only the high 8 bits are valid. */
2337                 dest = SET_APIC_LOGICAL_ID(dest);
2338                 __target_IO_APIC_irq(irq, dest, cfg);
2339                 ret = 0;
2340         }
2341         spin_unlock_irqrestore(&ioapic_lock, flags);
2342
2343         return ret;
2344 }
2345
2346 static int
2347 set_ioapic_affinity_irq(unsigned int irq, const struct cpumask *mask)
2348 {
2349         struct irq_desc *desc;
2350
2351         desc = irq_to_desc(irq);
2352
2353         return set_ioapic_affinity_irq_desc(desc, mask);
2354 }
2355
2356 #ifdef CONFIG_INTR_REMAP
2357
2358 /*
2359  * Migrate the IO-APIC irq in the presence of intr-remapping.
2360  *
2361  * For both level and edge triggered, irq migration is a simple atomic
2362  * update(of vector and cpu destination) of IRTE and flush the hardware cache.
2363  *
2364  * For level triggered, we eliminate the io-apic RTE modification (with the
2365  * updated vector information), by using a virtual vector (io-apic pin number).
2366  * Real vector that is used for interrupting cpu will be coming from
2367  * the interrupt-remapping table entry.
2368  */
2369 static int
2370 migrate_ioapic_irq_desc(struct irq_desc *desc, const struct cpumask *mask)
2371 {
2372         struct irq_cfg *cfg;
2373         struct irte irte;
2374         unsigned int dest;
2375         unsigned int irq;
2376         int ret = -1;
2377
2378         if (!cpumask_intersects(mask, cpu_online_mask))
2379                 return ret;
2380
2381         irq = desc->irq;
2382         if (get_irte(irq, &irte))
2383                 return ret;
2384
2385         cfg = desc->chip_data;
2386         if (assign_irq_vector(irq, cfg, mask))
2387                 return ret;
2388
2389         dest = apic->cpu_mask_to_apicid_and(cfg->domain, mask);
2390
2391         irte.vector = cfg->vector;
2392         irte.dest_id = IRTE_DEST(dest);
2393
2394         /*
2395          * Modified the IRTE and flushes the Interrupt entry cache.
2396          */
2397         modify_irte(irq, &irte);
2398
2399         if (cfg->move_in_progress)
2400                 send_cleanup_vector(cfg);
2401
2402         cpumask_copy(desc->affinity, mask);
2403
2404         return 0;
2405 }
2406
2407 /*
2408  * Migrates the IRQ destination in the process context.
2409  */
2410 static int set_ir_ioapic_affinity_irq_desc(struct irq_desc *desc,
2411                                             const struct cpumask *mask)
2412 {
2413         return migrate_ioapic_irq_desc(desc, mask);
2414 }
2415 static int set_ir_ioapic_affinity_irq(unsigned int irq,
2416                                        const struct cpumask *mask)
2417 {
2418         struct irq_desc *desc = irq_to_desc(irq);
2419
2420         return set_ir_ioapic_affinity_irq_desc(desc, mask);
2421 }
2422 #else
2423 static inline int set_ir_ioapic_affinity_irq_desc(struct irq_desc *desc,
2424                                                    const struct cpumask *mask)
2425 {
2426         return 0;
2427 }
2428 #endif
2429
2430 asmlinkage void smp_irq_move_cleanup_interrupt(void)
2431 {
2432         unsigned vector, me;
2433
2434         ack_APIC_irq();
2435         exit_idle();
2436         irq_enter();
2437
2438         me = smp_processor_id();
2439         for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
2440                 unsigned int irq;
2441                 unsigned int irr;
2442                 struct irq_desc *desc;
2443                 struct irq_cfg *cfg;
2444                 irq = __get_cpu_var(vector_irq)[vector];
2445
2446                 if (irq == -1)
2447                         continue;
2448
2449                 desc = irq_to_desc(irq);
2450                 if (!desc)
2451                         continue;
2452
2453                 cfg = irq_cfg(irq);
2454                 spin_lock(&desc->lock);
2455                 if (!cfg->move_cleanup_count)
2456                         goto unlock;
2457
2458                 if (vector == cfg->vector && cpumask_test_cpu(me, cfg->domain))
2459                         goto unlock;
2460
2461                 irr = apic_read(APIC_IRR + (vector / 32 * 0x10));
2462                 /*
2463                  * Check if the vector that needs to be cleanedup is
2464                  * registered at the cpu's IRR. If so, then this is not
2465                  * the best time to clean it up. Lets clean it up in the
2466                  * next attempt by sending another IRQ_MOVE_CLEANUP_VECTOR
2467                  * to myself.
2468                  */
2469                 if (irr  & (1 << (vector % 32))) {
2470                         apic->send_IPI_self(IRQ_MOVE_CLEANUP_VECTOR);
2471                         goto unlock;
2472                 }
2473                 __get_cpu_var(vector_irq)[vector] = -1;
2474                 cfg->move_cleanup_count--;
2475 unlock:
2476                 spin_unlock(&desc->lock);
2477         }
2478
2479         irq_exit();
2480 }
2481
2482 static void irq_complete_move(struct irq_desc **descp)
2483 {
2484         struct irq_desc *desc = *descp;
2485         struct irq_cfg *cfg = desc->chip_data;
2486         unsigned vector, me;
2487
2488         if (likely(!cfg->move_in_progress))
2489                 return;
2490
2491         vector = ~get_irq_regs()->orig_ax;
2492         me = smp_processor_id();
2493
2494         if (vector == cfg->vector && cpumask_test_cpu(me, cfg->domain))
2495                 send_cleanup_vector(cfg);
2496 }
2497 #else
2498 static inline void irq_complete_move(struct irq_desc **descp) {}
2499 #endif
2500
2501 static void ack_apic_edge(unsigned int irq)
2502 {
2503         struct irq_desc *desc = irq_to_desc(irq);
2504
2505         irq_complete_move(&desc);
2506         move_native_irq(irq);
2507         ack_APIC_irq();
2508 }
2509
2510 atomic_t irq_mis_count;
2511
2512 static void ack_apic_level(unsigned int irq)
2513 {
2514         struct irq_desc *desc = irq_to_desc(irq);
2515
2516 #ifdef CONFIG_X86_32
2517         unsigned long v;
2518         int i;
2519 #endif
2520         struct irq_cfg *cfg;
2521         int do_unmask_irq = 0;
2522
2523         irq_complete_move(&desc);
2524 #ifdef CONFIG_GENERIC_PENDING_IRQ
2525         /* If we are moving the irq we need to mask it */
2526         if (unlikely(desc->status & IRQ_MOVE_PENDING)) {
2527                 do_unmask_irq = 1;
2528                 mask_IO_APIC_irq_desc(desc);
2529         }
2530 #endif
2531
2532 #ifdef CONFIG_X86_32
2533         /*
2534         * It appears there is an erratum which affects at least version 0x11
2535         * of I/O APIC (that's the 82093AA and cores integrated into various
2536         * chipsets).  Under certain conditions a level-triggered interrupt is
2537         * erroneously delivered as edge-triggered one but the respective IRR
2538         * bit gets set nevertheless.  As a result the I/O unit expects an EOI
2539         * message but it will never arrive and further interrupts are blocked
2540         * from the source.  The exact reason is so far unknown, but the
2541         * phenomenon was observed when two consecutive interrupt requests
2542         * from a given source get delivered to the same CPU and the source is
2543         * temporarily disabled in between.
2544         *
2545         * A workaround is to simulate an EOI message manually.  We achieve it
2546         * by setting the trigger mode to edge and then to level when the edge
2547         * trigger mode gets detected in the TMR of a local APIC for a
2548         * level-triggered interrupt.  We mask the source for the time of the
2549         * operation to prevent an edge-triggered interrupt escaping meanwhile.
2550         * The idea is from Manfred Spraul.  --macro
2551         */
2552         cfg = desc->chip_data;
2553         i = cfg->vector;
2554
2555         v = apic_read(APIC_TMR + ((i & ~0x1f) >> 1));
2556 #endif
2557
2558         /*
2559          * We must acknowledge the irq before we move it or the acknowledge will
2560          * not propagate properly.
2561          */
2562         ack_APIC_irq();
2563
2564         /* Now we can move and renable the irq */
2565         if (unlikely(do_unmask_irq)) {
2566                 /* Only migrate the irq if the ack has been received.
2567                  *
2568                  * On rare occasions the broadcast level triggered ack gets
2569                  * delayed going to ioapics, and if we reprogram the
2570                  * vector while Remote IRR is still set the irq will never
2571                  * fire again.
2572                  *
2573                  * To prevent this scenario we read the Remote IRR bit
2574                  * of the ioapic.  This has two effects.
2575                  * - On any sane system the read of the ioapic will
2576                  *   flush writes (and acks) going to the ioapic from
2577                  *   this cpu.
2578                  * - We get to see if the ACK has actually been delivered.
2579                  *
2580                  * Based on failed experiments of reprogramming the
2581                  * ioapic entry from outside of irq context starting
2582                  * with masking the ioapic entry and then polling until
2583                  * Remote IRR was clear before reprogramming the
2584                  * ioapic I don't trust the Remote IRR bit to be
2585                  * completey accurate.
2586                  *
2587                  * However there appears to be no other way to plug
2588                  * this race, so if the Remote IRR bit is not
2589                  * accurate and is causing problems then it is a hardware bug
2590                  * and you can go talk to the chipset vendor about it.
2591                  */
2592                 cfg = desc->chip_data;
2593                 if (!io_apic_level_ack_pending(cfg))
2594                         move_masked_irq(irq);
2595                 unmask_IO_APIC_irq_desc(desc);
2596         }
2597
2598 #ifdef CONFIG_X86_32
2599         if (!(v & (1 << (i & 0x1f)))) {
2600                 atomic_inc(&irq_mis_count);
2601                 spin_lock(&ioapic_lock);
2602                 __mask_and_edge_IO_APIC_irq(cfg);
2603                 __unmask_and_level_IO_APIC_irq(cfg);
2604                 spin_unlock(&ioapic_lock);
2605         }
2606 #endif
2607 }
2608
2609 #ifdef CONFIG_INTR_REMAP
2610 static void __eoi_ioapic_irq(unsigned int irq, struct irq_cfg *cfg)
2611 {
2612         int apic, pin;
2613         struct irq_pin_list *entry;
2614
2615         entry = cfg->irq_2_pin;
2616         for (;;) {
2617
2618                 if (!entry)
2619                         break;
2620
2621                 apic = entry->apic;
2622                 pin = entry->pin;
2623                 io_apic_eoi(apic, pin);
2624                 entry = entry->next;
2625         }
2626 }
2627
2628 static void
2629 eoi_ioapic_irq(struct irq_desc *desc)
2630 {
2631         struct irq_cfg *cfg;
2632         unsigned long flags;
2633         unsigned int irq;
2634
2635         irq = desc->irq;
2636         cfg = desc->chip_data;
2637
2638         spin_lock_irqsave(&ioapic_lock, flags);
2639         __eoi_ioapic_irq(irq, cfg);
2640         spin_unlock_irqrestore(&ioapic_lock, flags);
2641 }
2642
2643 static void ir_ack_apic_edge(unsigned int irq)
2644 {
2645         ack_APIC_irq();
2646 }
2647
2648 static void ir_ack_apic_level(unsigned int irq)
2649 {
2650         struct irq_desc *desc = irq_to_desc(irq);
2651
2652         ack_APIC_irq();
2653         eoi_ioapic_irq(desc);
2654 }
2655 #endif /* CONFIG_INTR_REMAP */
2656
2657 static struct irq_chip ioapic_chip __read_mostly = {
2658         .name           = "IO-APIC",
2659         .startup        = startup_ioapic_irq,
2660         .mask           = mask_IO_APIC_irq,
2661         .unmask         = unmask_IO_APIC_irq,
2662         .ack            = ack_apic_edge,
2663         .eoi            = ack_apic_level,
2664 #ifdef CONFIG_SMP
2665         .set_affinity   = set_ioapic_affinity_irq,
2666 #endif
2667         .retrigger      = ioapic_retrigger_irq,
2668 };
2669
2670 static struct irq_chip ir_ioapic_chip __read_mostly = {
2671         .name           = "IR-IO-APIC",
2672         .startup        = startup_ioapic_irq,
2673         .mask           = mask_IO_APIC_irq,
2674         .unmask         = unmask_IO_APIC_irq,
2675 #ifdef CONFIG_INTR_REMAP
2676         .ack            = ir_ack_apic_edge,
2677         .eoi            = ir_ack_apic_level,
2678 #ifdef CONFIG_SMP
2679         .set_affinity   = set_ir_ioapic_affinity_irq,
2680 #endif
2681 #endif
2682         .retrigger      = ioapic_retrigger_irq,
2683 };
2684
2685 static inline void init_IO_APIC_traps(void)
2686 {
2687         int irq;
2688         struct irq_desc *desc;
2689         struct irq_cfg *cfg;
2690
2691         /*
2692          * NOTE! The local APIC isn't very good at handling
2693          * multiple interrupts at the same interrupt level.
2694          * As the interrupt level is determined by taking the
2695          * vector number and shifting that right by 4, we
2696          * want to spread these out a bit so that they don't
2697          * all fall in the same interrupt level.
2698          *
2699          * Also, we've got to be careful not to trash gate
2700          * 0x80, because int 0x80 is hm, kind of importantish. ;)
2701          */
2702         for_each_irq_desc(irq, desc) {
2703                 cfg = desc->chip_data;
2704                 if (IO_APIC_IRQ(irq) && cfg && !cfg->vector) {
2705                         /*
2706                          * Hmm.. We don't have an entry for this,
2707                          * so default to an old-fashioned 8259
2708                          * interrupt if we can..
2709                          */
2710                         if (irq < NR_IRQS_LEGACY)
2711                                 make_8259A_irq(irq);
2712                         else
2713                                 /* Strange. Oh, well.. */
2714                                 desc->chip = &no_irq_chip;
2715                 }
2716         }
2717 }
2718
2719 /*
2720  * The local APIC irq-chip implementation:
2721  */
2722
2723 static void mask_lapic_irq(unsigned int irq)
2724 {
2725         unsigned long v;
2726
2727         v = apic_read(APIC_LVT0);
2728         apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
2729 }
2730
2731 static void unmask_lapic_irq(unsigned int irq)
2732 {
2733         unsigned long v;
2734
2735         v = apic_read(APIC_LVT0);
2736         apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
2737 }
2738
2739 static void ack_lapic_irq(unsigned int irq)
2740 {
2741         ack_APIC_irq();
2742 }
2743
2744 static struct irq_chip lapic_chip __read_mostly = {
2745         .name           = "local-APIC",
2746         .mask           = mask_lapic_irq,
2747         .unmask         = unmask_lapic_irq,
2748         .ack            = ack_lapic_irq,
2749 };
2750
2751 static void lapic_register_intr(int irq, struct irq_desc *desc)
2752 {
2753         desc->status &= ~IRQ_LEVEL;
2754         set_irq_chip_and_handler_name(irq, &lapic_chip, handle_edge_irq,
2755                                       "edge");
2756 }
2757
2758 static void __init setup_nmi(void)
2759 {
2760         /*
2761          * Dirty trick to enable the NMI watchdog ...
2762          * We put the 8259A master into AEOI mode and
2763          * unmask on all local APICs LVT0 as NMI.
2764          *
2765          * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
2766          * is from Maciej W. Rozycki - so we do not have to EOI from
2767          * the NMI handler or the timer interrupt.
2768          */
2769         apic_printk(APIC_VERBOSE, KERN_INFO "activating NMI Watchdog ...");
2770
2771         enable_NMI_through_LVT0();
2772
2773         apic_printk(APIC_VERBOSE, " done.\n");
2774 }
2775
2776 /*
2777  * This looks a bit hackish but it's about the only one way of sending
2778  * a few INTA cycles to 8259As and any associated glue logic.  ICR does
2779  * not support the ExtINT mode, unfortunately.  We need to send these
2780  * cycles as some i82489DX-based boards have glue logic that keeps the
2781  * 8259A interrupt line asserted until INTA.  --macro
2782  */
2783 static inline void __init unlock_ExtINT_logic(void)
2784 {
2785         int apic, pin, i;
2786         struct IO_APIC_route_entry entry0, entry1;
2787         unsigned char save_control, save_freq_select;
2788
2789         pin  = find_isa_irq_pin(8, mp_INT);
2790         if (pin == -1) {
2791                 WARN_ON_ONCE(1);
2792                 return;
2793         }
2794         apic = find_isa_irq_apic(8, mp_INT);
2795         if (apic == -1) {
2796                 WARN_ON_ONCE(1);
2797                 return;
2798         }
2799
2800         entry0 = ioapic_read_entry(apic, pin);
2801         clear_IO_APIC_pin(apic, pin);
2802
2803         memset(&entry1, 0, sizeof(entry1));
2804
2805         entry1.dest_mode = 0;                   /* physical delivery */
2806         entry1.mask = 0;                        /* unmask IRQ now */
2807         entry1.dest = hard_smp_processor_id();
2808         entry1.delivery_mode = dest_ExtINT;
2809         entry1.polarity = entry0.polarity;
2810         entry1.trigger = 0;
2811         entry1.vector = 0;
2812
2813         ioapic_write_entry(apic, pin, entry1);
2814
2815         save_control = CMOS_READ(RTC_CONTROL);
2816         save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
2817         CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
2818                    RTC_FREQ_SELECT);
2819         CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
2820
2821         i = 100;
2822         while (i-- > 0) {
2823                 mdelay(10);
2824                 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
2825                         i -= 10;
2826         }
2827
2828         CMOS_WRITE(save_control, RTC_CONTROL);
2829         CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
2830         clear_IO_APIC_pin(apic, pin);
2831
2832         ioapic_write_entry(apic, pin, entry0);
2833 }
2834
2835 static int disable_timer_pin_1 __initdata;
2836 /* Actually the next is obsolete, but keep it for paranoid reasons -AK */
2837 static int __init disable_timer_pin_setup(char *arg)
2838 {
2839         disable_timer_pin_1 = 1;
2840         return 0;
2841 }
2842 early_param("disable_timer_pin_1", disable_timer_pin_setup);
2843
2844 int timer_through_8259 __initdata;
2845
2846 /*
2847  * This code may look a bit paranoid, but it's supposed to cooperate with
2848  * a wide range of boards and BIOS bugs.  Fortunately only the timer IRQ
2849  * is so screwy.  Thanks to Brian Perkins for testing/hacking this beast
2850  * fanatically on his truly buggy board.
2851  *
2852  * FIXME: really need to revamp this for all platforms.
2853  */
2854 static inline void __init check_timer(void)
2855 {
2856         struct irq_desc *desc = irq_to_desc(0);
2857         struct irq_cfg *cfg = desc->chip_data;
2858         int node = cpu_to_node(boot_cpu_id);
2859         int apic1, pin1, apic2, pin2;
2860         unsigned long flags;
2861         int no_pin1 = 0;
2862
2863         local_irq_save(flags);
2864
2865         /*
2866          * get/set the timer IRQ vector:
2867          */
2868         disable_8259A_irq(0);
2869         assign_irq_vector(0, cfg, apic->target_cpus());
2870
2871         /*
2872          * As IRQ0 is to be enabled in the 8259A, the virtual
2873          * wire has to be disabled in the local APIC.  Also
2874          * timer interrupts need to be acknowledged manually in
2875          * the 8259A for the i82489DX when using the NMI
2876          * watchdog as that APIC treats NMIs as level-triggered.
2877          * The AEOI mode will finish them in the 8259A
2878          * automatically.
2879          */
2880         apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
2881         init_8259A(1);
2882 #ifdef CONFIG_X86_32
2883         {
2884                 unsigned int ver;
2885
2886                 ver = apic_read(APIC_LVR);
2887                 ver = GET_APIC_VERSION(ver);
2888                 timer_ack = (nmi_watchdog == NMI_IO_APIC && !APIC_INTEGRATED(ver));
2889         }
2890 #endif
2891
2892         pin1  = find_isa_irq_pin(0, mp_INT);
2893         apic1 = find_isa_irq_apic(0, mp_INT);
2894         pin2  = ioapic_i8259.pin;
2895         apic2 = ioapic_i8259.apic;
2896
2897         apic_printk(APIC_QUIET, KERN_INFO "..TIMER: vector=0x%02X "
2898                     "apic1=%d pin1=%d apic2=%d pin2=%d\n",
2899                     cfg->vector, apic1, pin1, apic2, pin2);
2900
2901         /*
2902          * Some BIOS writers are clueless and report the ExtINTA
2903          * I/O APIC input from the cascaded 8259A as the timer
2904          * interrupt input.  So just in case, if only one pin
2905          * was found above, try it both directly and through the
2906          * 8259A.
2907          */
2908         if (pin1 == -1) {
2909                 if (intr_remapping_enabled)
2910                         panic("BIOS bug: timer not connected to IO-APIC");
2911                 pin1 = pin2;
2912                 apic1 = apic2;
2913                 no_pin1 = 1;
2914         } else if (pin2 == -1) {
2915                 pin2 = pin1;
2916                 apic2 = apic1;
2917         }
2918
2919         if (pin1 != -1) {
2920                 /*
2921                  * Ok, does IRQ0 through the IOAPIC work?
2922                  */
2923                 if (no_pin1) {
2924                         add_pin_to_irq_node(cfg, node, apic1, pin1);
2925                         setup_timer_IRQ0_pin(apic1, pin1, cfg->vector);
2926                 } else {
2927                         /* for edge trigger, setup_IO_APIC_irq already
2928                          * leave it unmasked.
2929                          * so only need to unmask if it is level-trigger
2930                          * do we really have level trigger timer?
2931                          */
2932                         int idx;
2933                         idx = find_irq_entry(apic1, pin1, mp_INT);
2934                         if (idx != -1 && irq_trigger(idx))
2935                                 unmask_IO_APIC_irq_desc(desc);
2936                 }
2937                 if (timer_irq_works()) {
2938                         if (nmi_watchdog == NMI_IO_APIC) {
2939                                 setup_nmi();
2940                                 enable_8259A_irq(0);
2941                         }
2942                         if (disable_timer_pin_1 > 0)
2943                                 clear_IO_APIC_pin(0, pin1);
2944                         goto out;
2945                 }
2946                 if (intr_remapping_enabled)
2947                         panic("timer doesn't work through Interrupt-remapped IO-APIC");
2948                 local_irq_disable();
2949                 clear_IO_APIC_pin(apic1, pin1);
2950                 if (!no_pin1)
2951                         apic_printk(APIC_QUIET, KERN_ERR "..MP-BIOS bug: "
2952                                     "8254 timer not connected to IO-APIC\n");
2953
2954                 apic_printk(APIC_QUIET, KERN_INFO "...trying to set up timer "
2955                             "(IRQ0) through the 8259A ...\n");
2956                 apic_printk(APIC_QUIET, KERN_INFO
2957                             "..... (found apic %d pin %d) ...\n", apic2, pin2);
2958                 /*
2959                  * legacy devices should be connected to IO APIC #0
2960                  */
2961                 replace_pin_at_irq_node(cfg, node, apic1, pin1, apic2, pin2);
2962                 setup_timer_IRQ0_pin(apic2, pin2, cfg->vector);
2963                 enable_8259A_irq(0);
2964                 if (timer_irq_works()) {
2965                         apic_printk(APIC_QUIET, KERN_INFO "....... works.\n");
2966                         timer_through_8259 = 1;
2967                         if (nmi_watchdog == NMI_IO_APIC) {
2968                                 disable_8259A_irq(0);
2969                                 setup_nmi();
2970                                 enable_8259A_irq(0);
2971                         }
2972                         goto out;
2973                 }
2974                 /*
2975                  * Cleanup, just in case ...
2976                  */
2977                 local_irq_disable();
2978                 disable_8259A_irq(0);
2979                 clear_IO_APIC_pin(apic2, pin2);
2980                 apic_printk(APIC_QUIET, KERN_INFO "....... failed.\n");
2981         }
2982
2983         if (nmi_watchdog == NMI_IO_APIC) {
2984                 apic_printk(APIC_QUIET, KERN_WARNING "timer doesn't work "
2985                             "through the IO-APIC - disabling NMI Watchdog!\n");
2986                 nmi_watchdog = NMI_NONE;
2987         }
2988 #ifdef CONFIG_X86_32
2989         timer_ack = 0;
2990 #endif
2991
2992         apic_printk(APIC_QUIET, KERN_INFO
2993                     "...trying to set up timer as Virtual Wire IRQ...\n");
2994
2995         lapic_register_intr(0, desc);
2996         apic_write(APIC_LVT0, APIC_DM_FIXED | cfg->vector);     /* Fixed mode */
2997         enable_8259A_irq(0);
2998
2999         if (timer_irq_works()) {
3000                 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
3001                 goto out;
3002         }
3003         local_irq_disable();
3004         disable_8259A_irq(0);
3005         apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | cfg->vector);
3006         apic_printk(APIC_QUIET, KERN_INFO "..... failed.\n");
3007
3008         apic_printk(APIC_QUIET, KERN_INFO
3009                     "...trying to set up timer as ExtINT IRQ...\n");
3010
3011         init_8259A(0);
3012         make_8259A_irq(0);
3013         apic_write(APIC_LVT0, APIC_DM_EXTINT);
3014
3015         unlock_ExtINT_logic();
3016
3017         if (timer_irq_works()) {
3018                 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
3019                 goto out;
3020         }
3021         local_irq_disable();
3022         apic_printk(APIC_QUIET, KERN_INFO "..... failed :(.\n");
3023         panic("IO-APIC + timer doesn't work!  Boot with apic=debug and send a "
3024                 "report.  Then try booting with the 'noapic' option.\n");
3025 out:
3026         local_irq_restore(flags);
3027 }
3028
3029 /*
3030  * Traditionally ISA IRQ2 is the cascade IRQ, and is not available
3031  * to devices.  However there may be an I/O APIC pin available for
3032  * this interrupt regardless.  The pin may be left unconnected, but
3033  * typically it will be reused as an ExtINT cascade interrupt for
3034  * the master 8259A.  In the MPS case such a pin will normally be
3035  * reported as an ExtINT interrupt in the MP table.  With ACPI
3036  * there is no provision for ExtINT interrupts, and in the absence
3037  * of an override it would be treated as an ordinary ISA I/O APIC
3038  * interrupt, that is edge-triggered and unmasked by default.  We
3039  * used to do this, but it caused problems on some systems because
3040  * of the NMI watchdog and sometimes IRQ0 of the 8254 timer using
3041  * the same ExtINT cascade interrupt to drive the local APIC of the
3042  * bootstrap processor.  Therefore we refrain from routing IRQ2 to
3043  * the I/O APIC in all cases now.  No actual device should request
3044  * it anyway.  --macro
3045  */
3046 #define PIC_IRQS        (1 << PIC_CASCADE_IR)
3047
3048 void __init setup_IO_APIC(void)
3049 {
3050
3051         /*
3052          * calling enable_IO_APIC() is moved to setup_local_APIC for BP
3053          */
3054
3055         io_apic_irqs = ~PIC_IRQS;
3056
3057         apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
3058         /*
3059          * Set up IO-APIC IRQ routing.
3060          */
3061 #ifdef CONFIG_X86_32
3062         if (!acpi_ioapic)
3063                 setup_ioapic_ids_from_mpc();
3064 #endif
3065         sync_Arb_IDs();
3066         setup_IO_APIC_irqs();
3067         init_IO_APIC_traps();
3068         check_timer();
3069 }
3070
3071 /*
3072  *      Called after all the initialization is done. If we didnt find any
3073  *      APIC bugs then we can allow the modify fast path
3074  */
3075
3076 static int __init io_apic_bug_finalize(void)
3077 {
3078         if (sis_apic_bug == -1)
3079                 sis_apic_bug = 0;
3080         return 0;
3081 }
3082
3083 late_initcall(io_apic_bug_finalize);
3084
3085 struct sysfs_ioapic_data {
3086         struct sys_device dev;
3087         struct IO_APIC_route_entry entry[0];
3088 };
3089 static struct sysfs_ioapic_data * mp_ioapic_data[MAX_IO_APICS];
3090
3091 static int ioapic_suspend(struct sys_device *dev, pm_message_t state)
3092 {
3093         struct IO_APIC_route_entry *entry;
3094         struct sysfs_ioapic_data *data;
3095         int i;
3096
3097         data = container_of(dev, struct sysfs_ioapic_data, dev);
3098         entry = data->entry;
3099         for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ )
3100                 *entry = ioapic_read_entry(dev->id, i);
3101
3102         return 0;
3103 }
3104
3105 static int ioapic_resume(struct sys_device *dev)
3106 {
3107         struct IO_APIC_route_entry *entry;
3108         struct sysfs_ioapic_data *data;
3109         unsigned long flags;
3110         union IO_APIC_reg_00 reg_00;
3111         int i;
3112
3113         data = container_of(dev, struct sysfs_ioapic_data, dev);
3114         entry = data->entry;
3115
3116         spin_lock_irqsave(&ioapic_lock, flags);
3117         reg_00.raw = io_apic_read(dev->id, 0);
3118         if (reg_00.bits.ID != mp_ioapics[dev->id].apicid) {
3119                 reg_00.bits.ID = mp_ioapics[dev->id].apicid;
3120                 io_apic_write(dev->id, 0, reg_00.raw);
3121         }
3122         spin_unlock_irqrestore(&ioapic_lock, flags);
3123         for (i = 0; i < nr_ioapic_registers[dev->id]; i++)
3124                 ioapic_write_entry(dev->id, i, entry[i]);
3125
3126         return 0;
3127 }
3128
3129 static struct sysdev_class ioapic_sysdev_class = {
3130         .name = "ioapic",
3131         .suspend = ioapic_suspend,
3132         .resume = ioapic_resume,
3133 };
3134
3135 static int __init ioapic_init_sysfs(void)
3136 {
3137         struct sys_device * dev;
3138         int i, size, error;
3139
3140         error = sysdev_class_register(&ioapic_sysdev_class);
3141         if (error)
3142                 return error;
3143
3144         for (i = 0; i < nr_ioapics; i++ ) {
3145                 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
3146                         * sizeof(struct IO_APIC_route_entry);
3147                 mp_ioapic_data[i] = kzalloc(size, GFP_KERNEL);
3148                 if (!mp_ioapic_data[i]) {
3149                         printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
3150                         continue;
3151                 }
3152                 dev = &mp_ioapic_data[i]->dev;
3153                 dev->id = i;
3154                 dev->cls = &ioapic_sysdev_class;
3155                 error = sysdev_register(dev);
3156                 if (error) {
3157                         kfree(mp_ioapic_data[i]);
3158                         mp_ioapic_data[i] = NULL;
3159                         printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
3160                         continue;
3161                 }
3162         }
3163
3164         return 0;
3165 }
3166
3167 device_initcall(ioapic_init_sysfs);
3168
3169 static int nr_irqs_gsi = NR_IRQS_LEGACY;
3170 /*
3171  * Dynamic irq allocate and deallocation
3172  */
3173 unsigned int create_irq_nr(unsigned int irq_want, int node)
3174 {
3175         /* Allocate an unused irq */
3176         unsigned int irq;
3177         unsigned int new;
3178         unsigned long flags;
3179         struct irq_cfg *cfg_new = NULL;
3180         struct irq_desc *desc_new = NULL;
3181
3182         irq = 0;
3183         if (irq_want < nr_irqs_gsi)
3184                 irq_want = nr_irqs_gsi;
3185
3186         spin_lock_irqsave(&vector_lock, flags);
3187         for (new = irq_want; new < nr_irqs; new++) {
3188                 desc_new = irq_to_desc_alloc_node(new, node);
3189                 if (!desc_new) {
3190                         printk(KERN_INFO "can not get irq_desc for %d\n", new);
3191                         continue;
3192                 }
3193                 cfg_new = desc_new->chip_data;
3194
3195                 if (cfg_new->vector != 0)
3196                         continue;
3197
3198                 desc_new = move_irq_desc(desc_new, node);
3199
3200                 if (__assign_irq_vector(new, cfg_new, apic->target_cpus()) == 0)
3201                         irq = new;
3202                 break;
3203         }
3204         spin_unlock_irqrestore(&vector_lock, flags);
3205
3206         if (irq > 0) {
3207                 dynamic_irq_init(irq);
3208                 /* restore it, in case dynamic_irq_init clear it */
3209                 if (desc_new)
3210                         desc_new->chip_data = cfg_new;
3211         }
3212         return irq;
3213 }
3214
3215 int create_irq(void)
3216 {
3217         int node = cpu_to_node(boot_cpu_id);
3218         unsigned int irq_want;
3219         int irq;
3220
3221         irq_want = nr_irqs_gsi;
3222         irq = create_irq_nr(irq_want, node);
3223
3224         if (irq == 0)
3225                 irq = -1;
3226
3227         return irq;
3228 }
3229
3230 void destroy_irq(unsigned int irq)
3231 {
3232         unsigned long flags;
3233         struct irq_cfg *cfg;
3234         struct irq_desc *desc;
3235
3236         /* store it, in case dynamic_irq_cleanup clear it */
3237         desc = irq_to_desc(irq);
3238         cfg = desc->chip_data;
3239         dynamic_irq_cleanup(irq);
3240         /* connect back irq_cfg */
3241         if (desc)
3242                 desc->chip_data = cfg;
3243
3244         free_irte(irq);
3245         spin_lock_irqsave(&vector_lock, flags);
3246         __clear_irq_vector(irq, cfg);
3247         spin_unlock_irqrestore(&vector_lock, flags);
3248 }
3249
3250 /*
3251  * MSI message composition
3252  */
3253 #ifdef CONFIG_PCI_MSI
3254 static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq, struct msi_msg *msg)
3255 {
3256         struct irq_cfg *cfg;
3257         int err;
3258         unsigned dest;
3259
3260         if (disable_apic)
3261                 return -ENXIO;
3262
3263         cfg = irq_cfg(irq);
3264         err = assign_irq_vector(irq, cfg, apic->target_cpus());
3265         if (err)
3266                 return err;
3267
3268         dest = apic->cpu_mask_to_apicid_and(cfg->domain, apic->target_cpus());
3269
3270         if (irq_remapped(irq)) {
3271                 struct irte irte;
3272                 int ir_index;
3273                 u16 sub_handle;
3274
3275                 ir_index = map_irq_to_irte_handle(irq, &sub_handle);
3276                 BUG_ON(ir_index == -1);
3277
3278                 memset (&irte, 0, sizeof(irte));
3279
3280                 irte.present = 1;
3281                 irte.dst_mode = apic->irq_dest_mode;
3282                 irte.trigger_mode = 0; /* edge */
3283                 irte.dlvry_mode = apic->irq_delivery_mode;
3284                 irte.vector = cfg->vector;
3285                 irte.dest_id = IRTE_DEST(dest);
3286
3287                 modify_irte(irq, &irte);
3288
3289                 msg->address_hi = MSI_ADDR_BASE_HI;
3290                 msg->data = sub_handle;
3291                 msg->address_lo = MSI_ADDR_BASE_LO | MSI_ADDR_IR_EXT_INT |
3292                                   MSI_ADDR_IR_SHV |
3293                                   MSI_ADDR_IR_INDEX1(ir_index) |
3294                                   MSI_ADDR_IR_INDEX2(ir_index);
3295         } else {
3296                 if (x2apic_enabled())
3297                         msg->address_hi = MSI_ADDR_BASE_HI |
3298                                           MSI_ADDR_EXT_DEST_ID(dest);
3299                 else
3300                         msg->address_hi = MSI_ADDR_BASE_HI;
3301
3302                 msg->address_lo =
3303                         MSI_ADDR_BASE_LO |
3304                         ((apic->irq_dest_mode == 0) ?
3305                                 MSI_ADDR_DEST_MODE_PHYSICAL:
3306                                 MSI_ADDR_DEST_MODE_LOGICAL) |
3307                         ((apic->irq_delivery_mode != dest_LowestPrio) ?
3308                                 MSI_ADDR_REDIRECTION_CPU:
3309                                 MSI_ADDR_REDIRECTION_LOWPRI) |
3310                         MSI_ADDR_DEST_ID(dest);
3311
3312                 msg->data =
3313                         MSI_DATA_TRIGGER_EDGE |
3314                         MSI_DATA_LEVEL_ASSERT |
3315                         ((apic->irq_delivery_mode != dest_LowestPrio) ?
3316                                 MSI_DATA_DELIVERY_FIXED:
3317                                 MSI_DATA_DELIVERY_LOWPRI) |
3318                         MSI_DATA_VECTOR(cfg->vector);
3319         }
3320         return err;
3321 }
3322
3323 #ifdef CONFIG_SMP
3324 static int set_msi_irq_affinity(unsigned int irq, const struct cpumask *mask)
3325 {
3326         struct irq_desc *desc = irq_to_desc(irq);
3327         struct irq_cfg *cfg;
3328         struct msi_msg msg;
3329         unsigned int dest;
3330
3331         dest = set_desc_affinity(desc, mask);
3332         if (dest == BAD_APICID)
3333                 return -1;
3334
3335         cfg = desc->chip_data;
3336
3337         read_msi_msg_desc(desc, &msg);
3338
3339         msg.data &= ~MSI_DATA_VECTOR_MASK;
3340         msg.data |= MSI_DATA_VECTOR(cfg->vector);
3341         msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3342         msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3343
3344         write_msi_msg_desc(desc, &msg);
3345
3346         return 0;
3347 }
3348 #ifdef CONFIG_INTR_REMAP
3349 /*
3350  * Migrate the MSI irq to another cpumask. This migration is
3351  * done in the process context using interrupt-remapping hardware.
3352  */
3353 static int
3354 ir_set_msi_irq_affinity(unsigned int irq, const struct cpumask *mask)
3355 {
3356         struct irq_desc *desc = irq_to_desc(irq);
3357         struct irq_cfg *cfg = desc->chip_data;
3358         unsigned int dest;
3359         struct irte irte;
3360
3361         if (get_irte(irq, &irte))
3362                 return -1;
3363
3364         dest = set_desc_affinity(desc, mask);
3365         if (dest == BAD_APICID)
3366                 return -1;
3367
3368         irte.vector = cfg->vector;
3369         irte.dest_id = IRTE_DEST(dest);
3370
3371         /*
3372          * atomically update the IRTE with the new destination and vector.
3373          */
3374         modify_irte(irq, &irte);
3375
3376         /*
3377          * After this point, all the interrupts will start arriving
3378          * at the new destination. So, time to cleanup the previous
3379          * vector allocation.
3380          */
3381         if (cfg->move_in_progress)
3382                 send_cleanup_vector(cfg);
3383
3384         return 0;
3385 }
3386
3387 #endif
3388 #endif /* CONFIG_SMP */
3389
3390 /*
3391  * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
3392  * which implement the MSI or MSI-X Capability Structure.
3393  */
3394 static struct irq_chip msi_chip = {
3395         .name           = "PCI-MSI",
3396         .unmask         = unmask_msi_irq,
3397         .mask           = mask_msi_irq,
3398         .ack            = ack_apic_edge,
3399 #ifdef CONFIG_SMP
3400         .set_affinity   = set_msi_irq_affinity,
3401 #endif
3402         .retrigger      = ioapic_retrigger_irq,
3403 };
3404
3405 static struct irq_chip msi_ir_chip = {
3406         .name           = "IR-PCI-MSI",
3407         .unmask         = unmask_msi_irq,
3408         .mask           = mask_msi_irq,
3409 #ifdef CONFIG_INTR_REMAP
3410         .ack            = ir_ack_apic_edge,
3411 #ifdef CONFIG_SMP
3412         .set_affinity   = ir_set_msi_irq_affinity,
3413 #endif
3414 #endif
3415         .retrigger      = ioapic_retrigger_irq,
3416 };
3417
3418 /*
3419  * Map the PCI dev to the corresponding remapping hardware unit
3420  * and allocate 'nvec' consecutive interrupt-remapping table entries
3421  * in it.
3422  */
3423 static int msi_alloc_irte(struct pci_dev *dev, int irq, int nvec)
3424 {
3425         struct intel_iommu *iommu;
3426         int index;
3427
3428         iommu = map_dev_to_ir(dev);
3429         if (!iommu) {
3430                 printk(KERN_ERR
3431                        "Unable to map PCI %s to iommu\n", pci_name(dev));
3432                 return -ENOENT;
3433         }
3434
3435         index = alloc_irte(iommu, irq, nvec);
3436         if (index < 0) {
3437                 printk(KERN_ERR
3438                        "Unable to allocate %d IRTE for PCI %s\n", nvec,
3439                        pci_name(dev));
3440                 return -ENOSPC;
3441         }
3442         return index;
3443 }
3444
3445 static int setup_msi_irq(struct pci_dev *dev, struct msi_desc *msidesc, int irq)
3446 {
3447         int ret;
3448         struct msi_msg msg;
3449
3450         ret = msi_compose_msg(dev, irq, &msg);
3451         if (ret < 0)
3452                 return ret;
3453
3454         set_irq_msi(irq, msidesc);
3455         write_msi_msg(irq, &msg);
3456
3457         if (irq_remapped(irq)) {
3458                 struct irq_desc *desc = irq_to_desc(irq);
3459                 /*
3460                  * irq migration in process context
3461                  */
3462                 desc->status |= IRQ_MOVE_PCNTXT;
3463                 set_irq_chip_and_handler_name(irq, &msi_ir_chip, handle_edge_irq, "edge");
3464         } else
3465                 set_irq_chip_and_handler_name(irq, &msi_chip, handle_edge_irq, "edge");
3466
3467         dev_printk(KERN_DEBUG, &dev->dev, "irq %d for MSI/MSI-X\n", irq);
3468
3469         return 0;
3470 }
3471
3472 int arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
3473 {
3474         unsigned int irq;
3475         int ret, sub_handle;
3476         struct msi_desc *msidesc;
3477         unsigned int irq_want;
3478         struct intel_iommu *iommu = NULL;
3479         int index = 0;
3480         int node;
3481
3482         /* x86 doesn't support multiple MSI yet */
3483         if (type == PCI_CAP_ID_MSI && nvec > 1)
3484                 return 1;
3485
3486         node = dev_to_node(&dev->dev);
3487         irq_want = nr_irqs_gsi;
3488         sub_handle = 0;
3489         list_for_each_entry(msidesc, &dev->msi_list, list) {
3490                 irq = create_irq_nr(irq_want, node);
3491                 if (irq == 0)
3492                         return -1;
3493                 irq_want = irq + 1;
3494                 if (!intr_remapping_enabled)
3495                         goto no_ir;
3496
3497                 if (!sub_handle) {
3498                         /*
3499                          * allocate the consecutive block of IRTE's
3500                          * for 'nvec'
3501                          */
3502                         index = msi_alloc_irte(dev, irq, nvec);
3503                         if (index < 0) {
3504                                 ret = index;
3505                                 goto error;
3506                         }
3507                 } else {
3508                         iommu = map_dev_to_ir(dev);
3509                         if (!iommu) {
3510                                 ret = -ENOENT;
3511                                 goto error;
3512                         }
3513                         /*
3514                          * setup the mapping between the irq and the IRTE
3515                          * base index, the sub_handle pointing to the
3516                          * appropriate interrupt remap table entry.
3517                          */
3518                         set_irte_irq(irq, iommu, index, sub_handle);
3519                 }
3520 no_ir:
3521                 ret = setup_msi_irq(dev, msidesc, irq);
3522                 if (ret < 0)
3523                         goto error;
3524                 sub_handle++;
3525         }
3526         return 0;
3527
3528 error:
3529         destroy_irq(irq);
3530         return ret;
3531 }
3532
3533 void arch_teardown_msi_irq(unsigned int irq)
3534 {
3535         destroy_irq(irq);
3536 }
3537
3538 #if defined (CONFIG_DMAR) || defined (CONFIG_INTR_REMAP)
3539 #ifdef CONFIG_SMP
3540 static int dmar_msi_set_affinity(unsigned int irq, const struct cpumask *mask)
3541 {
3542         struct irq_desc *desc = irq_to_desc(irq);
3543         struct irq_cfg *cfg;
3544         struct msi_msg msg;
3545         unsigned int dest;
3546
3547         dest = set_desc_affinity(desc, mask);
3548         if (dest == BAD_APICID)
3549                 return -1;
3550
3551         cfg = desc->chip_data;
3552
3553         dmar_msi_read(irq, &msg);
3554
3555         msg.data &= ~MSI_DATA_VECTOR_MASK;
3556         msg.data |= MSI_DATA_VECTOR(cfg->vector);
3557         msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3558         msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3559
3560         dmar_msi_write(irq, &msg);
3561
3562         return 0;
3563 }
3564
3565 #endif /* CONFIG_SMP */
3566
3567 struct irq_chip dmar_msi_type = {
3568         .name = "DMAR_MSI",
3569         .unmask = dmar_msi_unmask,
3570         .mask = dmar_msi_mask,
3571         .ack = ack_apic_edge,
3572 #ifdef CONFIG_SMP
3573         .set_affinity = dmar_msi_set_affinity,
3574 #endif
3575         .retrigger = ioapic_retrigger_irq,
3576 };
3577
3578 int arch_setup_dmar_msi(unsigned int irq)
3579 {
3580         int ret;
3581         struct msi_msg msg;
3582
3583         ret = msi_compose_msg(NULL, irq, &msg);
3584         if (ret < 0)
3585                 return ret;
3586         dmar_msi_write(irq, &msg);
3587         set_irq_chip_and_handler_name(irq, &dmar_msi_type, handle_edge_irq,
3588                 "edge");
3589         return 0;
3590 }
3591 #endif
3592
3593 #ifdef CONFIG_HPET_TIMER
3594
3595 #ifdef CONFIG_SMP
3596 static int hpet_msi_set_affinity(unsigned int irq, const struct cpumask *mask)
3597 {
3598         struct irq_desc *desc = irq_to_desc(irq);
3599         struct irq_cfg *cfg;
3600         struct msi_msg msg;
3601         unsigned int dest;
3602
3603         dest = set_desc_affinity(desc, mask);
3604         if (dest == BAD_APICID)
3605                 return -1;
3606
3607         cfg = desc->chip_data;
3608
3609         hpet_msi_read(irq, &msg);
3610
3611         msg.data &= ~MSI_DATA_VECTOR_MASK;
3612         msg.data |= MSI_DATA_VECTOR(cfg->vector);
3613         msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3614         msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3615
3616         hpet_msi_write(irq, &msg);
3617
3618         return 0;
3619 }
3620
3621 #endif /* CONFIG_SMP */
3622
3623 static struct irq_chip hpet_msi_type = {
3624         .name = "HPET_MSI",
3625         .unmask = hpet_msi_unmask,
3626         .mask = hpet_msi_mask,
3627         .ack = ack_apic_edge,
3628 #ifdef CONFIG_SMP
3629         .set_affinity = hpet_msi_set_affinity,
3630 #endif
3631         .retrigger = ioapic_retrigger_irq,
3632 };
3633
3634 int arch_setup_hpet_msi(unsigned int irq)
3635 {
3636         int ret;
3637         struct msi_msg msg;
3638         struct irq_desc *desc = irq_to_desc(irq);
3639
3640         ret = msi_compose_msg(NULL, irq, &msg);
3641         if (ret < 0)
3642                 return ret;
3643
3644         hpet_msi_write(irq, &msg);
3645         desc->status |= IRQ_MOVE_PCNTXT;
3646         set_irq_chip_and_handler_name(irq, &hpet_msi_type, handle_edge_irq,
3647                 "edge");
3648
3649         return 0;
3650 }
3651 #endif
3652
3653 #endif /* CONFIG_PCI_MSI */
3654 /*
3655  * Hypertransport interrupt support
3656  */
3657 #ifdef CONFIG_HT_IRQ
3658
3659 #ifdef CONFIG_SMP
3660
3661 static void target_ht_irq(unsigned int irq, unsigned int dest, u8 vector)
3662 {
3663         struct ht_irq_msg msg;
3664         fetch_ht_irq_msg(irq, &msg);
3665
3666         msg.address_lo &= ~(HT_IRQ_LOW_VECTOR_MASK | HT_IRQ_LOW_DEST_ID_MASK);
3667         msg.address_hi &= ~(HT_IRQ_HIGH_DEST_ID_MASK);
3668
3669         msg.address_lo |= HT_IRQ_LOW_VECTOR(vector) | HT_IRQ_LOW_DEST_ID(dest);
3670         msg.address_hi |= HT_IRQ_HIGH_DEST_ID(dest);
3671
3672         write_ht_irq_msg(irq, &msg);
3673 }
3674
3675 static int set_ht_irq_affinity(unsigned int irq, const struct cpumask *mask)
3676 {
3677         struct irq_desc *desc = irq_to_desc(irq);
3678         struct irq_cfg *cfg;
3679         unsigned int dest;
3680
3681         dest = set_desc_affinity(desc, mask);
3682         if (dest == BAD_APICID)
3683                 return -1;
3684
3685         cfg = desc->chip_data;
3686
3687         target_ht_irq(irq, dest, cfg->vector);
3688
3689         return 0;
3690 }
3691
3692 #endif
3693
3694 static struct irq_chip ht_irq_chip = {
3695         .name           = "PCI-HT",
3696         .mask           = mask_ht_irq,
3697         .unmask         = unmask_ht_irq,
3698         .ack            = ack_apic_edge,
3699 #ifdef CONFIG_SMP
3700         .set_affinity   = set_ht_irq_affinity,
3701 #endif
3702         .retrigger      = ioapic_retrigger_irq,
3703 };
3704
3705 int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev)
3706 {
3707         struct irq_cfg *cfg;
3708         int err;
3709
3710         if (disable_apic)
3711                 return -ENXIO;
3712
3713         cfg = irq_cfg(irq);
3714         err = assign_irq_vector(irq, cfg, apic->target_cpus());
3715         if (!err) {
3716                 struct ht_irq_msg msg;
3717                 unsigned dest;
3718
3719                 dest = apic->cpu_mask_to_apicid_and(cfg->domain,
3720                                                     apic->target_cpus());
3721
3722                 msg.address_hi = HT_IRQ_HIGH_DEST_ID(dest);
3723
3724                 msg.address_lo =
3725                         HT_IRQ_LOW_BASE |
3726                         HT_IRQ_LOW_DEST_ID(dest) |
3727                         HT_IRQ_LOW_VECTOR(cfg->vector) |
3728                         ((apic->irq_dest_mode == 0) ?
3729                                 HT_IRQ_LOW_DM_PHYSICAL :
3730                                 HT_IRQ_LOW_DM_LOGICAL) |
3731                         HT_IRQ_LOW_RQEOI_EDGE |
3732                         ((apic->irq_delivery_mode != dest_LowestPrio) ?
3733                                 HT_IRQ_LOW_MT_FIXED :
3734                                 HT_IRQ_LOW_MT_ARBITRATED) |
3735                         HT_IRQ_LOW_IRQ_MASKED;
3736
3737                 write_ht_irq_msg(irq, &msg);
3738
3739                 set_irq_chip_and_handler_name(irq, &ht_irq_chip,
3740                                               handle_edge_irq, "edge");
3741
3742                 dev_printk(KERN_DEBUG, &dev->dev, "irq %d for HT\n", irq);
3743         }
3744         return err;
3745 }
3746 #endif /* CONFIG_HT_IRQ */
3747
3748 #ifdef CONFIG_X86_UV
3749 /*
3750  * Re-target the irq to the specified CPU and enable the specified MMR located
3751  * on the specified blade to allow the sending of MSIs to the specified CPU.
3752  */
3753 int arch_enable_uv_irq(char *irq_name, unsigned int irq, int cpu, int mmr_blade,
3754                        unsigned long mmr_offset)
3755 {
3756         const struct cpumask *eligible_cpu = cpumask_of(cpu);
3757         struct irq_cfg *cfg;
3758         int mmr_pnode;
3759         unsigned long mmr_value;
3760         struct uv_IO_APIC_route_entry *entry;
3761         unsigned long flags;
3762         int err;
3763
3764         BUILD_BUG_ON(sizeof(struct uv_IO_APIC_route_entry) != sizeof(unsigned long));
3765
3766         cfg = irq_cfg(irq);
3767
3768         err = assign_irq_vector(irq, cfg, eligible_cpu);
3769         if (err != 0)
3770                 return err;
3771
3772         spin_lock_irqsave(&vector_lock, flags);
3773         set_irq_chip_and_handler_name(irq, &uv_irq_chip, handle_percpu_irq,
3774                                       irq_name);
3775         spin_unlock_irqrestore(&vector_lock, flags);
3776
3777         mmr_value = 0;
3778         entry = (struct uv_IO_APIC_route_entry *)&mmr_value;
3779         entry->vector           = cfg->vector;
3780         entry->delivery_mode    = apic->irq_delivery_mode;
3781         entry->dest_mode        = apic->irq_dest_mode;
3782         entry->polarity         = 0;
3783         entry->trigger          = 0;
3784         entry->mask             = 0;
3785         entry->dest             = apic->cpu_mask_to_apicid(eligible_cpu);
3786
3787         mmr_pnode = uv_blade_to_pnode(mmr_blade);
3788         uv_write_global_mmr64(mmr_pnode, mmr_offset, mmr_value);
3789
3790         return irq;
3791 }
3792
3793 /*
3794  * Disable the specified MMR located on the specified blade so that MSIs are
3795  * longer allowed to be sent.
3796  */
3797 void arch_disable_uv_irq(int mmr_blade, unsigned long mmr_offset)
3798 {
3799         unsigned long mmr_value;
3800         struct uv_IO_APIC_route_entry *entry;
3801         int mmr_pnode;
3802
3803         BUILD_BUG_ON(sizeof(struct uv_IO_APIC_route_entry) != sizeof(unsigned long));
3804
3805         mmr_value = 0;
3806         entry = (struct uv_IO_APIC_route_entry *)&mmr_value;
3807         entry->mask = 1;
3808
3809         mmr_pnode = uv_blade_to_pnode(mmr_blade);
3810         uv_write_global_mmr64(mmr_pnode, mmr_offset, mmr_value);
3811 }
3812 #endif /* CONFIG_X86_64 */
3813
3814 int __init io_apic_get_redir_entries (int ioapic)
3815 {
3816         union IO_APIC_reg_01    reg_01;
3817         unsigned long flags;
3818
3819         spin_lock_irqsave(&ioapic_lock, flags);
3820         reg_01.raw = io_apic_read(ioapic, 1);
3821         spin_unlock_irqrestore(&ioapic_lock, flags);
3822
3823         return reg_01.bits.entries;
3824 }
3825
3826 void __init probe_nr_irqs_gsi(void)
3827 {
3828         int nr = 0;
3829
3830         nr = acpi_probe_gsi();
3831         if (nr > nr_irqs_gsi) {
3832                 nr_irqs_gsi = nr;
3833         } else {
3834                 /* for acpi=off or acpi is not compiled in */
3835                 int idx;
3836
3837                 nr = 0;
3838                 for (idx = 0; idx < nr_ioapics; idx++)
3839                         nr += io_apic_get_redir_entries(idx) + 1;
3840
3841                 if (nr > nr_irqs_gsi)
3842                         nr_irqs_gsi = nr;
3843         }
3844
3845         printk(KERN_DEBUG "nr_irqs_gsi: %d\n", nr_irqs_gsi);
3846 }
3847
3848 #ifdef CONFIG_SPARSE_IRQ
3849 int __init arch_probe_nr_irqs(void)
3850 {
3851         int nr;
3852
3853         if (nr_irqs > (NR_VECTORS * nr_cpu_ids))
3854                 nr_irqs = NR_VECTORS * nr_cpu_ids;
3855
3856         nr = nr_irqs_gsi + 8 * nr_cpu_ids;
3857 #if defined(CONFIG_PCI_MSI) || defined(CONFIG_HT_IRQ)
3858         /*
3859          * for MSI and HT dyn irq
3860          */
3861         nr += nr_irqs_gsi * 16;
3862 #endif
3863         if (nr < nr_irqs)
3864                 nr_irqs = nr;
3865
3866         return 0;
3867 }
3868 #endif
3869
3870 static int __io_apic_set_pci_routing(struct device *dev, int irq,
3871                                 struct io_apic_irq_attr *irq_attr)
3872 {
3873         struct irq_desc *desc;
3874         struct irq_cfg *cfg;
3875         int node;
3876         int ioapic, pin;
3877         int trigger, polarity;
3878
3879         ioapic = irq_attr->ioapic;
3880         if (!IO_APIC_IRQ(irq)) {
3881                 apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
3882                         ioapic);
3883                 return -EINVAL;
3884         }
3885
3886         if (dev)
3887                 node = dev_to_node(dev);
3888         else
3889                 node = cpu_to_node(boot_cpu_id);
3890
3891         desc = irq_to_desc_alloc_node(irq, node);
3892         if (!desc) {
3893                 printk(KERN_INFO "can not get irq_desc %d\n", irq);
3894                 return 0;
3895         }
3896
3897         pin = irq_attr->ioapic_pin;
3898         trigger = irq_attr->trigger;
3899         polarity = irq_attr->polarity;
3900
3901         /*
3902          * IRQs < 16 are already in the irq_2_pin[] map
3903          */
3904         if (irq >= NR_IRQS_LEGACY) {
3905                 cfg = desc->chip_data;
3906                 add_pin_to_irq_node(cfg, node, ioapic, pin);
3907         }
3908
3909         setup_IO_APIC_irq(ioapic, pin, irq, desc, trigger, polarity);
3910
3911         return 0;
3912 }
3913
3914 int io_apic_set_pci_routing(struct device *dev, int irq,
3915                                 struct io_apic_irq_attr *irq_attr)
3916 {
3917         int ioapic, pin;
3918         /*
3919          * Avoid pin reprogramming.  PRTs typically include entries
3920          * with redundant pin->gsi mappings (but unique PCI devices);
3921          * we only program the IOAPIC on the first.
3922          */
3923         ioapic = irq_attr->ioapic;
3924         pin = irq_attr->ioapic_pin;
3925         if (test_bit(pin, mp_ioapic_routing[ioapic].pin_programmed)) {
3926                 pr_debug("Pin %d-%d already programmed\n",
3927                          mp_ioapics[ioapic].apicid, pin);
3928                 return 0;
3929         }
3930         set_bit(pin, mp_ioapic_routing[ioapic].pin_programmed);
3931
3932         return __io_apic_set_pci_routing(dev, irq, irq_attr);
3933 }
3934
3935 /* --------------------------------------------------------------------------
3936                           ACPI-based IOAPIC Configuration
3937    -------------------------------------------------------------------------- */
3938
3939 #ifdef CONFIG_ACPI
3940
3941 #ifdef CONFIG_X86_32
3942 int __init io_apic_get_unique_id(int ioapic, int apic_id)
3943 {
3944         union IO_APIC_reg_00 reg_00;
3945         static physid_mask_t apic_id_map = PHYSID_MASK_NONE;
3946         physid_mask_t tmp;
3947         unsigned long flags;
3948         int i = 0;
3949
3950         /*
3951          * The P4 platform supports up to 256 APIC IDs on two separate APIC
3952          * buses (one for LAPICs, one for IOAPICs), where predecessors only
3953          * supports up to 16 on one shared APIC bus.
3954          *
3955          * TBD: Expand LAPIC/IOAPIC support on P4-class systems to take full
3956          *      advantage of new APIC bus architecture.
3957          */
3958
3959         if (physids_empty(apic_id_map))
3960                 apic_id_map = apic->ioapic_phys_id_map(phys_cpu_present_map);
3961
3962         spin_lock_irqsave(&ioapic_lock, flags);
3963         reg_00.raw = io_apic_read(ioapic, 0);
3964         spin_unlock_irqrestore(&ioapic_lock, flags);
3965
3966         if (apic_id >= get_physical_broadcast()) {
3967                 printk(KERN_WARNING "IOAPIC[%d]: Invalid apic_id %d, trying "
3968                         "%d\n", ioapic, apic_id, reg_00.bits.ID);
3969                 apic_id = reg_00.bits.ID;
3970         }
3971
3972         /*
3973          * Every APIC in a system must have a unique ID or we get lots of nice
3974          * 'stuck on smp_invalidate_needed IPI wait' messages.
3975          */
3976         if (apic->check_apicid_used(apic_id_map, apic_id)) {
3977
3978                 for (i = 0; i < get_physical_broadcast(); i++) {
3979                         if (!apic->check_apicid_used(apic_id_map, i))
3980                                 break;
3981                 }
3982
3983                 if (i == get_physical_broadcast())
3984                         panic("Max apic_id exceeded!\n");
3985
3986                 printk(KERN_WARNING "IOAPIC[%d]: apic_id %d already used, "
3987                         "trying %d\n", ioapic, apic_id, i);
3988
3989                 apic_id = i;
3990         }
3991
3992         tmp = apic->apicid_to_cpu_present(apic_id);
3993         physids_or(apic_id_map, apic_id_map, tmp);
3994
3995         if (reg_00.bits.ID != apic_id) {
3996                 reg_00.bits.ID = apic_id;
3997
3998                 spin_lock_irqsave(&ioapic_lock, flags);
3999                 io_apic_write(ioapic, 0, reg_00.raw);
4000                 reg_00.raw = io_apic_read(ioapic, 0);
4001                 spin_unlock_irqrestore(&ioapic_lock, flags);
4002
4003                 /* Sanity check */
4004                 if (reg_00.bits.ID != apic_id) {
4005                         printk("IOAPIC[%d]: Unable to change apic_id!\n", ioapic);
4006                         return -1;
4007                 }
4008         }
4009
4010         apic_printk(APIC_VERBOSE, KERN_INFO
4011                         "IOAPIC[%d]: Assigned apic_id %d\n", ioapic, apic_id);
4012
4013         return apic_id;
4014 }
4015
4016 int __init io_apic_get_version(int ioapic)
4017 {
4018         union IO_APIC_reg_01    reg_01;
4019         unsigned long flags;
4020
4021         spin_lock_irqsave(&ioapic_lock, flags);
4022         reg_01.raw = io_apic_read(ioapic, 1);
4023         spin_unlock_irqrestore(&ioapic_lock, flags);
4024
4025         return reg_01.bits.version;
4026 }
4027 #endif
4028
4029 int acpi_get_override_irq(int bus_irq, int *trigger, int *polarity)
4030 {
4031         int i;
4032
4033         if (skip_ioapic_setup)
4034                 return -1;
4035
4036         for (i = 0; i < mp_irq_entries; i++)
4037                 if (mp_irqs[i].irqtype == mp_INT &&
4038                     mp_irqs[i].srcbusirq == bus_irq)
4039                         break;
4040         if (i >= mp_irq_entries)
4041                 return -1;
4042
4043         *trigger = irq_trigger(i);
4044         *polarity = irq_polarity(i);
4045         return 0;
4046 }
4047
4048 #endif /* CONFIG_ACPI */
4049
4050 /*
4051  * This function currently is only a helper for the i386 smp boot process where
4052  * we need to reprogram the ioredtbls to cater for the cpus which have come online
4053  * so mask in all cases should simply be apic->target_cpus()
4054  */
4055 #ifdef CONFIG_SMP
4056 void __init setup_ioapic_dest(void)
4057 {
4058         int pin, ioapic = 0, irq, irq_entry;
4059         struct irq_desc *desc;
4060         const struct cpumask *mask;
4061
4062         if (skip_ioapic_setup == 1)
4063                 return;
4064
4065 #ifdef CONFIG_ACPI
4066         if (!acpi_disabled && acpi_ioapic) {
4067                 ioapic = mp_find_ioapic(0);
4068                 if (ioapic < 0)
4069                         ioapic = 0;
4070         }
4071 #endif
4072
4073         for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
4074                 irq_entry = find_irq_entry(ioapic, pin, mp_INT);
4075                 if (irq_entry == -1)
4076                         continue;
4077                 irq = pin_2_irq(irq_entry, ioapic, pin);
4078
4079                 desc = irq_to_desc(irq);
4080
4081                 /*
4082                  * Honour affinities which have been set in early boot
4083                  */
4084                 if (desc->status &
4085                     (IRQ_NO_BALANCING | IRQ_AFFINITY_SET))
4086                         mask = desc->affinity;
4087                 else
4088                         mask = apic->target_cpus();
4089
4090                 if (intr_remapping_enabled)
4091                         set_ir_ioapic_affinity_irq_desc(desc, mask);
4092                 else
4093                         set_ioapic_affinity_irq_desc(desc, mask);
4094         }
4095
4096 }
4097 #endif
4098
4099 #define IOAPIC_RESOURCE_NAME_SIZE 11
4100
4101 static struct resource *ioapic_resources;
4102
4103 static struct resource * __init ioapic_setup_resources(void)
4104 {
4105         unsigned long n;
4106         struct resource *res;
4107         char *mem;
4108         int i;
4109
4110         if (nr_ioapics <= 0)
4111                 return NULL;
4112
4113         n = IOAPIC_RESOURCE_NAME_SIZE + sizeof(struct resource);
4114         n *= nr_ioapics;
4115
4116         mem = alloc_bootmem(n);
4117         res = (void *)mem;
4118
4119         if (mem != NULL) {
4120                 mem += sizeof(struct resource) * nr_ioapics;
4121
4122                 for (i = 0; i < nr_ioapics; i++) {
4123                         res[i].name = mem;
4124                         res[i].flags = IORESOURCE_MEM | IORESOURCE_BUSY;
4125                         sprintf(mem,  "IOAPIC %u", i);
4126                         mem += IOAPIC_RESOURCE_NAME_SIZE;
4127                 }
4128         }
4129
4130         ioapic_resources = res;
4131
4132         return res;
4133 }
4134
4135 void __init ioapic_init_mappings(void)
4136 {
4137         unsigned long ioapic_phys, idx = FIX_IO_APIC_BASE_0;
4138         struct resource *ioapic_res;
4139         int i;
4140
4141         ioapic_res = ioapic_setup_resources();
4142         for (i = 0; i < nr_ioapics; i++) {
4143                 if (smp_found_config) {
4144                         ioapic_phys = mp_ioapics[i].apicaddr;
4145 #ifdef CONFIG_X86_32
4146                         if (!ioapic_phys) {
4147                                 printk(KERN_ERR
4148                                        "WARNING: bogus zero IO-APIC "
4149                                        "address found in MPTABLE, "
4150                                        "disabling IO/APIC support!\n");
4151                                 smp_found_config = 0;
4152                                 skip_ioapic_setup = 1;
4153                                 goto fake_ioapic_page;
4154                         }
4155 #endif
4156                 } else {
4157 #ifdef CONFIG_X86_32
4158 fake_ioapic_page:
4159 #endif
4160                         ioapic_phys = (unsigned long)
4161                                 alloc_bootmem_pages(PAGE_SIZE);
4162                         ioapic_phys = __pa(ioapic_phys);
4163                 }
4164                 set_fixmap_nocache(idx, ioapic_phys);
4165                 apic_printk(APIC_VERBOSE,
4166                             "mapped IOAPIC to %08lx (%08lx)\n",
4167                             __fix_to_virt(idx), ioapic_phys);
4168                 idx++;
4169
4170                 if (ioapic_res != NULL) {
4171                         ioapic_res->start = ioapic_phys;
4172                         ioapic_res->end = ioapic_phys + (4 * 1024) - 1;
4173                         ioapic_res++;
4174                 }
4175         }
4176 }
4177
4178 static int __init ioapic_insert_resources(void)
4179 {
4180         int i;
4181         struct resource *r = ioapic_resources;
4182
4183         if (!r) {
4184                 if (nr_ioapics > 0) {
4185                         printk(KERN_ERR
4186                                 "IO APIC resources couldn't be allocated.\n");
4187                         return -1;
4188                 }
4189                 return 0;
4190         }
4191
4192         for (i = 0; i < nr_ioapics; i++) {
4193                 insert_resource(&iomem_resource, r);
4194                 r++;
4195         }
4196
4197         return 0;
4198 }
4199
4200 /* Insert the IO APIC resources after PCI initialization has occured to handle
4201  * IO APICS that are mapped in on a BAR in PCI space. */
4202 late_initcall(ioapic_insert_resources);