Linux kernel & device driver programming

Cross-Referenced Linux and Device Driver Code

[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ]
Version: [ 2.6.11.8 ] [ 2.6.25 ] [ 2.6.25.8 ] [ 2.6.31.13 ] Architecture: [ i386 ]
  1 /*
  2  *      Intel IO-APIC support for multi-Pentium hosts.
  3  *
  4  *      Copyright (C) 1997, 1998, 1999, 2000, 2009 Ingo Molnar, Hajnalka Szabo
  5  *
  6  *      Many thanks to Stig Venaas for trying out countless experimental
  7  *      patches and reporting/debugging problems patiently!
  8  *
  9  *      (c) 1999, Multiple IO-APIC support, developed by
 10  *      Ken-ichi Yaku <yaku@css1.kbnes.nec.co.jp> and
 11  *      Hidemi Kishimoto <kisimoto@css1.kbnes.nec.co.jp>,
 12  *      further tested and cleaned up by Zach Brown <zab@redhat.com>
 13  *      and Ingo Molnar <mingo@redhat.com>
 14  *
 15  *      Fixes
 16  *      Maciej W. Rozycki       :       Bits for genuine 82489DX APICs;
 17  *                                      thanks to Eric Gilmore
 18  *                                      and Rolf G. Tews
 19  *                                      for testing these extensively
 20  *      Paul Diefenbaugh        :       Added full ACPI support
 21  */
 22 
 23 #include <linux/mm.h>
 24 #include <linux/interrupt.h>
 25 #include <linux/init.h>
 26 #include <linux/delay.h>
 27 #include <linux/sched.h>
 28 #include <linux/pci.h>
 29 #include <linux/mc146818rtc.h>
 30 #include <linux/compiler.h>
 31 #include <linux/acpi.h>
 32 #include <linux/module.h>
 33 #include <linux/sysdev.h>
 34 #include <linux/msi.h>
 35 #include <linux/htirq.h>
 36 #include <linux/freezer.h>
 37 #include <linux/kthread.h>
 38 #include <linux/jiffies.h>      /* time_after() */
 39 #ifdef CONFIG_ACPI
 40 #include <acpi/acpi_bus.h>
 41 #endif
 42 #include <linux/bootmem.h>
 43 #include <linux/dmar.h>
 44 #include <linux/hpet.h>
 45 
 46 #include <asm/idle.h>
 47 #include <asm/io.h>
 48 #include <asm/smp.h>
 49 #include <asm/cpu.h>
 50 #include <asm/desc.h>
 51 #include <asm/proto.h>
 52 #include <asm/acpi.h>
 53 #include <asm/dma.h>
 54 #include <asm/timer.h>
 55 #include <asm/i8259.h>
 56 #include <asm/nmi.h>
 57 #include <asm/msidef.h>
 58 #include <asm/hypertransport.h>
 59 #include <asm/setup.h>
 60 #include <asm/irq_remapping.h>
 61 #include <asm/hpet.h>
 62 #include <asm/hw_irq.h>
 63 #include <asm/uv/uv_hub.h>
 64 #include <asm/uv/uv_irq.h>
 65 
 66 #include <asm/apic.h>
 67 
 68 #define __apicdebuginit(type) static type __init
 69 
 70 /*
 71  *      Is the SiS APIC rmw bug present ?
 72  *      -1 = don't know, 0 = no, 1 = yes
 73  */
 74 int sis_apic_bug = -1;
 75 
 76 static DEFINE_SPINLOCK(ioapic_lock);
 77 static DEFINE_SPINLOCK(vector_lock);
 78 
 79 /*
 80  * # of IRQ routing registers
 81  */
 82 int nr_ioapic_registers[MAX_IO_APICS];
 83 
 84 /* I/O APIC entries */
 85 struct mpc_ioapic mp_ioapics[MAX_IO_APICS];
 86 int nr_ioapics;
 87 
 88 /* MP IRQ source entries */
 89 struct mpc_intsrc mp_irqs[MAX_IRQ_SOURCES];
 90 
 91 /* # of MP IRQ source entries */
 92 int mp_irq_entries;
 93 
 94 #if defined (CONFIG_MCA) || defined (CONFIG_EISA)
 95 int mp_bus_id_to_type[MAX_MP_BUSSES];
 96 #endif
 97 
 98 DECLARE_BITMAP(mp_bus_not_pci, MAX_MP_BUSSES);
 99 
100 int skip_ioapic_setup;
101 
102 void arch_disable_smp_support(void)
103 {
104 #ifdef CONFIG_PCI
105         noioapicquirk = 1;
106         noioapicreroute = -1;
107 #endif
108         skip_ioapic_setup = 1;
109 }
110 
111 static int __init parse_noapic(char *str)
112 {
113         /* disable IO-APIC */
114         arch_disable_smp_support();
115         return 0;
116 }
117 early_param("noapic", parse_noapic);
118 
119 struct irq_pin_list;
120 
121 /*
122  * This is performance-critical, we want to do it O(1)
123  *
124  * the indexing order of this array favors 1:1 mappings
125  * between pins and IRQs.
126  */
127 
128 struct irq_pin_list {
129         int apic, pin;
130         struct irq_pin_list *next;
131 };
132 
133 static struct irq_pin_list *get_one_free_irq_2_pin(int node)
134 {
135         struct irq_pin_list *pin;
136 
137         pin = kzalloc_node(sizeof(*pin), GFP_ATOMIC, node);
138 
139         return pin;
140 }
141 
142 struct irq_cfg {
143         struct irq_pin_list *irq_2_pin;
144         cpumask_var_t domain;
145         cpumask_var_t old_domain;
146         unsigned move_cleanup_count;
147         u8 vector;
148         u8 move_in_progress : 1;
149 };
150 
151 /* irq_cfg is indexed by the sum of all RTEs in all I/O APICs. */
152 #ifdef CONFIG_SPARSE_IRQ
153 static struct irq_cfg irq_cfgx[] = {
154 #else
155 static struct irq_cfg irq_cfgx[NR_IRQS] = {
156 #endif
157         [0]  = { .vector = IRQ0_VECTOR,  },
158         [1]  = { .vector = IRQ1_VECTOR,  },
159         [2]  = { .vector = IRQ2_VECTOR,  },
160         [3]  = { .vector = IRQ3_VECTOR,  },
161         [4]  = { .vector = IRQ4_VECTOR,  },
162         [5]  = { .vector = IRQ5_VECTOR,  },
163         [6]  = { .vector = IRQ6_VECTOR,  },
164         [7]  = { .vector = IRQ7_VECTOR,  },
165         [8]  = { .vector = IRQ8_VECTOR,  },
166         [9]  = { .vector = IRQ9_VECTOR,  },
167         [10] = { .vector = IRQ10_VECTOR, },
168         [11] = { .vector = IRQ11_VECTOR, },
169         [12] = { .vector = IRQ12_VECTOR, },
170         [13] = { .vector = IRQ13_VECTOR, },
171         [14] = { .vector = IRQ14_VECTOR, },
172         [15] = { .vector = IRQ15_VECTOR, },
173 };
174 
175 int __init arch_early_irq_init(void)
176 {
177         struct irq_cfg *cfg;
178         struct irq_desc *desc;
179         int count;
180         int node;
181         int i;
182 
183         cfg = irq_cfgx;
184         count = ARRAY_SIZE(irq_cfgx);
185         node= cpu_to_node(boot_cpu_id);
186 
187         for (i = 0; i < count; i++) {
188                 desc = irq_to_desc(i);
189                 desc->chip_data = &cfg[i];
190                 zalloc_cpumask_var_node(&cfg[i].domain, GFP_NOWAIT, node);
191                 zalloc_cpumask_var_node(&cfg[i].old_domain, GFP_NOWAIT, node);
192                 if (i < NR_IRQS_LEGACY)
193                         cpumask_setall(cfg[i].domain);
194         }
195 
196         return 0;
197 }
198 
199 #ifdef CONFIG_SPARSE_IRQ
200 static struct irq_cfg *irq_cfg(unsigned int irq)
201 {
202         struct irq_cfg *cfg = NULL;
203         struct irq_desc *desc;
204 
205         desc = irq_to_desc(irq);
206         if (desc)
207                 cfg = desc->chip_data;
208 
209         return cfg;
210 }
211 
212 static struct irq_cfg *get_one_free_irq_cfg(int node)
213 {
214         struct irq_cfg *cfg;
215 
216         cfg = kzalloc_node(sizeof(*cfg), GFP_ATOMIC, node);
217         if (cfg) {
218                 if (!alloc_cpumask_var_node(&cfg->domain, GFP_ATOMIC, node)) {
219                         kfree(cfg);
220                         cfg = NULL;
221                 } else if (!alloc_cpumask_var_node(&cfg->old_domain,
222                                                           GFP_ATOMIC, node)) {
223                         free_cpumask_var(cfg->domain);
224                         kfree(cfg);
225                         cfg = NULL;
226                 } else {
227                         cpumask_clear(cfg->domain);
228                         cpumask_clear(cfg->old_domain);
229                 }
230         }
231 
232         return cfg;
233 }
234 
235 int arch_init_chip_data(struct irq_desc *desc, int node)
236 {
237         struct irq_cfg *cfg;
238 
239         cfg = desc->chip_data;
240         if (!cfg) {
241                 desc->chip_data = get_one_free_irq_cfg(node);
242                 if (!desc->chip_data) {
243                         printk(KERN_ERR "can not alloc irq_cfg\n");
244                         BUG_ON(1);
245                 }
246         }
247 
248         return 0;
249 }
250 
251 /* for move_irq_desc */
252 static void
253 init_copy_irq_2_pin(struct irq_cfg *old_cfg, struct irq_cfg *cfg, int node)
254 {
255         struct irq_pin_list *old_entry, *head, *tail, *entry;
256 
257         cfg->irq_2_pin = NULL;
258         old_entry = old_cfg->irq_2_pin;
259         if (!old_entry)
260                 return;
261 
262         entry = get_one_free_irq_2_pin(node);
263         if (!entry)
264                 return;
265 
266         entry->apic     = old_entry->apic;
267         entry->pin      = old_entry->pin;
268         head            = entry;
269         tail            = entry;
270         old_entry       = old_entry->next;
271         while (old_entry) {
272                 entry = get_one_free_irq_2_pin(node);
273                 if (!entry) {
274                         entry = head;
275                         while (entry) {
276                                 head = entry->next;
277                                 kfree(entry);
278                                 entry = head;
279                         }
280                         /* still use the old one */
281                         return;
282                 }
283                 entry->apic     = old_entry->apic;
284                 entry->pin      = old_entry->pin;
285                 tail->next      = entry;
286                 tail            = entry;
287                 old_entry       = old_entry->next;
288         }
289 
290         tail->next = NULL;
291         cfg->irq_2_pin = head;
292 }
293 
294 static void free_irq_2_pin(struct irq_cfg *old_cfg, struct irq_cfg *cfg)
295 {
296         struct irq_pin_list *entry, *next;
297 
298         if (old_cfg->irq_2_pin == cfg->irq_2_pin)
299                 return;
300 
301         entry = old_cfg->irq_2_pin;
302 
303         while (entry) {
304                 next = entry->next;
305                 kfree(entry);
306                 entry = next;
307         }
308         old_cfg->irq_2_pin = NULL;
309 }
310 
311 void arch_init_copy_chip_data(struct irq_desc *old_desc,
312                                  struct irq_desc *desc, int node)
313 {
314         struct irq_cfg *cfg;
315         struct irq_cfg *old_cfg;
316 
317         cfg = get_one_free_irq_cfg(node);
318 
319         if (!cfg)
320                 return;
321 
322         desc->chip_data = cfg;
323 
324         old_cfg = old_desc->chip_data;
325 
326         memcpy(cfg, old_cfg, sizeof(struct irq_cfg));
327 
328         init_copy_irq_2_pin(old_cfg, cfg, node);
329 }
330 
331 static void free_irq_cfg(struct irq_cfg *old_cfg)
332 {
333         kfree(old_cfg);
334 }
335 
336 void arch_free_chip_data(struct irq_desc *old_desc, struct irq_desc *desc)
337 {
338         struct irq_cfg *old_cfg, *cfg;
339 
340         old_cfg = old_desc->chip_data;
341         cfg = desc->chip_data;
342 
343         if (old_cfg == cfg)
344                 return;
345 
346         if (old_cfg) {
347                 free_irq_2_pin(old_cfg, cfg);
348                 free_irq_cfg(old_cfg);
349                 old_desc->chip_data = NULL;
350         }
351 }
352 /* end for move_irq_desc */
353 
354 #else
355 static struct irq_cfg *irq_cfg(unsigned int irq)
356 {
357         return irq < nr_irqs ? irq_cfgx + irq : NULL;
358 }
359 
360 #endif
361 
362 struct io_apic {
363         unsigned int index;
364         unsigned int unused[3];
365         unsigned int data;
366         unsigned int unused2[11];
367         unsigned int eoi;
368 };
369 
370 static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
371 {
372         return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
373                 + (mp_ioapics[idx].apicaddr & ~PAGE_MASK);
374 }
375 
376 static inline void io_apic_eoi(unsigned int apic, unsigned int vector)
377 {
378         struct io_apic __iomem *io_apic = io_apic_base(apic);
379         writel(vector, &io_apic->eoi);
380 }
381 
382 static inline unsigned int io_apic_read(unsigned int apic, unsigned int reg)
383 {
384         struct io_apic __iomem *io_apic = io_apic_base(apic);
385         writel(reg, &io_apic->index);
386         return readl(&io_apic->data);
387 }
388 
389 static inline void io_apic_write(unsigned int apic, unsigned int reg, unsigned int value)
390 {
391         struct io_apic __iomem *io_apic = io_apic_base(apic);
392         writel(reg, &io_apic->index);
393         writel(value, &io_apic->data);
394 }
395 
396 /*
397  * Re-write a value: to be used for read-modify-write
398  * cycles where the read already set up the index register.
399  *
400  * Older SiS APIC requires we rewrite the index register
401  */
402 static inline void io_apic_modify(unsigned int apic, unsigned int reg, unsigned int value)
403 {
404         struct io_apic __iomem *io_apic = io_apic_base(apic);
405 
406         if (sis_apic_bug)
407                 writel(reg, &io_apic->index);
408         writel(value, &io_apic->data);
409 }
410 
411 static bool io_apic_level_ack_pending(struct irq_cfg *cfg)
412 {
413         struct irq_pin_list *entry;
414         unsigned long flags;
415 
416         spin_lock_irqsave(&ioapic_lock, flags);
417         entry = cfg->irq_2_pin;
418         for (;;) {
419                 unsigned int reg;
420                 int pin;
421 
422                 if (!entry)
423                         break;
424                 pin = entry->pin;
425                 reg = io_apic_read(entry->apic, 0x10 + pin*2);
426                 /* Is the remote IRR bit set? */
427                 if (reg & IO_APIC_REDIR_REMOTE_IRR) {
428                         spin_unlock_irqrestore(&ioapic_lock, flags);
429                         return true;
430                 }
431                 if (!entry->next)
432                         break;
433                 entry = entry->next;
434         }
435         spin_unlock_irqrestore(&ioapic_lock, flags);
436 
437         return false;
438 }
439 
440 union entry_union {
441         struct { u32 w1, w2; };
442         struct IO_APIC_route_entry entry;
443 };
444 
445 static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
446 {
447         union entry_union eu;
448         unsigned long flags;
449         spin_lock_irqsave(&ioapic_lock, flags);
450         eu.w1 = io_apic_read(apic, 0x10 + 2 * pin);
451         eu.w2 = io_apic_read(apic, 0x11 + 2 * pin);
452         spin_unlock_irqrestore(&ioapic_lock, flags);
453         return eu.entry;
454 }
455 
456 /*
457  * When we write a new IO APIC routing entry, we need to write the high
458  * word first! If the mask bit in the low word is clear, we will enable
459  * the interrupt, and we need to make sure the entry is fully populated
460  * before that happens.
461  */
462 static void
463 __ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
464 {
465         union entry_union eu = {{0, 0}};
466 
467         eu.entry = e;
468         io_apic_write(apic, 0x11 + 2*pin, eu.w2);
469         io_apic_write(apic, 0x10 + 2*pin, eu.w1);
470 }
471 
472 void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
473 {
474         unsigned long flags;
475         spin_lock_irqsave(&ioapic_lock, flags);
476         __ioapic_write_entry(apic, pin, e);
477         spin_unlock_irqrestore(&ioapic_lock, flags);
478 }
479 
480 /*
481  * When we mask an IO APIC routing entry, we need to write the low
482  * word first, in order to set the mask bit before we change the
483  * high bits!
484  */
485 static void ioapic_mask_entry(int apic, int pin)
486 {
487         unsigned long flags;
488         union entry_union eu = { .entry.mask = 1 };
489 
490         spin_lock_irqsave(&ioapic_lock, flags);
491         io_apic_write(apic, 0x10 + 2*pin, eu.w1);
492         io_apic_write(apic, 0x11 + 2*pin, eu.w2);
493         spin_unlock_irqrestore(&ioapic_lock, flags);
494 }
495 
496 /*
497  * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
498  * shared ISA-space IRQs, so we have to support them. We are super
499  * fast in the common case, and fast for shared ISA-space IRQs.
500  */
501 static void add_pin_to_irq_node(struct irq_cfg *cfg, int node, int apic, int pin)
502 {
503         struct irq_pin_list *entry;
504 
505         entry = cfg->irq_2_pin;
506         if (!entry) {
507                 entry = get_one_free_irq_2_pin(node);
508                 if (!entry) {
509                         printk(KERN_ERR "can not alloc irq_2_pin to add %d - %d\n",
510                                         apic, pin);
511                         return;
512                 }
513                 cfg->irq_2_pin = entry;
514                 entry->apic = apic;
515                 entry->pin = pin;
516                 return;
517         }
518 
519         while (entry->next) {
520                 /* not again, please */
521                 if (entry->apic == apic && entry->pin == pin)
522                         return;
523 
524                 entry = entry->next;
525         }
526 
527         entry->next = get_one_free_irq_2_pin(node);
528         entry = entry->next;
529         entry->apic = apic;
530         entry->pin = pin;
531 }
532 
533 /*
534  * Reroute an IRQ to a different pin.
535  */
536 static void __init replace_pin_at_irq_node(struct irq_cfg *cfg, int node,
537                                       int oldapic, int oldpin,
538                                       int newapic, int newpin)
539 {
540         struct irq_pin_list *entry = cfg->irq_2_pin;
541         int replaced = 0;
542 
543         while (entry) {
544                 if (entry->apic == oldapic && entry->pin == oldpin) {
545                         entry->apic = newapic;
546                         entry->pin = newpin;
547                         replaced = 1;
548                         /* every one is different, right? */
549                         break;
550                 }
551                 entry = entry->next;
552         }
553 
554         /* why? call replace before add? */
555         if (!replaced)
556                 add_pin_to_irq_node(cfg, node, newapic, newpin);
557 }
558 
559 static inline void io_apic_modify_irq(struct irq_cfg *cfg,
560                                 int mask_and, int mask_or,
561                                 void (*final)(struct irq_pin_list *entry))
562 {
563         int pin;
564         struct irq_pin_list *entry;
565 
566         for (entry = cfg->irq_2_pin; entry != NULL; entry = entry->next) {
567                 unsigned int reg;
568                 pin = entry->pin;
569                 reg = io_apic_read(entry->apic, 0x10 + pin * 2);
570                 reg &= mask_and;
571                 reg |= mask_or;
572                 io_apic_modify(entry->apic, 0x10 + pin * 2, reg);
573                 if (final)
574                         final(entry);
575         }
576 }
577 
578 static void __unmask_IO_APIC_irq(struct irq_cfg *cfg)
579 {
580         io_apic_modify_irq(cfg, ~IO_APIC_REDIR_MASKED, 0, NULL);
581 }
582 
583 #ifdef CONFIG_X86_64
584 static void io_apic_sync(struct irq_pin_list *entry)
585 {
586         /*
587          * Synchronize the IO-APIC and the CPU by doing
588          * a dummy read from the IO-APIC
589          */
590         struct io_apic __iomem *io_apic;
591         io_apic = io_apic_base(entry->apic);
592         readl(&io_apic->data);
593 }
594 
595 static void __mask_IO_APIC_irq(struct irq_cfg *cfg)
596 {
597         io_apic_modify_irq(cfg, ~0, IO_APIC_REDIR_MASKED, &io_apic_sync);
598 }
599 #else /* CONFIG_X86_32 */
600 static void __mask_IO_APIC_irq(struct irq_cfg *cfg)
601 {
602         io_apic_modify_irq(cfg, ~0, IO_APIC_REDIR_MASKED, NULL);
603 }
604 
605 static void __mask_and_edge_IO_APIC_irq(struct irq_cfg *cfg)
606 {
607         io_apic_modify_irq(cfg, ~IO_APIC_REDIR_LEVEL_TRIGGER,
608                         IO_APIC_REDIR_MASKED, NULL);
609 }
610 
611 static void __unmask_and_level_IO_APIC_irq(struct irq_cfg *cfg)
612 {
613         io_apic_modify_irq(cfg, ~IO_APIC_REDIR_MASKED,
614                         IO_APIC_REDIR_LEVEL_TRIGGER, NULL);
615 }
616 #endif /* CONFIG_X86_32 */
617 
618 static void mask_IO_APIC_irq_desc(struct irq_desc *desc)
619 {
620         struct irq_cfg *cfg = desc->chip_data;
621         unsigned long flags;
622 
623         BUG_ON(!cfg);
624 
625         spin_lock_irqsave(&ioapic_lock, flags);
626         __mask_IO_APIC_irq(cfg);
627         spin_unlock_irqrestore(&ioapic_lock, flags);
628 }
629 
630 static void unmask_IO_APIC_irq_desc(struct irq_desc *desc)
631 {
632         struct irq_cfg *cfg = desc->chip_data;
633         unsigned long flags;
634 
635         spin_lock_irqsave(&ioapic_lock, flags);
636         __unmask_IO_APIC_irq(cfg);
637         spin_unlock_irqrestore(&ioapic_lock, flags);
638 }
639 
640 static void mask_IO_APIC_irq(unsigned int irq)
641 {
642         struct irq_desc *desc = irq_to_desc(irq);
643 
644         mask_IO_APIC_irq_desc(desc);
645 }
646 static void unmask_IO_APIC_irq(unsigned int irq)
647 {
648         struct irq_desc *desc = irq_to_desc(irq);
649 
650         unmask_IO_APIC_irq_desc(desc);
651 }
652 
653 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
654 {
655         struct IO_APIC_route_entry entry;
656 
657         /* Check delivery_mode to be sure we're not clearing an SMI pin */
658         entry = ioapic_read_entry(apic, pin);
659         if (entry.delivery_mode == dest_SMI)
660                 return;
661         /*
662          * Disable it in the IO-APIC irq-routing table:
663          */
664         ioapic_mask_entry(apic, pin);
665 }
666 
667 static void clear_IO_APIC (void)
668 {
669         int apic, pin;
670 
671         for (apic = 0; apic < nr_ioapics; apic++)
672                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
673                         clear_IO_APIC_pin(apic, pin);
674 }
675 
676 #ifdef CONFIG_X86_32
677 /*
678  * support for broken MP BIOSs, enables hand-redirection of PIRQ0-7 to
679  * specific CPU-side IRQs.
680  */
681 
682 #define MAX_PIRQS 8
683 static int pirq_entries[MAX_PIRQS] = {
684         [0 ... MAX_PIRQS - 1] = -1
685 };
686 
687 static int __init ioapic_pirq_setup(char *str)
688 {
689         int i, max;
690         int ints[MAX_PIRQS+1];
691 
692         get_options(str, ARRAY_SIZE(ints), ints);
693 
694         apic_printk(APIC_VERBOSE, KERN_INFO
695                         "PIRQ redirection, working around broken MP-BIOS.\n");
696         max = MAX_PIRQS;
697         if (ints[0] < MAX_PIRQS)
698                 max = ints[0];
699 
700         for (i = 0; i < max; i++) {
701                 apic_printk(APIC_VERBOSE, KERN_DEBUG
702                                 "... PIRQ%d -> IRQ %d\n", i, ints[i+1]);
703                 /*
704                  * PIRQs are mapped upside down, usually.
705                  */
706                 pirq_entries[MAX_PIRQS-i-1] = ints[i+1];
707         }
708         return 1;
709 }
710 
711 __setup("pirq=", ioapic_pirq_setup);
712 #endif /* CONFIG_X86_32 */
713 
714 struct IO_APIC_route_entry **alloc_ioapic_entries(void)
715 {
716         int apic;
717         struct IO_APIC_route_entry **ioapic_entries;
718 
719         ioapic_entries = kzalloc(sizeof(*ioapic_entries) * nr_ioapics,
720                                 GFP_ATOMIC);
721         if (!ioapic_entries)
722                 return 0;
723 
724         for (apic = 0; apic < nr_ioapics; apic++) {
725                 ioapic_entries[apic] =
726                         kzalloc(sizeof(struct IO_APIC_route_entry) *
727                                 nr_ioapic_registers[apic], GFP_ATOMIC);
728                 if (!ioapic_entries[apic])
729                         goto nomem;
730         }
731 
732         return ioapic_entries;
733 
734 nomem:
735         while (--apic >= 0)
736                 kfree(ioapic_entries[apic]);
737         kfree(ioapic_entries);
738 
739         return 0;
740 }
741 
742 /*
743  * Saves all the IO-APIC RTE's
744  */
745 int save_IO_APIC_setup(struct IO_APIC_route_entry **ioapic_entries)
746 {
747         int apic, pin;
748 
749         if (!ioapic_entries)
750                 return -ENOMEM;
751 
752         for (apic = 0; apic < nr_ioapics; apic++) {
753                 if (!ioapic_entries[apic])
754                         return -ENOMEM;
755 
756                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
757                         ioapic_entries[apic][pin] =
758                                 ioapic_read_entry(apic, pin);
759         }
760 
761         return 0;
762 }
763 
764 /*
765  * Mask all IO APIC entries.
766  */
767 void mask_IO_APIC_setup(struct IO_APIC_route_entry **ioapic_entries)
768 {
769         int apic, pin;
770 
771         if (!ioapic_entries)
772                 return;
773 
774         for (apic = 0; apic < nr_ioapics; apic++) {
775                 if (!ioapic_entries[apic])
776                         break;
777 
778                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
779                         struct IO_APIC_route_entry entry;
780 
781                         entry = ioapic_entries[apic][pin];
782                         if (!entry.mask) {
783                                 entry.mask = 1;
784                                 ioapic_write_entry(apic, pin, entry);
785                         }
786                 }
787         }
788 }
789 
790 /*
791  * Restore IO APIC entries which was saved in ioapic_entries.
792  */
793 int restore_IO_APIC_setup(struct IO_APIC_route_entry **ioapic_entries)
794 {
795         int apic, pin;
796 
797         if (!ioapic_entries)
798                 return -ENOMEM;
799 
800         for (apic = 0; apic < nr_ioapics; apic++) {
801                 if (!ioapic_entries[apic])
802                         return -ENOMEM;
803 
804                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
805                         ioapic_write_entry(apic, pin,
806                                         ioapic_entries[apic][pin]);
807         }
808         return 0;
809 }
810 
811 void free_ioapic_entries(struct IO_APIC_route_entry **ioapic_entries)
812 {
813         int apic;
814 
815         for (apic = 0; apic < nr_ioapics; apic++)
816                 kfree(ioapic_entries[apic]);
817 
818         kfree(ioapic_entries);
819 }
820 
821 /*
822  * Find the IRQ entry number of a certain pin.
823  */
824 static int find_irq_entry(int apic, int pin, int type)
825 {
826         int i;
827 
828         for (i = 0; i < mp_irq_entries; i++)
829                 if (mp_irqs[i].irqtype == type &&
830                     (mp_irqs[i].dstapic == mp_ioapics[apic].apicid ||
831                      mp_irqs[i].dstapic == MP_APIC_ALL) &&
832                     mp_irqs[i].dstirq == pin)
833                         return i;
834 
835         return -1;
836 }
837 
838 /*
839  * Find the pin to which IRQ[irq] (ISA) is connected
840  */
841 static int __init find_isa_irq_pin(int irq, int type)
842 {
843         int i;
844 
845         for (i = 0; i < mp_irq_entries; i++) {
846                 int lbus = mp_irqs[i].srcbus;
847 
848                 if (test_bit(lbus, mp_bus_not_pci) &&
849                     (mp_irqs[i].irqtype == type) &&
850                     (mp_irqs[i].srcbusirq == irq))
851 
852                         return mp_irqs[i].dstirq;
853         }
854         return -1;
855 }
856 
857 static int __init find_isa_irq_apic(int irq, int type)
858 {
859         int i;
860 
861         for (i = 0; i < mp_irq_entries; i++) {
862                 int lbus = mp_irqs[i].srcbus;
863 
864                 if (test_bit(lbus, mp_bus_not_pci) &&
865                     (mp_irqs[i].irqtype == type) &&
866                     (mp_irqs[i].srcbusirq == irq))
867                         break;
868         }
869         if (i < mp_irq_entries) {
870                 int apic;
871                 for(apic = 0; apic < nr_ioapics; apic++) {
872                         if (mp_ioapics[apic].apicid == mp_irqs[i].dstapic)
873                                 return apic;
874                 }
875         }
876 
877         return -1;
878 }
879 
880 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
881 /*
882  * EISA Edge/Level control register, ELCR
883  */
884 static int EISA_ELCR(unsigned int irq)
885 {
886         if (irq < NR_IRQS_LEGACY) {
887                 unsigned int port = 0x4d0 + (irq >> 3);
888                 return (inb(port) >> (irq & 7)) & 1;
889         }
890         apic_printk(APIC_VERBOSE, KERN_INFO
891                         "Broken MPtable reports ISA irq %d\n", irq);
892         return 0;
893 }
894 
895 #endif
896 
897 /* ISA interrupts are always polarity zero edge triggered,
898  * when listed as conforming in the MP table. */
899 
900 #define default_ISA_trigger(idx)        (0)
901 #define default_ISA_polarity(idx)       (0)
902 
903 /* EISA interrupts are always polarity zero and can be edge or level
904  * trigger depending on the ELCR value.  If an interrupt is listed as
905  * EISA conforming in the MP table, that means its trigger type must
906  * be read in from the ELCR */
907 
908 #define default_EISA_trigger(idx)       (EISA_ELCR(mp_irqs[idx].srcbusirq))
909 #define default_EISA_polarity(idx)      default_ISA_polarity(idx)
910 
911 /* PCI interrupts are always polarity one level triggered,
912  * when listed as conforming in the MP table. */
913 
914 #define default_PCI_trigger(idx)        (1)
915 #define default_PCI_polarity(idx)       (1)
916 
917 /* MCA interrupts are always polarity zero level triggered,
918  * when listed as conforming in the MP table. */
919 
920 #define default_MCA_trigger(idx)        (1)
921 #define default_MCA_polarity(idx)       default_ISA_polarity(idx)
922 
923 static int MPBIOS_polarity(int idx)
924 {
925         int bus = mp_irqs[idx].srcbus;
926         int polarity;
927 
928         /*
929          * Determine IRQ line polarity (high active or low active):
930          */
931         switch (mp_irqs[idx].irqflag & 3)
932         {
933                 case 0: /* conforms, ie. bus-type dependent polarity */
934                         if (test_bit(bus, mp_bus_not_pci))
935                                 polarity = default_ISA_polarity(idx);
936                         else
937                                 polarity = default_PCI_polarity(idx);
938                         break;
939                 case 1: /* high active */
940                 {
941                         polarity = 0;
942                         break;
943                 }
944                 case 2: /* reserved */
945                 {
946                         printk(KERN_WARNING "broken BIOS!!\n");
947                         polarity = 1;
948                         break;
949                 }
950                 case 3: /* low active */
951                 {
952                         polarity = 1;
953                         break;
954                 }
955                 default: /* invalid */
956                 {
957                         printk(KERN_WARNING "broken BIOS!!\n");
958                         polarity = 1;
959                         break;
960                 }
961         }
962         return polarity;
963 }
964 
965 static int MPBIOS_trigger(int idx)
966 {
967         int bus = mp_irqs[idx].srcbus;
968         int trigger;
969 
970         /*
971          * Determine IRQ trigger mode (edge or level sensitive):
972          */
973         switch ((mp_irqs[idx].irqflag>>2) & 3)
974         {
975                 case 0: /* conforms, ie. bus-type dependent */
976                         if (test_bit(bus, mp_bus_not_pci))
977                                 trigger = default_ISA_trigger(idx);
978                         else
979                                 trigger = default_PCI_trigger(idx);
980 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
981                         switch (mp_bus_id_to_type[bus]) {
982                                 case MP_BUS_ISA: /* ISA pin */
983                                 {
984                                         /* set before the switch */
985                                         break;
986                                 }
987                                 case MP_BUS_EISA: /* EISA pin */
988                                 {
989                                         trigger = default_EISA_trigger(idx);
990                                         break;
991                                 }
992                                 case MP_BUS_PCI: /* PCI pin */
993                                 {
994                                         /* set before the switch */
995                                         break;
996                                 }
997                                 case MP_BUS_MCA: /* MCA pin */
998                                 {
999                                         trigger = default_MCA_trigger(idx);
1000                                         break;
1001                                 }
1002                                 default:
1003                                 {
1004                                         printk(KERN_WARNING "broken BIOS!!\n");
1005                                         trigger = 1;
1006                                         break;
1007                                 }
1008                         }
1009 #endif
1010                         break;
1011                 case 1: /* edge */
1012                 {
1013                         trigger = 0;
1014                         break;
1015                 }
1016                 case 2: /* reserved */
1017                 {
1018                         printk(KERN_WARNING "broken BIOS!!\n");
1019                         trigger = 1;
1020                         break;
1021                 }
1022                 case 3: /* level */
1023                 {
1024                         trigger = 1;
1025                         break;
1026                 }
1027                 default: /* invalid */
1028                 {
1029                         printk(KERN_WARNING "broken BIOS!!\n");
1030                         trigger = 0;
1031                         break;
1032                 }
1033         }
1034         return trigger;
1035 }
1036 
1037 static inline int irq_polarity(int idx)
1038 {
1039         return MPBIOS_polarity(idx);
1040 }
1041 
1042 static inline int irq_trigger(int idx)
1043 {
1044         return MPBIOS_trigger(idx);
1045 }
1046 
1047 int (*ioapic_renumber_irq)(int ioapic, int irq);
1048 static int pin_2_irq(int idx, int apic, int pin)
1049 {
1050         int irq, i;
1051         int bus = mp_irqs[idx].srcbus;
1052 
1053         /*
1054          * Debugging check, we are in big trouble if this message pops up!
1055          */
1056         if (mp_irqs[idx].dstirq != pin)
1057                 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
1058 
1059         if (test_bit(bus, mp_bus_not_pci)) {
1060                 irq = mp_irqs[idx].srcbusirq;
1061         } else {
1062                 /*
1063                  * PCI IRQs are mapped in order
1064                  */
1065                 i = irq = 0;
1066                 while (i < apic)
1067                         irq += nr_ioapic_registers[i++];
1068                 irq += pin;
1069                 /*
1070                  * For MPS mode, so far only needed by ES7000 platform
1071                  */
1072                 if (ioapic_renumber_irq)
1073                         irq = ioapic_renumber_irq(apic, irq);
1074         }
1075 
1076 #ifdef CONFIG_X86_32
1077         /*
1078          * PCI IRQ command line redirection. Yes, limits are hardcoded.
1079          */
1080         if ((pin >= 16) && (pin <= 23)) {
1081                 if (pirq_entries[pin-16] != -1) {
1082                         if (!pirq_entries[pin-16]) {
1083                                 apic_printk(APIC_VERBOSE, KERN_DEBUG
1084                                                 "disabling PIRQ%d\n", pin-16);
1085                         } else {
1086                                 irq = pirq_entries[pin-16];
1087                                 apic_printk(APIC_VERBOSE, KERN_DEBUG
1088                                                 "using PIRQ%d -> IRQ %d\n",
1089                                                 pin-16, irq);
1090                         }
1091                 }
1092         }
1093 #endif
1094 
1095         return irq;
1096 }
1097 
1098 /*
1099  * Find a specific PCI IRQ entry.
1100  * Not an __init, possibly needed by modules
1101  */
1102 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin,
1103                                 struct io_apic_irq_attr *irq_attr)
1104 {
1105         int apic, i, best_guess = -1;
1106 
1107         apic_printk(APIC_DEBUG,
1108                     "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
1109                     bus, slot, pin);
1110         if (test_bit(bus, mp_bus_not_pci)) {
1111                 apic_printk(APIC_VERBOSE,
1112                             "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
1113                 return -1;
1114         }
1115         for (i = 0; i < mp_irq_entries; i++) {
1116                 int lbus = mp_irqs[i].srcbus;
1117 
1118                 for (apic = 0; apic < nr_ioapics; apic++)
1119                         if (mp_ioapics[apic].apicid == mp_irqs[i].dstapic ||
1120                             mp_irqs[i].dstapic == MP_APIC_ALL)
1121                                 break;
1122 
1123                 if (!test_bit(lbus, mp_bus_not_pci) &&
1124                     !mp_irqs[i].irqtype &&
1125                     (bus == lbus) &&
1126                     (slot == ((mp_irqs[i].srcbusirq >> 2) & 0x1f))) {
1127                         int irq = pin_2_irq(i, apic, mp_irqs[i].dstirq);
1128 
1129                         if (!(apic || IO_APIC_IRQ(irq)))
1130                                 continue;
1131 
1132                         if (pin == (mp_irqs[i].srcbusirq & 3)) {
1133                                 set_io_apic_irq_attr(irq_attr, apic,
1134                                                      mp_irqs[i].dstirq,
1135                                                      irq_trigger(i),
1136                                                      irq_polarity(i));
1137                                 return irq;
1138                         }
1139                         /*
1140                          * Use the first all-but-pin matching entry as a
1141                          * best-guess fuzzy result for broken mptables.
1142                          */
1143                         if (best_guess < 0) {
1144                                 set_io_apic_irq_attr(irq_attr, apic,
1145                                                      mp_irqs[i].dstirq,
1146                                                      irq_trigger(i),
1147                                                      irq_polarity(i));
1148                                 best_guess = irq;
1149                         }
1150                 }
1151         }
1152         return best_guess;
1153 }
1154 EXPORT_SYMBOL(IO_APIC_get_PCI_irq_vector);
1155 
1156 void lock_vector_lock(void)
1157 {
1158         /* Used to the online set of cpus does not change
1159          * during assign_irq_vector.
1160          */
1161         spin_lock(&vector_lock);
1162 }
1163 
1164 void unlock_vector_lock(void)
1165 {
1166         spin_unlock(&vector_lock);
1167 }
1168 
1169 static int
1170 __assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask)
1171 {
1172         /*
1173          * NOTE! The local APIC isn't very good at handling
1174          * multiple interrupts at the same interrupt level.
1175          * As the interrupt level is determined by taking the
1176          * vector number and shifting that right by 4, we
1177          * want to spread these out a bit so that they don't
1178          * all fall in the same interrupt level.
1179          *
1180          * Also, we've got to be careful not to trash gate
1181          * 0x80, because int 0x80 is hm, kind of importantish. ;)
1182          */
1183         static int current_vector = FIRST_DEVICE_VECTOR, current_offset = 0;
1184         unsigned int old_vector;
1185         int cpu, err;
1186         cpumask_var_t tmp_mask;
1187 
1188         if ((cfg->move_in_progress) || cfg->move_cleanup_count)
1189                 return -EBUSY;
1190 
1191         if (!alloc_cpumask_var(&tmp_mask, GFP_ATOMIC))
1192                 return -ENOMEM;
1193 
1194         old_vector = cfg->vector;
1195         if (old_vector) {
1196                 cpumask_and(tmp_mask, mask, cpu_online_mask);
1197                 cpumask_and(tmp_mask, cfg->domain, tmp_mask);
1198                 if (!cpumask_empty(tmp_mask)) {
1199                         free_cpumask_var(tmp_mask);
1200                         return 0;
1201                 }
1202         }
1203 
1204         /* Only try and allocate irqs on cpus that are present */
1205         err = -ENOSPC;
1206         for_each_cpu_and(cpu, mask, cpu_online_mask) {
1207                 int new_cpu;
1208                 int vector, offset;
1209 
1210                 apic->vector_allocation_domain(cpu, tmp_mask);
1211 
1212                 vector = current_vector;
1213                 offset = current_offset;
1214 next:
1215                 vector += 8;
1216                 if (vector >= first_system_vector) {
1217                         /* If out of vectors on large boxen, must share them. */
1218                         offset = (offset + 1) % 8;
1219                         vector = FIRST_DEVICE_VECTOR + offset;
1220                 }
1221                 if (unlikely(current_vector == vector))
1222                         continue;
1223 
1224                 if (test_bit(vector, used_vectors))
1225                         goto next;
1226 
1227                 for_each_cpu_and(new_cpu, tmp_mask, cpu_online_mask)
1228                         if (per_cpu(vector_irq, new_cpu)[vector] != -1)
1229                                 goto next;
1230                 /* Found one! */
1231                 current_vector = vector;
1232                 current_offset = offset;
1233                 if (old_vector) {
1234                         cfg->move_in_progress = 1;
1235                         cpumask_copy(cfg->old_domain, cfg->domain);
1236                 }
1237                 for_each_cpu_and(new_cpu, tmp_mask, cpu_online_mask)
1238                         per_cpu(vector_irq, new_cpu)[vector] = irq;
1239                 cfg->vector = vector;
1240                 cpumask_copy(cfg->domain, tmp_mask);
1241                 err = 0;
1242                 break;
1243         }
1244         free_cpumask_var(tmp_mask);
1245         return err;
1246 }
1247 
1248 static int
1249 assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask)
1250 {
1251         int err;
1252         unsigned long flags;
1253 
1254         spin_lock_irqsave(&vector_lock, flags);
1255         err = __assign_irq_vector(irq, cfg, mask);
1256         spin_unlock_irqrestore(&vector_lock, flags);
1257         return err;
1258 }
1259 
1260 static void __clear_irq_vector(int irq, struct irq_cfg *cfg)
1261 {
1262         int cpu, vector;
1263 
1264         BUG_ON(!cfg->vector);
1265 
1266         vector = cfg->vector;
1267         for_each_cpu_and(cpu, cfg->domain, cpu_online_mask)
1268                 per_cpu(vector_irq, cpu)[vector] = -1;
1269 
1270         cfg->vector = 0;
1271         cpumask_clear(cfg->domain);
1272 
1273         if (likely(!cfg->move_in_progress))
1274                 return;
1275         for_each_cpu_and(cpu, cfg->old_domain, cpu_online_mask) {
1276                 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS;
1277                                                                 vector++) {
1278                         if (per_cpu(vector_irq, cpu)[vector] != irq)
1279                                 continue;
1280                         per_cpu(vector_irq, cpu)[vector] = -1;
1281                         break;
1282                 }
1283         }
1284         cfg->move_in_progress = 0;
1285 }
1286 
1287 void __setup_vector_irq(int cpu)
1288 {
1289         /* Initialize vector_irq on a new cpu */
1290         /* This function must be called with vector_lock held */
1291         int irq, vector;
1292         struct irq_cfg *cfg;
1293         struct irq_desc *desc;
1294 
1295         /* Mark the inuse vectors */
1296         for_each_irq_desc(irq, desc) {
1297                 cfg = desc->chip_data;
1298                 if (!cpumask_test_cpu(cpu, cfg->domain))
1299                         continue;
1300                 vector = cfg->vector;
1301                 per_cpu(vector_irq, cpu)[vector] = irq;
1302         }
1303         /* Mark the free vectors */
1304         for (vector = 0; vector < NR_VECTORS; ++vector) {
1305                 irq = per_cpu(vector_irq, cpu)[vector];
1306                 if (irq < 0)
1307                         continue;
1308 
1309                 cfg = irq_cfg(irq);
1310                 if (!cpumask_test_cpu(cpu, cfg->domain))
1311                         per_cpu(vector_irq, cpu)[vector] = -1;
1312         }
1313 }
1314 
1315 static struct irq_chip ioapic_chip;
1316 static struct irq_chip ir_ioapic_chip;
1317 
1318 #define IOAPIC_AUTO     -1
1319 #define IOAPIC_EDGE     0
1320 #define IOAPIC_LEVEL    1
1321 
1322 #ifdef CONFIG_X86_32
1323 static inline int IO_APIC_irq_trigger(int irq)
1324 {
1325         int apic, idx, pin;
1326 
1327         for (apic = 0; apic < nr_ioapics; apic++) {
1328                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1329                         idx = find_irq_entry(apic, pin, mp_INT);
1330                         if ((idx != -1) && (irq == pin_2_irq(idx, apic, pin)))
1331                                 return irq_trigger(idx);
1332                 }
1333         }
1334         /*
1335          * nonexistent IRQs are edge default
1336          */
1337         return 0;
1338 }
1339 #else
1340 static inline int IO_APIC_irq_trigger(int irq)
1341 {
1342         return 1;
1343 }
1344 #endif
1345 
1346 static void ioapic_register_intr(int irq, struct irq_desc *desc, unsigned long trigger)
1347 {
1348 
1349         if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1350             trigger == IOAPIC_LEVEL)
1351                 desc->status |= IRQ_LEVEL;
1352         else
1353                 desc->status &= ~IRQ_LEVEL;
1354 
1355         if (irq_remapped(irq)) {
1356                 desc->status |= IRQ_MOVE_PCNTXT;
1357                 if (trigger)
1358                         set_irq_chip_and_handler_name(irq, &ir_ioapic_chip,
1359                                                       handle_fasteoi_irq,
1360                                                      "fasteoi");
1361                 else
1362                         set_irq_chip_and_handler_name(irq, &ir_ioapic_chip,
1363                                                       handle_edge_irq, "edge");
1364                 return;
1365         }
1366 
1367         if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1368             trigger == IOAPIC_LEVEL)
1369                 set_irq_chip_and_handler_name(irq, &ioapic_chip,
1370                                               handle_fasteoi_irq,
1371                                               "fasteoi");
1372         else
1373                 set_irq_chip_and_handler_name(irq, &ioapic_chip,
1374                                               handle_edge_irq, "edge");
1375 }
1376 
1377 int setup_ioapic_entry(int apic_id, int irq,
1378                        struct IO_APIC_route_entry *entry,
1379                        unsigned int destination, int trigger,
1380                        int polarity, int vector, int pin)
1381 {
1382         /*
1383          * add it to the IO-APIC irq-routing table:
1384          */
1385         memset(entry,0,sizeof(*entry));
1386 
1387         if (intr_remapping_enabled) {
1388                 struct intel_iommu *iommu = map_ioapic_to_ir(apic_id);
1389                 struct irte irte;
1390                 struct IR_IO_APIC_route_entry *ir_entry =
1391                         (struct IR_IO_APIC_route_entry *) entry;
1392                 int index;
1393 
1394                 if (!iommu)
1395                         panic("No mapping iommu for ioapic %d\n", apic_id);
1396 
1397                 index = alloc_irte(iommu, irq, 1);
1398                 if (index < 0)
1399                         panic("Failed to allocate IRTE for ioapic %d\n", apic_id);
1400 
1401                 memset(&irte, 0, sizeof(irte));
1402 
1403                 irte.present = 1;
1404                 irte.dst_mode = apic->irq_dest_mode;
1405                 /*
1406                  * Trigger mode in the IRTE will always be edge, and the
1407                  * actual level or edge trigger will be setup in the IO-APIC
1408                  * RTE. This will help simplify level triggered irq migration.
1409                  * For more details, see the comments above explainig IO-APIC
1410                  * irq migration in the presence of interrupt-remapping.
1411                  */
1412                 irte.trigger_mode = 0;
1413                 irte.dlvry_mode = apic->irq_delivery_mode;
1414                 irte.vector = vector;
1415                 irte.dest_id = IRTE_DEST(destination);
1416 
1417                 /* Set source-id of interrupt request */
1418                 set_ioapic_sid(&irte, apic_id);
1419 
1420                 modify_irte(irq, &irte);
1421 
1422                 ir_entry->index2 = (index >> 15) & 0x1;
1423                 ir_entry->zero = 0;
1424                 ir_entry->format = 1;
1425                 ir_entry->index = (index & 0x7fff);
1426                 /*
1427                  * IO-APIC RTE will be configured with virtual vector.
1428                  * irq handler will do the explicit EOI to the io-apic.
1429                  */
1430                 ir_entry->vector = pin;
1431         } else {
1432                 entry->delivery_mode = apic->irq_delivery_mode;
1433                 entry->dest_mode = apic->irq_dest_mode;
1434                 entry->dest = destination;
1435                 entry->vector = vector;
1436         }
1437 
1438         entry->mask = 0;                                /* enable IRQ */
1439         entry->trigger = trigger;
1440         entry->polarity = polarity;
1441 
1442         /* Mask level triggered irqs.
1443          * Use IRQ_DELAYED_DISABLE for edge triggered irqs.
1444          */
1445         if (trigger)
1446                 entry->mask = 1;
1447         return 0;
1448 }
1449 
1450 static void setup_IO_APIC_irq(int apic_id, int pin, unsigned int irq, struct irq_desc *desc,
1451                               int trigger, int polarity)
1452 {
1453         struct irq_cfg *cfg;
1454         struct IO_APIC_route_entry entry;
1455         unsigned int dest;
1456 
1457         if (!IO_APIC_IRQ(irq))
1458                 return;
1459 
1460         cfg = desc->chip_data;
1461 
1462         if (assign_irq_vector(irq, cfg, apic->target_cpus()))
1463                 return;
1464 
1465         dest = apic->cpu_mask_to_apicid_and(cfg->domain, apic->target_cpus());
1466 
1467         apic_printk(APIC_VERBOSE,KERN_DEBUG
1468                     "IOAPIC[%d]: Set routing entry (%d-%d -> 0x%x -> "
1469                     "IRQ %d Mode:%i Active:%i)\n",
1470                     apic_id, mp_ioapics[apic_id].apicid, pin, cfg->vector,
1471                     irq, trigger, polarity);
1472 
1473 
1474         if (setup_ioapic_entry(mp_ioapics[apic_id].apicid, irq, &entry,
1475                                dest, trigger, polarity, cfg->vector, pin)) {
1476                 printk("Failed to setup ioapic entry for ioapic  %d, pin %d\n",
1477                        mp_ioapics[apic_id].apicid, pin);
1478                 __clear_irq_vector(irq, cfg);
1479                 return;
1480         }
1481 
1482         ioapic_register_intr(irq, desc, trigger);
1483         if (irq < NR_IRQS_LEGACY)
1484                 disable_8259A_irq(irq);
1485 
1486         ioapic_write_entry(apic_id, pin, entry);
1487 }
1488 
1489 static struct {
1490         DECLARE_BITMAP(pin_programmed, MP_MAX_IOAPIC_PIN + 1);
1491 } mp_ioapic_routing[MAX_IO_APICS];
1492 
1493 static void __init setup_IO_APIC_irqs(void)
1494 {
1495         int apic_id = 0, pin, idx, irq;
1496         int notcon = 0;
1497         struct irq_desc *desc;
1498         struct irq_cfg *cfg;
1499         int node = cpu_to_node(boot_cpu_id);
1500 
1501         apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
1502 
1503 #ifdef CONFIG_ACPI
1504         if (!acpi_disabled && acpi_ioapic) {
1505                 apic_id = mp_find_ioapic(0);
1506                 if (apic_id < 0)
1507                         apic_id = 0;
1508         }
1509 #endif
1510 
1511         for (pin = 0; pin < nr_ioapic_registers[apic_id]; pin++) {
1512                 idx = find_irq_entry(apic_id, pin, mp_INT);
1513                 if (idx == -1) {
1514                         if (!notcon) {
1515                                 notcon = 1;
1516                                 apic_printk(APIC_VERBOSE,
1517                                         KERN_DEBUG " %d-%d",
1518                                         mp_ioapics[apic_id].apicid, pin);
1519                         } else
1520                                 apic_printk(APIC_VERBOSE, " %d-%d",
1521                                         mp_ioapics[apic_id].apicid, pin);
1522                         continue;
1523                 }
1524                 if (notcon) {
1525                         apic_printk(APIC_VERBOSE,
1526                                 " (apicid-pin) not connected\n");
1527                         notcon = 0;
1528                 }
1529 
1530                 irq = pin_2_irq(idx, apic_id, pin);
1531 
1532                 /*
1533                  * Skip the timer IRQ if there's a quirk handler
1534                  * installed and if it returns 1:
1535                  */
1536                 if (apic->multi_timer_check &&
1537                                 apic->multi_timer_check(apic_id, irq))
1538                         continue;
1539 
1540                 desc = irq_to_desc_alloc_node(irq, node);
1541                 if (!desc) {
1542                         printk(KERN_INFO "can not get irq_desc for %d\n", irq);
1543                         continue;
1544                 }
1545                 cfg = desc->chip_data;
1546                 add_pin_to_irq_node(cfg, node, apic_id, pin);
1547                 /*
1548                  * don't mark it in pin_programmed, so later acpi could
1549                  * set it correctly when irq < 16
1550                  */
1551                 setup_IO_APIC_irq(apic_id, pin, irq, desc,
1552                                 irq_trigger(idx), irq_polarity(idx));
1553         }
1554 
1555         if (notcon)
1556                 apic_printk(APIC_VERBOSE,
1557                         " (apicid-pin) not connected\n");
1558 }
1559 
1560 /*
1561  * for the gsit that is not in first ioapic
1562  * but could not use acpi_register_gsi()
1563  * like some special sci in IBM x3330
1564  */
1565 void setup_IO_APIC_irq_extra(u32 gsi)
1566 {
1567         int apic_id = 0, pin, idx, irq;
1568         int node = cpu_to_node(boot_cpu_id);
1569         struct irq_desc *desc;
1570         struct irq_cfg *cfg;
1571 
1572         /*
1573          * Convert 'gsi' to 'ioapic.pin'.
1574          */
1575         apic_id = mp_find_ioapic(gsi);
1576         if (apic_id < 0)
1577                 return;
1578 
1579         pin = mp_find_ioapic_pin(apic_id, gsi);
1580         idx = find_irq_entry(apic_id, pin, mp_INT);
1581         if (idx == -1)
1582                 return;
1583 
1584         irq = pin_2_irq(idx, apic_id, pin);
1585 #ifdef CONFIG_SPARSE_IRQ
1586         desc = irq_to_desc(irq);
1587         if (desc)
1588                 return;
1589 #endif
1590         desc = irq_to_desc_alloc_node(irq, node);
1591         if (!desc) {
1592                 printk(KERN_INFO "can not get irq_desc for %d\n", irq);
1593                 return;
1594         }
1595 
1596         cfg = desc->chip_data;
1597         add_pin_to_irq_node(cfg, node, apic_id, pin);
1598 
1599         if (test_bit(pin, mp_ioapic_routing[apic_id].pin_programmed)) {
1600                 pr_debug("Pin %d-%d already programmed\n",
1601                          mp_ioapics[apic_id].apicid, pin);
1602                 return;
1603         }
1604         set_bit(pin, mp_ioapic_routing[apic_id].pin_programmed);
1605 
1606         setup_IO_APIC_irq(apic_id, pin, irq, desc,
1607                         irq_trigger(idx), irq_polarity(idx));
1608 }
1609 
1610 /*
1611  * Set up the timer pin, possibly with the 8259A-master behind.
1612  */
1613 static void __init setup_timer_IRQ0_pin(unsigned int apic_id, unsigned int pin,
1614                                         int vector)
1615 {
1616         struct IO_APIC_route_entry entry;
1617 
1618         if (intr_remapping_enabled)
1619                 return;
1620 
1621         memset(&entry, 0, sizeof(entry));
1622 
1623         /*
1624          * We use logical delivery to get the timer IRQ
1625          * to the first CPU.
1626          */
1627         entry.dest_mode = apic->irq_dest_mode;
1628         entry.mask = 0;                 /* don't mask IRQ for edge */
1629         entry.dest = apic->cpu_mask_to_apicid(apic->target_cpus());
1630         entry.delivery_mode = apic->irq_delivery_mode;
1631         entry.polarity = 0;
1632         entry.trigger = 0;
1633         entry.vector = vector;
1634 
1635         /*
1636          * The timer IRQ doesn't have to know that behind the
1637          * scene we may have a 8259A-master in AEOI mode ...
1638          */
1639         set_irq_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq, "edge");
1640 
1641         /*
1642          * Add it to the IO-APIC irq-routing table:
1643          */
1644         ioapic_write_entry(apic_id, pin, entry);
1645 }
1646 
1647 
1648 __apicdebuginit(void) print_IO_APIC(void)
1649 {
1650         int apic, i;
1651         union IO_APIC_reg_00 reg_00;
1652         union IO_APIC_reg_01 reg_01;
1653         union IO_APIC_reg_02 reg_02;
1654         union IO_APIC_reg_03 reg_03;
1655         unsigned long flags;
1656         struct irq_cfg *cfg;
1657         struct irq_desc *desc;
1658         unsigned int irq;
1659 
1660         if (apic_verbosity == APIC_QUIET)
1661                 return;
1662 
1663         printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
1664         for (i = 0; i < nr_ioapics; i++)
1665                 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
1666                        mp_ioapics[i].apicid, nr_ioapic_registers[i]);
1667 
1668         /*
1669          * We are a bit conservative about what we expect.  We have to
1670          * know about every hardware change ASAP.
1671          */
1672         printk(KERN_INFO "testing the IO APIC.......................\n");
1673 
1674         for (apic = 0; apic < nr_ioapics; apic++) {
1675 
1676         spin_lock_irqsave(&ioapic_lock, flags);
1677         reg_00.raw = io_apic_read(apic, 0);
1678         reg_01.raw = io_apic_read(apic, 1);
1679         if (reg_01.bits.version >= 0x10)
1680                 reg_02.raw = io_apic_read(apic, 2);
1681         if (reg_01.bits.version >= 0x20)
1682                 reg_03.raw = io_apic_read(apic, 3);
1683         spin_unlock_irqrestore(&ioapic_lock, flags);
1684 
1685         printk("\n");
1686         printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].apicid);
1687         printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
1688         printk(KERN_DEBUG ".......    : physical APIC id: %02X\n", reg_00.bits.ID);
1689         printk(KERN_DEBUG ".......    : Delivery Type: %X\n", reg_00.bits.delivery_type);
1690         printk(KERN_DEBUG ".......    : LTS          : %X\n", reg_00.bits.LTS);
1691 
1692         printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)&reg_01);
1693         printk(KERN_DEBUG ".......     : max redirection entries: %04X\n", reg_01.bits.entries);
1694 
1695         printk(KERN_DEBUG ".......     : PRQ implemented: %X\n", reg_01.bits.PRQ);
1696         printk(KERN_DEBUG ".......     : IO APIC version: %04X\n", reg_01.bits.version);
1697 
1698         /*
1699          * Some Intel chipsets with IO APIC VERSION of 0x1? don't have reg_02,
1700          * but the value of reg_02 is read as the previous read register
1701          * value, so ignore it if reg_02 == reg_01.
1702          */
1703         if (reg_01.bits.version >= 0x10 && reg_02.raw != reg_01.raw) {
1704                 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
1705                 printk(KERN_DEBUG ".......     : arbitration: %02X\n", reg_02.bits.arbitration);
1706         }
1707 
1708         /*
1709          * Some Intel chipsets with IO APIC VERSION of 0x2? don't have reg_02
1710          * or reg_03, but the value of reg_0[23] is read as the previous read
1711          * register value, so ignore it if reg_03 == reg_0[12].
1712          */
1713         if (reg_01.bits.version >= 0x20 && reg_03.raw != reg_02.raw &&
1714             reg_03.raw != reg_01.raw) {
1715                 printk(KERN_DEBUG ".... register #03: %08X\n", reg_03.raw);
1716                 printk(KERN_DEBUG ".......     : Boot DT    : %X\n", reg_03.bits.boot_DT);
1717         }
1718 
1719         printk(KERN_DEBUG ".... IRQ redirection table:\n");
1720 
1721         printk(KERN_DEBUG " NR Dst Mask Trig IRR Pol"
1722                           " Stat Dmod Deli Vect:   \n");
1723 
1724         for (i = 0; i <= reg_01.bits.entries; i++) {
1725                 struct IO_APIC_route_entry entry;
1726 
1727                 entry = ioapic_read_entry(apic, i);
1728 
1729                 printk(KERN_DEBUG " %02x %03X ",
1730                         i,
1731                         entry.dest
1732                 );
1733 
1734                 printk("%1d    %1d    %1d   %1d   %1d    %1d    %1d    %02X\n",
1735                         entry.mask,
1736                         entry.trigger,
1737                         entry.irr,
1738                         entry.polarity,
1739                         entry.delivery_status,
1740                         entry.dest_mode,
1741                         entry.delivery_mode,
1742                         entry.vector
1743                 );
1744         }
1745         }
1746         printk(KERN_DEBUG "IRQ to pin mappings:\n");
1747         for_each_irq_desc(irq, desc) {
1748                 struct irq_pin_list *entry;
1749 
1750                 cfg = desc->chip_data;
1751                 entry = cfg->irq_2_pin;
1752                 if (!entry)
1753                         continue;
1754                 printk(KERN_DEBUG "IRQ%d ", irq);
1755                 for (;;) {
1756                         printk("-> %d:%d", entry->apic, entry->pin);
1757                         if (!entry->next)
1758                                 break;
1759                         entry = entry->next;
1760                 }
1761                 printk("\n");
1762         }
1763 
1764         printk(KERN_INFO ".................................... done.\n");
1765 
1766         return;
1767 }
1768 
1769 __apicdebuginit(void) print_APIC_field(int base)
1770 {
1771         int i;
1772 
1773         if (apic_verbosity == APIC_QUIET)
1774                 return;
1775 
1776         printk(KERN_DEBUG);
1777 
1778         for (i = 0; i < 8; i++)
1779                 printk(KERN_CONT "%08x", apic_read(base + i*0x10));
1780 
1781         printk(KERN_CONT "\n");
1782 }
1783 
1784 __apicdebuginit(void) print_local_APIC(void *dummy)
1785 {
1786         unsigned int i, v, ver, maxlvt;
1787         u64 icr;
1788 
1789         if (apic_verbosity == APIC_QUIET)
1790                 return;
1791 
1792         printk(KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1793                 smp_processor_id(), hard_smp_processor_id());
1794         v = apic_read(APIC_ID);
1795         printk(KERN_INFO "... APIC ID:      %08x (%01x)\n", v, read_apic_id());
1796         v = apic_read(APIC_LVR);
1797         printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1798         ver = GET_APIC_VERSION(v);
1799         maxlvt = lapic_get_maxlvt();
1800 
1801         v = apic_read(APIC_TASKPRI);
1802         printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1803 
1804         if (APIC_INTEGRATED(ver)) {                     /* !82489DX */
1805                 if (!APIC_XAPIC(ver)) {
1806                         v = apic_read(APIC_ARBPRI);
1807                         printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
1808                                v & APIC_ARBPRI_MASK);
1809                 }
1810                 v = apic_read(APIC_PROCPRI);
1811                 printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1812         }
1813 
1814         /*
1815          * Remote read supported only in the 82489DX and local APIC for
1816          * Pentium processors.
1817          */
1818         if (!APIC_INTEGRATED(ver) || maxlvt == 3) {
1819                 v = apic_read(APIC_RRR);
1820                 printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1821         }
1822 
1823         v = apic_read(APIC_LDR);
1824         printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1825         if (!x2apic_enabled()) {
1826                 v = apic_read(APIC_DFR);
1827                 printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1828         }
1829         v = apic_read(APIC_SPIV);
1830         printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1831 
1832         printk(KERN_DEBUG "... APIC ISR field:\n");
1833         print_APIC_field(APIC_ISR);
1834         printk(KERN_DEBUG "... APIC TMR field:\n");
1835         print_APIC_field(APIC_TMR);
1836         printk(KERN_DEBUG "... APIC IRR field:\n");
1837         print_APIC_field(APIC_IRR);
1838 
1839         if (APIC_INTEGRATED(ver)) {             /* !82489DX */
1840                 if (maxlvt > 3)         /* Due to the Pentium erratum 3AP. */
1841                         apic_write(APIC_ESR, 0);
1842 
1843                 v = apic_read(APIC_ESR);
1844                 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1845         }
1846 
1847         icr = apic_icr_read();
1848         printk(KERN_DEBUG "... APIC ICR: %08x\n", (u32)icr);
1849         printk(KERN_DEBUG "... APIC ICR2: %08x\n", (u32)(icr >> 32));
1850 
1851         v = apic_read(APIC_LVTT);
1852         printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1853 
1854         if (maxlvt > 3) {                       /* PC is LVT#4. */
1855                 v = apic_read(APIC_LVTPC);
1856                 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1857         }
1858         v = apic_read(APIC_LVT0);
1859         printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1860         v = apic_read(APIC_LVT1);
1861         printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1862 
1863         if (maxlvt > 2) {                       /* ERR is LVT#3. */
1864                 v = apic_read(APIC_LVTERR);
1865                 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1866         }
1867 
1868         v = apic_read(APIC_TMICT);
1869         printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1870         v = apic_read(APIC_TMCCT);
1871         printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1872         v = apic_read(APIC_TDCR);
1873         printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1874 
1875         if (boot_cpu_has(X86_FEATURE_EXTAPIC)) {
1876                 v = apic_read(APIC_EFEAT);
1877                 maxlvt = (v >> 16) & 0xff;
1878                 printk(KERN_DEBUG "... APIC EFEAT: %08x\n", v);
1879                 v = apic_read(APIC_ECTRL);
1880                 printk(KERN_DEBUG "... APIC ECTRL: %08x\n", v);
1881                 for (i = 0; i < maxlvt; i++) {
1882                         v = apic_read(APIC_EILVTn(i));
1883                         printk(KERN_DEBUG "... APIC EILVT%d: %08x\n", i, v);
1884                 }
1885         }
1886         printk("\n");
1887 }
1888 
1889 __apicdebuginit(void) print_all_local_APICs(void)
1890 {
1891         int cpu;
1892 
1893         preempt_disable();
1894         for_each_online_cpu(cpu)
1895                 smp_call_function_single(cpu, print_local_APIC, NULL, 1);
1896         preempt_enable();
1897 }
1898 
1899 __apicdebuginit(void) print_PIC(void)
1900 {
1901         unsigned int v;
1902         unsigned long flags;
1903 
1904         if (apic_verbosity == APIC_QUIET)
1905                 return;
1906 
1907         printk(KERN_DEBUG "\nprinting PIC contents\n");
1908 
1909         spin_lock_irqsave(&i8259A_lock, flags);
1910 
1911         v = inb(0xa1) << 8 | inb(0x21);
1912         printk(KERN_DEBUG "... PIC  IMR: %04x\n", v);
1913 
1914         v = inb(0xa0) << 8 | inb(0x20);
1915         printk(KERN_DEBUG "... PIC  IRR: %04x\n", v);
1916 
1917         outb(0x0b,0xa0);
1918         outb(0x0b,0x20);
1919         v = inb(0xa0) << 8 | inb(0x20);
1920         outb(0x0a,0xa0);
1921         outb(0x0a,0x20);
1922 
1923         spin_unlock_irqrestore(&i8259A_lock, flags);
1924 
1925         printk(KERN_DEBUG "... PIC  ISR: %04x\n", v);
1926 
1927         v = inb(0x4d1) << 8 | inb(0x4d0);
1928         printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1929 }
1930 
1931 __apicdebuginit(int) print_all_ICs(void)
1932 {
1933         print_PIC();
1934 
1935         /* don't print out if apic is not there */
1936         if (!cpu_has_apic || disable_apic)
1937                 return 0;
1938 
1939         print_all_local_APICs();
1940         print_IO_APIC();
1941 
1942         return 0;
1943 }
1944 
1945 fs_initcall(print_all_ICs);
1946 
1947 
1948 /* Where if anywhere is the i8259 connect in external int mode */
1949 static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
1950 
1951 void __init enable_IO_APIC(void)
1952 {
1953         union IO_APIC_reg_01 reg_01;
1954         int i8259_apic, i8259_pin;
1955         int apic;
1956         unsigned long flags;
1957 
1958         /*
1959          * The number of IO-APIC IRQ registers (== #pins):
1960          */
1961         for (apic = 0; apic < nr_ioapics; apic++) {
1962                 spin_lock_irqsave(&ioapic_lock, flags);
1963                 reg_01.raw = io_apic_read(apic, 1);
1964                 spin_unlock_irqrestore(&ioapic_lock, flags);
1965                 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
1966         }
1967         for(apic = 0; apic < nr_ioapics; apic++) {
1968                 int pin;
1969                 /* See if any of the pins is in ExtINT mode */
1970                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1971                         struct IO_APIC_route_entry entry;
1972                         entry = ioapic_read_entry(apic, pin);
1973 
1974                         /* If the interrupt line is enabled and in ExtInt mode
1975                          * I have found the pin where the i8259 is connected.
1976                          */
1977                         if ((entry.mask == 0) && (entry.delivery_mode == dest_ExtINT)) {
1978                                 ioapic_i8259.apic = apic;
1979                                 ioapic_i8259.pin  = pin;
1980                                 goto found_i8259;
1981                         }
1982                 }
1983         }
1984  found_i8259:
1985         /* Look to see what if the MP table has reported the ExtINT */
1986         /* If we could not find the appropriate pin by looking at the ioapic
1987          * the i8259 probably is not connected the ioapic but give the
1988          * mptable a chance anyway.
1989          */
1990         i8259_pin  = find_isa_irq_pin(0, mp_ExtINT);
1991         i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
1992         /* Trust the MP table if nothing is setup in the hardware */
1993         if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
1994                 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
1995                 ioapic_i8259.pin  = i8259_pin;
1996                 ioapic_i8259.apic = i8259_apic;
1997         }
1998         /* Complain if the MP table and the hardware disagree */
1999         if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
2000                 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
2001         {
2002                 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
2003         }
2004 
2005         /*
2006          * Do not trust the IO-APIC being empty at bootup
2007          */
2008         clear_IO_APIC();
2009 }
2010 
2011 /*
2012  * Not an __init, needed by the reboot code
2013  */
2014 void disable_IO_APIC(void)
2015 {
2016         /*
2017          * Clear the IO-APIC before rebooting:
2018          */
2019         clear_IO_APIC();
2020 
2021         /*
2022          * If the i8259 is routed through an IOAPIC
2023          * Put that IOAPIC in virtual wire mode
2024          * so legacy interrupts can be delivered.
2025          *
2026          * With interrupt-remapping, for now we will use virtual wire A mode,
2027          * as virtual wire B is little complex (need to configure both
2028          * IOAPIC RTE aswell as interrupt-remapping table entry).
2029          * As this gets called during crash dump, keep this simple for now.
2030          */
2031         if (ioapic_i8259.pin != -1 && !intr_remapping_enabled) {
2032                 struct IO_APIC_route_entry entry;
2033 
2034                 memset(&entry, 0, sizeof(entry));
2035                 entry.mask            = 0; /* Enabled */
2036                 entry.trigger         = 0; /* Edge */
2037                 entry.irr             = 0;
2038                 entry.polarity        = 0; /* High */
2039                 entry.delivery_status = 0;
2040                 entry.dest_mode       = 0; /* Physical */
2041                 entry.delivery_mode   = dest_ExtINT; /* ExtInt */
2042                 entry.vector          = 0;
2043                 entry.dest            = read_apic_id();
2044 
2045                 /*
2046                  * Add it to the IO-APIC irq-routing table:
2047                  */
2048                 ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
2049         }
2050 
2051         /*
2052          * Use virtual wire A mode when interrupt remapping is enabled.
2053          */
2054         if (cpu_has_apic)
2055                 disconnect_bsp_APIC(!intr_remapping_enabled &&
2056                                 ioapic_i8259.pin != -1);
2057 }
2058 
2059 #ifdef CONFIG_X86_32
2060 /*
2061  * function to set the IO-APIC physical IDs based on the
2062  * values stored in the MPC table.
2063  *
2064  * by Matt Domsch <Matt_Domsch@dell.com>  Tue Dec 21 12:25:05 CST 1999
2065  */
2066 
2067 static void __init setup_ioapic_ids_from_mpc(void)
2068 {
2069         union IO_APIC_reg_00 reg_00;
2070         physid_mask_t phys_id_present_map;
2071         int apic_id;
2072         int i;
2073         unsigned char old_id;
2074         unsigned long flags;
2075 
2076         if (x86_quirks->setup_ioapic_ids && x86_quirks->setup_ioapic_ids())
2077                 return;
2078 
2079         /*
2080          * Don't check I/O APIC IDs for xAPIC systems.  They have
2081          * no meaning without the serial APIC bus.
2082          */
2083         if (!(boot_cpu_data.x86_vendor == X86_VENDOR_INTEL)
2084                 || APIC_XAPIC(apic_version[boot_cpu_physical_apicid]))
2085                 return;
2086         /*
2087          * This is broken; anything with a real cpu count has to
2088          * circumvent this idiocy regardless.
2089          */
2090         phys_id_present_map = apic->ioapic_phys_id_map(phys_cpu_present_map);
2091 
2092         /*
2093          * Set the IOAPIC ID to the value stored in the MPC table.
2094          */
2095         for (apic_id = 0; apic_id < nr_ioapics; apic_id++) {
2096 
2097                 /* Read the register 0 value */
2098                 spin_lock_irqsave(&ioapic_lock, flags);
2099                 reg_00.raw = io_apic_read(apic_id, 0);
2100                 spin_unlock_irqrestore(&ioapic_lock, flags);
2101 
2102                 old_id = mp_ioapics[apic_id].apicid;
2103 
2104                 if (mp_ioapics[apic_id].apicid >= get_physical_broadcast()) {
2105                         printk(KERN_ERR "BIOS bug, IO-APIC#%d ID is %d in the MPC table!...\n",
2106                                 apic_id, mp_ioapics[apic_id].apicid);
2107                         printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
2108                                 reg_00.bits.ID);
2109                         mp_ioapics[apic_id].apicid = reg_00.bits.ID;
2110                 }
2111 
2112                 /*
2113                  * Sanity check, is the ID really free? Every APIC in a
2114                  * system must have a unique ID or we get lots of nice
2115                  * 'stuck on smp_invalidate_needed IPI wait' messages.
2116                  */
2117                 if (apic->check_apicid_used(phys_id_present_map,
2118                                         mp_ioapics[apic_id].apicid)) {
2119                         printk(KERN_ERR "BIOS bug, IO-APIC#%d ID %d is already used!...\n",
2120                                 apic_id, mp_ioapics[apic_id].apicid);
2121                         for (i = 0; i < get_physical_broadcast(); i++)
2122                                 if (!physid_isset(i, phys_id_present_map))
2123                                         break;
2124                         if (i >= get_physical_broadcast())
2125                                 panic("Max APIC ID exceeded!\n");
2126                         printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
2127                                 i);
2128                         physid_set(i, phys_id_present_map);
2129                         mp_ioapics[apic_id].apicid = i;
2130                 } else {
2131                         physid_mask_t tmp;
2132                         tmp = apic->apicid_to_cpu_present(mp_ioapics[apic_id].apicid);
2133                         apic_printk(APIC_VERBOSE, "Setting %d in the "
2134                                         "phys_id_present_map\n",
2135                                         mp_ioapics[apic_id].apicid);
2136                         physids_or(phys_id_present_map, phys_id_present_map, tmp);
2137                 }
2138 
2139 
2140                 /*
2141                  * We need to adjust the IRQ routing table
2142                  * if the ID changed.
2143                  */
2144                 if (old_id != mp_ioapics[apic_id].apicid)
2145                         for (i = 0; i < mp_irq_entries; i++)
2146                                 if (mp_irqs[i].dstapic == old_id)
2147                                         mp_irqs[i].dstapic
2148                                                 = mp_ioapics[apic_id].apicid;
2149 
2150                 /*
2151                  * Read the right value from the MPC table and
2152                  * write it into the ID register.
2153                  */
2154                 apic_printk(APIC_VERBOSE, KERN_INFO
2155                         "...changing IO-APIC physical APIC ID to %d ...",
2156                         mp_ioapics[apic_id].apicid);
2157 
2158                 reg_00.bits.ID = mp_ioapics[apic_id].apicid;
2159                 spin_lock_irqsave(&ioapic_lock, flags);
2160                 io_apic_write(apic_id, 0, reg_00.raw);
2161                 spin_unlock_irqrestore(&ioapic_lock, flags);
2162 
2163                 /*
2164                  * Sanity check
2165                  */
2166                 spin_lock_irqsave(&ioapic_lock, flags);
2167                 reg_00.raw = io_apic_read(apic_id, 0);
2168                 spin_unlock_irqrestore(&ioapic_lock, flags);
2169                 if (reg_00.bits.ID != mp_ioapics[apic_id].apicid)
2170                         printk("could not set ID!\n");
2171                 else
2172                         apic_printk(APIC_VERBOSE, " ok.\n");
2173         }
2174 }
2175 #endif
2176 
2177 int no_timer_check __initdata;
2178 
2179 static int __init notimercheck(char *s)
2180 {
2181         no_timer_check = 1;
2182         return 1;
2183 }
2184 __setup("no_timer_check", notimercheck);
2185 
2186 /*
2187  * There is a nasty bug in some older SMP boards, their mptable lies
2188  * about the timer IRQ. We do the following to work around the situation:
2189  *
2190  *      - timer IRQ defaults to IO-APIC IRQ
2191  *      - if this function detects that timer IRQs are defunct, then we fall
2192  *        back to ISA timer IRQs
2193  */
2194 static int __init timer_irq_works(void)
2195 {
2196         unsigned long t1 = jiffies;
2197         unsigned long flags;
2198 
2199         if (no_timer_check)
2200                 return 1;
2201 
2202         local_save_flags(flags);
2203         local_irq_enable();
2204         /* Let ten ticks pass... */
2205         mdelay((10 * 1000) / HZ);
2206         local_irq_restore(flags);
2207 
2208         /*
2209          * Expect a few ticks at least, to be sure some possible
2210          * glue logic does not lock up after one or two first
2211          * ticks in a non-ExtINT mode.  Also the local APIC
2212          * might have cached one ExtINT interrupt.  Finally, at
2213          * least one tick may be lost due to delays.
2214          */
2215 
2216         /* jiffies wrap? */
2217         if (time_after(jiffies, t1 + 4))
2218                 return 1;
2219         return 0;
2220 }
2221 
2222 /*
2223  * In the SMP+IOAPIC case it might happen that there are an unspecified
2224  * number of pending IRQ events unhandled. These cases are very rare,
2225  * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
2226  * better to do it this way as thus we do not have to be aware of
2227  * 'pending' interrupts in the IRQ path, except at this point.
2228  */
2229 /*
2230  * Edge triggered needs to resend any interrupt
2231  * that was delayed but this is now handled in the device
2232  * independent code.
2233  */
2234 
2235 /*
2236  * Starting up a edge-triggered IO-APIC interrupt is
2237  * nasty - we need to make sure that we get the edge.
2238  * If it is already asserted for some reason, we need
2239  * return 1 to indicate that is was pending.
2240  *
2241  * This is not complete - we should be able to fake
2242  * an edge even if it isn't on the 8259A...
2243  */
2244 
2245 static unsigned int startup_ioapic_irq(unsigned int irq)
2246 {
2247         int was_pending = 0;
2248         unsigned long flags;
2249         struct irq_cfg *cfg;
2250 
2251         spin_lock_irqsave(&ioapic_lock, flags);
2252         if (irq < NR_IRQS_LEGACY) {
2253                 disable_8259A_irq(irq);
2254                 if (i8259A_irq_pending(irq))
2255                         was_pending = 1;
2256         }
2257         cfg = irq_cfg(irq);
2258         __unmask_IO_APIC_irq(cfg);
2259         spin_unlock_irqrestore(&ioapic_lock, flags);
2260 
2261         return was_pending;
2262 }
2263 
2264 #ifdef CONFIG_X86_64
2265 static int ioapic_retrigger_irq(unsigned int irq)
2266 {
2267 
2268         struct irq_cfg *cfg = irq_cfg(irq);
2269         unsigned long flags;
2270 
2271         spin_lock_irqsave(&vector_lock, flags);
2272         apic->send_IPI_mask(cpumask_of(cpumask_first(cfg->domain)), cfg->vector);
2273         spin_unlock_irqrestore(&vector_lock, flags);
2274 
2275         return 1;
2276 }
2277 #else
2278 static int ioapic_retrigger_irq(unsigned int irq)
2279 {
2280         apic->send_IPI_self(irq_cfg(irq)->vector);
2281 
2282         return 1;
2283 }
2284 #endif
2285 
2286 /*
2287  * Level and edge triggered IO-APIC interrupts need different handling,
2288  * so we use two separate IRQ descriptors. Edge triggered IRQs can be
2289  * handled with the level-triggered descriptor, but that one has slightly
2290  * more overhead. Level-triggered interrupts cannot be handled with the
2291  * edge-triggered handler, without risking IRQ storms and other ugly
2292  * races.
2293  */
2294 
2295 #ifdef CONFIG_SMP
2296 static void send_cleanup_vector(struct irq_cfg *cfg)
2297 {
2298         cpumask_var_t cleanup_mask;
2299 
2300         if (unlikely(!alloc_cpumask_var(&cleanup_mask, GFP_ATOMIC))) {
2301                 unsigned int i;
2302                 cfg->move_cleanup_count = 0;
2303                 for_each_cpu_and(i, cfg->old_domain, cpu_online_mask)
2304                         cfg->move_cleanup_count++;
2305                 for_each_cpu_and(i, cfg->old_domain, cpu_online_mask)
2306                         apic->send_IPI_mask(cpumask_of(i), IRQ_MOVE_CLEANUP_VECTOR);
2307         } else {
2308                 cpumask_and(cleanup_mask, cfg->old_domain, cpu_online_mask);
2309                 cfg->move_cleanup_count = cpumask_weight(cleanup_mask);
2310                 apic->send_IPI_mask(cleanup_mask, IRQ_MOVE_CLEANUP_VECTOR);
2311                 free_cpumask_var(cleanup_mask);
2312         }
2313         cfg->move_in_progress = 0;
2314 }
2315 
2316 static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, struct irq_cfg *cfg)
2317 {
2318         int apic, pin;
2319         struct irq_pin_list *entry;
2320         u8 vector = cfg->vector;
2321 
2322         entry = cfg->irq_2_pin;
2323         for (;;) {
2324                 unsigned int reg;
2325 
2326                 if (!entry)
2327                         break;
2328 
2329                 apic = entry->apic;
2330                 pin = entry->pin;
2331                 /*
2332                  * With interrupt-remapping, destination information comes
2333                  * from interrupt-remapping table entry.
2334                  */
2335                 if (!irq_remapped(irq))
2336                         io_apic_write(apic, 0x11 + pin*2, dest);
2337                 reg = io_apic_read(apic, 0x10 + pin*2);
2338                 reg &= ~IO_APIC_REDIR_VECTOR_MASK;
2339                 reg |= vector;
2340                 io_apic_modify(apic, 0x10 + pin*2, reg);
2341                 if (!entry->next)
2342                         break;
2343                 entry = entry->next;
2344         }
2345 }
2346 
2347 static int
2348 assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask);
2349 
2350 /*
2351  * Either sets desc->affinity to a valid value, and returns
2352  * ->cpu_mask_to_apicid of that, or returns BAD_APICID and
2353  * leaves desc->affinity untouched.
2354  */
2355 static unsigned int
2356 set_desc_affinity(struct irq_desc *desc, const struct cpumask *mask)
2357 {
2358         struct irq_cfg *cfg;
2359         unsigned int irq;
2360 
2361         if (!cpumask_intersects(mask, cpu_online_mask))
2362                 return BAD_APICID;
2363 
2364         irq = desc->irq;
2365         cfg = desc->chip_data;
2366         if (assign_irq_vector(irq, cfg, mask))
2367                 return BAD_APICID;
2368 
2369         cpumask_copy(desc->affinity, mask);
2370 
2371         return apic->cpu_mask_to_apicid_and(desc->affinity, cfg->domain);
2372 }
2373 
2374 static int
2375 set_ioapic_affinity_irq_desc(struct irq_desc *desc, const struct cpumask *mask)
2376 {
2377         struct irq_cfg *cfg;
2378         unsigned long flags;
2379         unsigned int dest;
2380         unsigned int irq;
2381         int ret = -1;
2382 
2383         irq = desc->irq;
2384         cfg = desc->chip_data;
2385 
2386         spin_lock_irqsave(&ioapic_lock, flags);
2387         dest = set_desc_affinity(desc, mask);
2388         if (dest != BAD_APICID) {
2389                 /* Only the high 8 bits are valid. */
2390                 dest = SET_APIC_LOGICAL_ID(dest);
2391                 __target_IO_APIC_irq(irq, dest, cfg);
2392                 ret = 0;
2393         }
2394         spin_unlock_irqrestore(&ioapic_lock, flags);
2395 
2396         return ret;
2397 }
2398 
2399 static int
2400 set_ioapic_affinity_irq(unsigned int irq, const struct cpumask *mask)
2401 {
2402         struct irq_desc *desc;
2403 
2404         desc = irq_to_desc(irq);
2405 
2406         return set_ioapic_affinity_irq_desc(desc, mask);
2407 }
2408 
2409 #ifdef CONFIG_INTR_REMAP
2410 
2411 /*
2412  * Migrate the IO-APIC irq in the presence of intr-remapping.
2413  *
2414  * For both level and edge triggered, irq migration is a simple atomic
2415  * update(of vector and cpu destination) of IRTE and flush the hardware cache.
2416  *
2417  * For level triggered, we eliminate the io-apic RTE modification (with the
2418  * updated vector information), by using a virtual vector (io-apic pin number).
2419  * Real vector that is used for interrupting cpu will be coming from
2420  * the interrupt-remapping table entry.
2421  */
2422 static int
2423 migrate_ioapic_irq_desc(struct irq_desc *desc, const struct cpumask *mask)
2424 {
2425         struct irq_cfg *cfg;
2426         struct irte irte;
2427         unsigned int dest;
2428         unsigned int irq;
2429         int ret = -1;
2430 
2431         if (!cpumask_intersects(mask, cpu_online_mask))
2432                 return ret;
2433 
2434         irq = desc->irq;
2435         if (get_irte(irq, &irte))
2436                 return ret;
2437 
2438         cfg = desc->chip_data;
2439         if (assign_irq_vector(irq, cfg, mask))
2440                 return ret;
2441 
2442         dest = apic->cpu_mask_to_apicid_and(cfg->domain, mask);
2443 
2444         irte.vector = cfg->vector;
2445         irte.dest_id = IRTE_DEST(dest);
2446 
2447         /*
2448          * Modified the IRTE and flushes the Interrupt entry cache.
2449          */
2450         modify_irte(irq, &irte);
2451 
2452         if (cfg->move_in_progress)
2453                 send_cleanup_vector(cfg);
2454 
2455         cpumask_copy(desc->affinity, mask);
2456 
2457         return 0;
2458 }
2459 
2460 /*
2461  * Migrates the IRQ destination in the process context.
2462  */
2463 static int set_ir_ioapic_affinity_irq_desc(struct irq_desc *desc,
2464                                             const struct cpumask *mask)
2465 {
2466         return migrate_ioapic_irq_desc(desc, mask);
2467 }
2468 static int set_ir_ioapic_affinity_irq(unsigned int irq,
2469                                        const struct cpumask *mask)
2470 {
2471         struct irq_desc *desc = irq_to_desc(irq);
2472 
2473         return set_ir_ioapic_affinity_irq_desc(desc, mask);
2474 }
2475 #else
2476 static inline int set_ir_ioapic_affinity_irq_desc(struct irq_desc *desc,
2477                                                    const struct cpumask *mask)
2478 {
2479         return 0;
2480 }
2481 #endif
2482 
2483 asmlinkage void smp_irq_move_cleanup_interrupt(void)
2484 {
2485         unsigned vector, me;
2486 
2487         ack_APIC_irq();
2488         exit_idle();
2489         irq_enter();
2490 
2491         me = smp_processor_id();
2492         for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
2493                 unsigned int irq;
2494                 unsigned int irr;
2495                 struct irq_desc *desc;
2496                 struct irq_cfg *cfg;
2497                 irq = __get_cpu_var(vector_irq)[vector];
2498 
2499                 if (irq == -1)
2500                         continue;
2501 
2502                 desc = irq_to_desc(irq);
2503                 if (!desc)
2504                         continue;
2505 
2506                 cfg = irq_cfg(irq);
2507                 spin_lock(&desc->lock);
2508                 if (!cfg->move_cleanup_count)
2509                         goto unlock;
2510 
2511                 if (vector == cfg->vector && cpumask_test_cpu(me, cfg->domain))
2512                         goto unlock;
2513 
2514                 irr = apic_read(APIC_IRR + (vector / 32 * 0x10));
2515                 /*
2516                  * Check if the vector that needs to be cleanedup is
2517                  * registered at the cpu's IRR. If so, then this is not
2518                  * the best time to clean it up. Lets clean it up in the
2519                  * next attempt by sending another IRQ_MOVE_CLEANUP_VECTOR
2520                  * to myself.
2521                  */
2522                 if (irr  & (1 << (vector % 32))) {
2523                         apic->send_IPI_self(IRQ_MOVE_CLEANUP_VECTOR);
2524                         goto unlock;
2525                 }
2526                 __get_cpu_var(vector_irq)[vector] = -1;
2527                 cfg->move_cleanup_count--;
2528 unlock:
2529                 spin_unlock(&desc->lock);
2530         }
2531 
2532         irq_exit();
2533 }
2534 
2535 static void irq_complete_move(struct irq_desc **descp)
2536 {
2537         struct irq_desc *desc = *descp;
2538         struct irq_cfg *cfg = desc->chip_data;
2539         unsigned vector, me;
2540 
2541         if (likely(!cfg->move_in_progress))
2542                 return;
2543 
2544         vector = ~get_irq_regs()->orig_ax;
2545         me = smp_processor_id();
2546 
2547         if (vector == cfg->vector && cpumask_test_cpu(me, cfg->domain))
2548                 send_cleanup_vector(cfg);
2549 }
2550 #else
2551 static inline void irq_complete_move(struct irq_desc **descp) {}
2552 #endif
2553 
2554 static void ack_apic_edge(unsigned int irq)
2555 {
2556         struct irq_desc *desc = irq_to_desc(irq);
2557 
2558         irq_complete_move(&desc);
2559         move_native_irq(irq);
2560         ack_APIC_irq();
2561 }
2562 
2563 atomic_t irq_mis_count;
2564 
2565 static void ack_apic_level(unsigned int irq)
2566 {
2567         struct irq_desc *desc = irq_to_desc(irq);
2568 
2569 #ifdef CONFIG_X86_32
2570         unsigned long v;
2571         int i;
2572 #endif
2573         struct irq_cfg *cfg;
2574         int do_unmask_irq = 0;
2575 
2576         irq_complete_move(&desc);
2577 #ifdef CONFIG_GENERIC_PENDING_IRQ
2578         /* If we are moving the irq we need to mask it */
2579         if (unlikely(desc->status & IRQ_MOVE_PENDING)) {
2580                 do_unmask_irq = 1;
2581                 mask_IO_APIC_irq_desc(desc);
2582         }
2583 #endif
2584 
2585 #ifdef CONFIG_X86_32
2586         /*
2587         * It appears there is an erratum which affects at least version 0x11
2588         * of I/O APIC (that's the 82093AA and cores integrated into various
2589         * chipsets).  Under certain conditions a level-triggered interrupt is
2590         * erroneously delivered as edge-triggered one but the respective IRR
2591         * bit gets set nevertheless.  As a result the I/O unit expects an EOI
2592         * message but it will never arrive and further interrupts are blocked
2593         * from the source.  The exact reason is so far unknown, but the
2594         * phenomenon was observed when two consecutive interrupt requests
2595         * from a given source get delivered to the same CPU and the source is
2596         * temporarily disabled in between.
2597         *
2598         * A workaround is to simulate an EOI message manually.  We achieve it
2599         * by setting the trigger mode to edge and then to level when the edge
2600         * trigger mode gets detected in the TMR of a local APIC for a
2601         * level-triggered interrupt.  We mask the source for the time of the
2602         * operation to prevent an edge-triggered interrupt escaping meanwhile.
2603         * The idea is from Manfred Spraul.  --macro
2604         */
2605         cfg = desc->chip_data;
2606         i = cfg->vector;
2607 
2608         v = apic_read(APIC_TMR + ((i & ~0x1f) >> 1));
2609 #endif
2610 
2611         /*
2612          * We must acknowledge the irq before we move it or the acknowledge will
2613          * not propagate properly.
2614          */
2615         ack_APIC_irq();
2616 
2617         /* Now we can move and renable the irq */
2618         if (unlikely(do_unmask_irq)) {
2619                 /* Only migrate the irq if the ack has been received.
2620                  *
2621                  * On rare occasions the broadcast level triggered ack gets
2622                  * delayed going to ioapics, and if we reprogram the
2623                  * vector while Remote IRR is still set the irq will never
2624                  * fire again.
2625                  *
2626                  * To prevent this scenario we read the Remote IRR bit
2627                  * of the ioapic.  This has two effects.
2628                  * - On any sane system the read of the ioapic will
2629                  *   flush writes (and acks) going to the ioapic from
2630                  *   this cpu.
2631                  * - We get to see if the ACK has actually been delivered.
2632                  *
2633                  * Based on failed experiments of reprogramming the
2634                  * ioapic entry from outside of irq context starting
2635                  * with masking the ioapic entry and then polling until
2636                  * Remote IRR was clear before reprogramming the
2637                  * ioapic I don't trust the Remote IRR bit to be
2638                  * completey accurate.
2639                  *
2640                  * However there appears to be no other way to plug
2641                  * this race, so if the Remote IRR bit is not
2642                  * accurate and is causing problems then it is a hardware bug
2643                  * and you can go talk to the chipset vendor about it.
2644                  */
2645                 cfg = desc->chip_data;
2646                 if (!io_apic_level_ack_pending(cfg))
2647                         move_masked_irq(irq);
2648                 unmask_IO_APIC_irq_desc(desc);
2649         }
2650 
2651 #ifdef CONFIG_X86_32
2652         if (!(v & (1 << (i & 0x1f)))) {
2653                 atomic_inc(&irq_mis_count);
2654                 spin_lock(&ioapic_lock);
2655                 __mask_and_edge_IO_APIC_irq(cfg);
2656                 __unmask_and_level_IO_APIC_irq(cfg);
2657                 spin_unlock(&ioapic_lock);
2658         }
2659 #endif
2660 }
2661 
2662 #ifdef CONFIG_INTR_REMAP
2663 static void __eoi_ioapic_irq(unsigned int irq, struct irq_cfg *cfg)
2664 {
2665         int apic, pin;
2666         struct irq_pin_list *entry;
2667 
2668         entry = cfg->irq_2_pin;
2669         for (;;) {
2670 
2671                 if (!entry)
2672                         break;
2673 
2674                 apic = entry->apic;
2675                 pin = entry->pin;
2676                 io_apic_eoi(apic, pin);
2677                 entry = entry->next;
2678         }
2679 }
2680 
2681 static void
2682 eoi_ioapic_irq(struct irq_desc *desc)
2683 {
2684         struct irq_cfg *cfg;
2685         unsigned long flags;
2686         unsigned int irq;
2687 
2688         irq = desc->irq;
2689         cfg = desc->chip_data;
2690 
2691         spin_lock_irqsave(&ioapic_lock, flags);
2692         __eoi_ioapic_irq(irq, cfg);
2693         spin_unlock_irqrestore(&ioapic_lock, flags);
2694 }
2695 
2696 static void ir_ack_apic_edge(unsigned int irq)
2697 {
2698         ack_APIC_irq();
2699 }
2700 
2701 static void ir_ack_apic_level(unsigned int irq)
2702 {
2703         struct irq_desc *desc = irq_to_desc(irq);
2704 
2705         ack_APIC_irq();
2706         eoi_ioapic_irq(desc);
2707 }
2708 #endif /* CONFIG_INTR_REMAP */
2709 
2710 static struct irq_chip ioapic_chip __read_mostly = {
2711         .name           = "IO-APIC",
2712         .startup        = startup_ioapic_irq,
2713         .mask           = mask_IO_APIC_irq,
2714         .unmask         = unmask_IO_APIC_irq,
2715         .ack            = ack_apic_edge,
2716         .eoi            = ack_apic_level,
2717 #ifdef CONFIG_SMP
2718         .set_affinity   = set_ioapic_affinity_irq,
2719 #endif
2720         .retrigger      = ioapic_retrigger_irq,
2721 };
2722 
2723 static struct irq_chip ir_ioapic_chip __read_mostly = {
2724         .name           = "IR-IO-APIC",
2725         .startup        = startup_ioapic_irq,
2726         .mask           = mask_IO_APIC_irq,
2727         .unmask         = unmask_IO_APIC_irq,
2728 #ifdef CONFIG_INTR_REMAP
2729         .ack            = ir_ack_apic_edge,
2730         .eoi            = ir_ack_apic_level,
2731 #ifdef CONFIG_SMP
2732         .set_affinity   = set_ir_ioapic_affinity_irq,
2733 #endif
2734 #endif
2735         .retrigger      = ioapic_retrigger_irq,
2736 };
2737 
2738 static inline void init_IO_APIC_traps(void)
2739 {
2740         int irq;
2741         struct irq_desc *desc;
2742         struct irq_cfg *cfg;
2743 
2744         /*
2745          * NOTE! The local APIC isn't very good at handling
2746          * multiple interrupts at the same interrupt level.
2747          * As the interrupt level is determined by taking the
2748          * vector number and shifting that right by 4, we
2749          * want to spread these out a bit so that they don't
2750          * all fall in the same interrupt level.
2751          *
2752          * Also, we've got to be careful not to trash gate
2753          * 0x80, because int 0x80 is hm, kind of importantish. ;)
2754          */
2755         for_each_irq_desc(irq, desc) {
2756                 cfg = desc->chip_data;
2757                 if (IO_APIC_IRQ(irq) && cfg && !cfg->vector) {
2758                         /*
2759                          * Hmm.. We don't have an entry for this,
2760                          * so default to an old-fashioned 8259
2761                          * interrupt if we can..
2762                          */
2763                         if (irq < NR_IRQS_LEGACY)
2764                                 make_8259A_irq(irq);
2765                         else
2766                                 /* Strange. Oh, well.. */
2767                                 desc->chip = &no_irq_chip;
2768                 }
2769         }
2770 }
2771 
2772 /*
2773  * The local APIC irq-chip implementation:
2774  */
2775 
2776 static void mask_lapic_irq(unsigned int irq)
2777 {
2778         unsigned long v;
2779 
2780         v = apic_read(APIC_LVT0);
2781         apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
2782 }
2783 
2784 static void unmask_lapic_irq(unsigned int irq)
2785 {
2786         unsigned long v;
2787 
2788         v = apic_read(APIC_LVT0);
2789         apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
2790 }
2791 
2792 static void ack_lapic_irq(unsigned int irq)
2793 {
2794         ack_APIC_irq();
2795 }
2796 
2797 static struct irq_chip lapic_chip __read_mostly = {
2798         .name           = "local-APIC",
2799         .mask           = mask_lapic_irq,
2800         .unmask         = unmask_lapic_irq,
2801         .ack            = ack_lapic_irq,
2802 };
2803 
2804 static void lapic_register_intr(int irq, struct irq_desc *desc)
2805 {
2806         desc->status &= ~IRQ_LEVEL;
2807         set_irq_chip_and_handler_name(irq, &lapic_chip, handle_edge_irq,
2808                                       "edge");
2809 }
2810 
2811 static void __init setup_nmi(void)
2812 {
2813         /*
2814          * Dirty trick to enable the NMI watchdog ...
2815          * We put the 8259A master into AEOI mode and
2816          * unmask on all local APICs LVT0 as NMI.
2817          *
2818          * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
2819          * is from Maciej W. Rozycki - so we do not have to EOI from
2820          * the NMI handler or the timer interrupt.
2821          */
2822         apic_printk(APIC_VERBOSE, KERN_INFO "activating NMI Watchdog ...");
2823 
2824         enable_NMI_through_LVT0();
2825 
2826         apic_printk(APIC_VERBOSE, " done.\n");
2827 }
2828 
2829 /*
2830  * This looks a bit hackish but it's about the only one way of sending
2831  * a few INTA cycles to 8259As and any associated glue logic.  ICR does
2832  * not support the ExtINT mode, unfortunately.  We need to send these
2833  * cycles as some i82489DX-based boards have glue logic that keeps the
2834  * 8259A interrupt line asserted until INTA.  --macro
2835  */
2836 static inline void __init unlock_ExtINT_logic(void)
2837 {
2838         int apic, pin, i;
2839         struct IO_APIC_route_entry entry0, entry1;
2840         unsigned char save_control, save_freq_select;
2841 
2842         pin  = find_isa_irq_pin(8, mp_INT);
2843         if (pin == -1) {
2844                 WARN_ON_ONCE(1);
2845                 return;
2846         }
2847         apic = find_isa_irq_apic(8, mp_INT);
2848         if (apic == -1) {
2849                 WARN_ON_ONCE(1);
2850                 return;
2851         }
2852 
2853         entry0 = ioapic_read_entry(apic, pin);
2854         clear_IO_APIC_pin(apic, pin);
2855 
2856         memset(&entry1, 0, sizeof(entry1));
2857 
2858         entry1.dest_mode = 0;                   /* physical delivery */
2859         entry1.mask = 0;                        /* unmask IRQ now */
2860         entry1.dest = hard_smp_processor_id();
2861         entry1.delivery_mode = dest_ExtINT;
2862         entry1.polarity = entry0.polarity;
2863         entry1.trigger = 0;
2864         entry1.vector = 0;
2865 
2866         ioapic_write_entry(apic, pin, entry1);
2867 
2868         save_control = CMOS_READ(RTC_CONTROL);
2869         save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
2870         CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
2871                    RTC_FREQ_SELECT);
2872         CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
2873 
2874         i = 100;
2875         while (i-- > 0) {
2876                 mdelay(10);
2877                 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
2878                         i -= 10;
2879         }
2880 
2881         CMOS_WRITE(save_control, RTC_CONTROL);
2882         CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
2883         clear_IO_APIC_pin(apic, pin);
2884 
2885         ioapic_write_entry(apic, pin, entry0);
2886 }
2887 
2888 static int disable_timer_pin_1 __initdata;
2889 /* Actually the next is obsolete, but keep it for paranoid reasons -AK */
2890 static int __init disable_timer_pin_setup(char *arg)
2891 {
2892         disable_timer_pin_1 = 1;
2893         return 0;
2894 }
2895 early_param("disable_timer_pin_1", disable_timer_pin_setup);
2896 
2897 int timer_through_8259 __initdata;
2898 
2899 /*
2900  * This code may look a bit paranoid, but it's supposed to cooperate with
2901  * a wide range of boards and BIOS bugs.  Fortunately only the timer IRQ
2902  * is so screwy.  Thanks to Brian Perkins for testing/hacking this beast
2903  * fanatically on his truly buggy board.
2904  *
2905  * FIXME: really need to revamp this for all platforms.
2906  */
2907 static inline void __init check_timer(void)
2908 {
2909         struct irq_desc *desc = irq_to_desc(0);
2910         struct irq_cfg *cfg = desc->chip_data;
2911         int node = cpu_to_node(boot_cpu_id);
2912         int apic1, pin1, apic2, pin2;
2913         unsigned long flags;
2914         int no_pin1 = 0;
2915 
2916         local_irq_save(flags);
2917 
2918         /*
2919          * get/set the timer IRQ vector:
2920          */
2921         disable_8259A_irq(0);
2922         assign_irq_vector(0, cfg, apic->target_cpus());
2923 
2924         /*
2925          * As IRQ0 is to be enabled in the 8259A, the virtual
2926          * wire has to be disabled in the local APIC.  Also
2927          * timer interrupts need to be acknowledged manually in
2928          * the 8259A for the i82489DX when using the NMI
2929          * watchdog as that APIC treats NMIs as level-triggered.
2930          * The AEOI mode will finish them in the 8259A
2931          * automatically.
2932          */
2933         apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
2934         init_8259A(1);
2935 #ifdef CONFIG_X86_32
2936         {
2937                 unsigned int ver;
2938 
2939                 ver = apic_read(APIC_LVR);
2940                 ver = GET_APIC_VERSION(ver);
2941                 timer_ack = (nmi_watchdog == NMI_IO_APIC && !APIC_INTEGRATED(ver));
2942         }
2943 #endif
2944 
2945         pin1  = find_isa_irq_pin(0, mp_INT);
2946         apic1 = find_isa_irq_apic(0, mp_INT);
2947         pin2  = ioapic_i8259.pin;
2948         apic2 = ioapic_i8259.apic;
2949 
2950         apic_printk(APIC_QUIET, KERN_INFO "..TIMER: vector=0x%02X "
2951                     "apic1=%d pin1=%d apic2=%d pin2=%d\n",
2952                     cfg->vector, apic1, pin1, apic2, pin2);
2953 
2954         /*
2955          * Some BIOS writers are clueless and report the ExtINTA
2956          * I/O APIC input from the cascaded 8259A as the timer
2957          * interrupt input.  So just in case, if only one pin
2958          * was found above, try it both directly and through the
2959          * 8259A.
2960          */
2961         if (pin1 == -1) {
2962                 if (intr_remapping_enabled)
2963                         panic("BIOS bug: timer not connected to IO-APIC");
2964                 pin1 = pin2;
2965                 apic1 = apic2;
2966                 no_pin1 = 1;
2967         } else if (pin2 == -1) {
2968                 pin2 = pin1;
2969                 apic2 = apic1;
2970         }
2971 
2972         if (pin1 != -1) {
2973                 /*
2974                  * Ok, does IRQ0 through the IOAPIC work?
2975                  */
2976                 if (no_pin1) {
2977                         add_pin_to_irq_node(cfg, node, apic1, pin1);
2978                         setup_timer_IRQ0_pin(apic1, pin1, cfg->vector);
2979                 } else {
2980                         /* for edge trigger, setup_IO_APIC_irq already
2981                          * leave it unmasked.
2982                          * so only need to unmask if it is level-trigger
2983                          * do we really have level trigger timer?
2984                          */
2985                         int idx;
2986                         idx = find_irq_entry(apic1, pin1, mp_INT);
2987                         if (idx != -1 && irq_trigger(idx))
2988                                 unmask_IO_APIC_irq_desc(desc);
2989                 }
2990                 if (timer_irq_works()) {
2991                         if (nmi_watchdog == NMI_IO_APIC) {
2992                                 setup_nmi();
2993                                 enable_8259A_irq(0);
2994                         }
2995                         if (disable_timer_pin_1 > 0)
2996                                 clear_IO_APIC_pin(0, pin1);
2997                         goto out;
2998                 }
2999                 if (intr_remapping_enabled)
3000                         panic("timer doesn't work through Interrupt-remapped IO-APIC");
3001                 local_irq_disable();
3002                 clear_IO_APIC_pin(apic1, pin1);
3003                 if (!no_pin1)
3004                         apic_printk(APIC_QUIET, KERN_ERR "..MP-BIOS bug: "
3005                                     "8254 timer not connected to IO-APIC\n");
3006 
3007                 apic_printk(APIC_QUIET, KERN_INFO "...trying to set up timer "
3008                             "(IRQ0) through the 8259A ...\n");
3009                 apic_printk(APIC_QUIET, KERN_INFO
3010                             "..... (found apic %d pin %d) ...\n", apic2, pin2);
3011                 /*
3012                  * legacy devices should be connected to IO APIC #0
3013                  */
3014                 replace_pin_at_irq_node(cfg, node, apic1, pin1, apic2, pin2);
3015                 setup_timer_IRQ0_pin(apic2, pin2, cfg->vector);
3016                 enable_8259A_irq(0);
3017                 if (timer_irq_works()) {
3018                         apic_printk(APIC_QUIET, KERN_INFO "....... works.\n");
3019                         timer_through_8259 = 1;
3020                         if (nmi_watchdog == NMI_IO_APIC) {
3021                                 disable_8259A_irq(0);
3022                                 setup_nmi();
3023                                 enable_8259A_irq(0);
3024                         }
3025                         goto out;
3026                 }
3027                 /*
3028                  * Cleanup, just in case ...
3029                  */
3030                 local_irq_disable();
3031                 disable_8259A_irq(0);
3032                 clear_IO_APIC_pin(apic2, pin2);
3033                 apic_printk(APIC_QUIET, KERN_INFO "....... failed.\n");
3034         }
3035 
3036         if (nmi_watchdog == NMI_IO_APIC) {
3037                 apic_printk(APIC_QUIET, KERN_WARNING "timer doesn't work "
3038                             "through the IO-APIC - disabling NMI Watchdog!\n");
3039                 nmi_watchdog = NMI_NONE;
3040         }
3041 #ifdef CONFIG_X86_32
3042         timer_ack = 0;
3043 #endif
3044 
3045         apic_printk(APIC_QUIET, KERN_INFO
3046                     "...trying to set up timer as Virtual Wire IRQ...\n");
3047 
3048         lapic_register_intr(0, desc);
3049         apic_write(APIC_LVT0, APIC_DM_FIXED | cfg->vector);     /* Fixed mode */
3050         enable_8259A_irq(0);
3051 
3052         if (timer_irq_works()) {
3053                 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
3054                 goto out;
3055         }
3056         local_irq_disable();
3057         disable_8259A_irq(0);
3058         apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | cfg->vector);
3059         apic_printk(APIC_QUIET, KERN_INFO "..... failed.\n");
3060 
3061         apic_printk(APIC_QUIET, KERN_INFO
3062                     "...trying to set up timer as ExtINT IRQ...\n");
3063 
3064         init_8259A(0);
3065         make_8259A_irq(0);
3066         apic_write(APIC_LVT0, APIC_DM_EXTINT);
3067 
3068         unlock_ExtINT_logic();
3069 
3070         if (timer_irq_works()) {
3071                 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
3072                 goto out;
3073         }
3074         local_irq_disable();
3075         apic_printk(APIC_QUIET, KERN_INFO "..... failed :(.\n");
3076         panic("IO-APIC + timer doesn't work!  Boot with apic=debug and send a "
3077                 "report.  Then try booting with the 'noapic' option.\n");
3078 out:
3079         local_irq_restore(flags);
3080 }
3081 
3082 /*
3083  * Traditionally ISA IRQ2 is the cascade IRQ, and is not available
3084  * to devices.  However there may be an I/O APIC pin available for
3085  * this interrupt regardless.  The pin may be left unconnected, but
3086  * typically it will be reused as an ExtINT cascade interrupt for
3087  * the master 8259A.  In the MPS case such a pin will normally be
3088  * reported as an ExtINT interrupt in the MP table.  With ACPI
3089  * there is no provision for ExtINT interrupts, and in the absence
3090  * of an override it would be treated as an ordinary ISA I/O APIC
3091  * interrupt, that is edge-triggered and unmasked by default.  We
3092  * used to do this, but it caused problems on some systems because
3093  * of the NMI watchdog and sometimes IRQ0 of the 8254 timer using
3094  * the same ExtINT cascade interrupt to drive the local APIC of the
3095  * bootstrap processor.  Therefore we refrain from routing IRQ2 to
3096  * the I/O APIC in all cases now.  No actual device should request
3097  * it anyway.  --macro
3098  */
3099 #define PIC_IRQS        (1 << PIC_CASCADE_IR)
3100 
3101 void __init setup_IO_APIC(void)
3102 {
3103 
3104         /*
3105          * calling enable_IO_APIC() is moved to setup_local_APIC for BP
3106          */
3107 
3108         io_apic_irqs = ~PIC_IRQS;
3109 
3110         apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
3111         /*
3112          * Set up IO-APIC IRQ routing.
3113          */
3114 #ifdef CONFIG_X86_32
3115         if (!acpi_ioapic)
3116                 setup_ioapic_ids_from_mpc();
3117 #endif
3118         sync_Arb_IDs();
3119         setup_IO_APIC_irqs();
3120         init_IO_APIC_traps();
3121         check_timer();
3122 }
3123 
3124 /*
3125  *      Called after all the initialization is done. If we didnt find any
3126  *      APIC bugs then we can allow the modify fast path
3127  */
3128 
3129 static int __init io_apic_bug_finalize(void)
3130 {
3131         if (sis_apic_bug == -1)
3132                 sis_apic_bug = 0;
3133         return 0;
3134 }
3135 
3136 late_initcall(io_apic_bug_finalize);
3137 
3138 struct sysfs_ioapic_data {
3139         struct sys_device dev;
3140         struct IO_APIC_route_entry entry[0];
3141 };
3142 static struct sysfs_ioapic_data * mp_ioapic_data[MAX_IO_APICS];
3143 
3144 static int ioapic_suspend(struct sys_device *dev, pm_message_t state)
3145 {
3146         struct IO_APIC_route_entry *entry;
3147         struct sysfs_ioapic_data *data;
3148         int i;
3149 
3150         data = container_of(dev, struct sysfs_ioapic_data, dev);
3151         entry = data->entry;
3152         for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ )
3153                 *entry = ioapic_read_entry(dev->id, i);
3154 
3155         return 0;
3156 }
3157 
3158 static int ioapic_resume(struct sys_device *dev)
3159 {
3160         struct IO_APIC_route_entry *entry;
3161         struct sysfs_ioapic_data *data;
3162         unsigned long flags;
3163         union IO_APIC_reg_00 reg_00;
3164         int i;
3165 
3166         data = container_of(dev, struct sysfs_ioapic_data, dev);
3167         entry = data->entry;
3168 
3169         spin_lock_irqsave(&ioapic_lock, flags);
3170         reg_00.raw = io_apic_read(dev->id, 0);
3171         if (reg_00.bits.ID != mp_ioapics[dev->id].apicid) {
3172                 reg_00.bits.ID = mp_ioapics[dev->id].apicid;
3173                 io_apic_write(dev->id, 0, reg_00.raw);
3174         }
3175         spin_unlock_irqrestore(&ioapic_lock, flags);
3176         for (i = 0; i < nr_ioapic_registers[dev->id]; i++)
3177                 ioapic_write_entry(dev->id, i, entry[i]);
3178 
3179         return 0;
3180 }
3181 
3182 static struct sysdev_class ioapic_sysdev_class = {
3183         .name = "ioapic",
3184         .suspend = ioapic_suspend,
3185         .resume = ioapic_resume,
3186 };
3187 
3188 static int __init ioapic_init_sysfs(void)
3189 {
3190         struct sys_device * dev;
3191         int i, size, error;
3192 
3193         error = sysdev_class_register(&ioapic_sysdev_class);
3194         if (error)
3195                 return error;
3196 
3197         for (i = 0; i < nr_ioapics; i++ ) {
3198                 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
3199                         * sizeof(struct IO_APIC_route_entry);
3200                 mp_ioapic_data[i] = kzalloc(size, GFP_KERNEL);
3201                 if (!mp_ioapic_data[i]) {
3202                         printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
3203                         continue;
3204                 }
3205                 dev = &mp_ioapic_data[i]->dev;
3206                 dev->id = i;
3207                 dev->cls = &ioapic_sysdev_class;
3208                 error = sysdev_register(dev);
3209                 if (error) {
3210                         kfree(mp_ioapic_data[i]);
3211                         mp_ioapic_data[i] = NULL;
3212                         printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
3213                         continue;
3214                 }
3215         }
3216 
3217         return 0;
3218 }
3219 
3220 device_initcall(ioapic_init_sysfs);
3221 
3222 static int nr_irqs_gsi = NR_IRQS_LEGACY;
3223 /*
3224  * Dynamic irq allocate and deallocation
3225  */
3226 unsigned int create_irq_nr(unsigned int irq_want, int node)
3227 {
3228         /* Allocate an unused irq */
3229         unsigned int irq;
3230         unsigned int new;
3231         unsigned long flags;
3232         struct irq_cfg *cfg_new = NULL;
3233         struct irq_desc *desc_new = NULL;
3234 
3235         irq = 0;
3236         if (irq_want < nr_irqs_gsi)
3237                 irq_want = nr_irqs_gsi;
3238 
3239         spin_lock_irqsave(&vector_lock, flags);
3240         for (new = irq_want; new < nr_irqs; new++) {
3241                 desc_new = irq_to_desc_alloc_node(new, node);
3242                 if (!desc_new) {
3243                         printk(KERN_INFO "can not get irq_desc for %d\n", new);
3244                         continue;
3245                 }
3246                 cfg_new = desc_new->chip_data;
3247 
3248                 if (cfg_new->vector != 0)
3249                         continue;
3250 
3251                 desc_new = move_irq_desc(desc_new, node);
3252 
3253                 if (__assign_irq_vector(new, cfg_new, apic->target_cpus()) == 0)
3254                         irq = new;
3255                 break;
3256         }
3257         spin_unlock_irqrestore(&vector_lock, flags);
3258 
3259         if (irq > 0) {
3260                 dynamic_irq_init(irq);
3261                 /* restore it, in case dynamic_irq_init clear it */
3262                 if (desc_new)
3263                         desc_new->chip_data = cfg_new;
3264         }
3265         return irq;
3266 }
3267 
3268 int create_irq(void)
3269 {
3270         int node = cpu_to_node(boot_cpu_id);
3271         unsigned int irq_want;
3272         int irq;
3273 
3274         irq_want = nr_irqs_gsi;
3275         irq = create_irq_nr(irq_want, node);
3276 
3277         if (irq == 0)
3278                 irq = -1;
3279 
3280         return irq;
3281 }
3282 
3283 void destroy_irq(unsigned int irq)
3284 {
3285         unsigned long flags;
3286         struct irq_cfg *cfg;
3287         struct irq_desc *desc;
3288 
3289         /* store it, in case dynamic_irq_cleanup clear it */
3290         desc = irq_to_desc(irq);
3291         cfg = desc->chip_data;
3292         dynamic_irq_cleanup(irq);
3293         /* connect back irq_cfg */
3294         if (desc)
3295                 desc->chip_data = cfg;
3296 
3297         free_irte(irq);
3298         spin_lock_irqsave(&vector_lock, flags);
3299         __clear_irq_vector(irq, cfg);
3300         spin_unlock_irqrestore(&vector_lock, flags);
3301 }
3302 
3303 /*
3304  * MSI message composition
3305  */
3306 #ifdef CONFIG_PCI_MSI
3307 static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq, struct msi_msg *msg)
3308 {
3309         struct irq_cfg *cfg;
3310         int err;
3311         unsigned dest;
3312 
3313         if (disable_apic)
3314                 return -ENXIO;
3315 
3316         cfg = irq_cfg(irq);
3317         err = assign_irq_vector(irq, cfg, apic->target_cpus());
3318         if (err)
3319                 return err;
3320 
3321         dest = apic->cpu_mask_to_apicid_and(cfg->domain, apic->target_cpus());
3322 
3323         if (irq_remapped(irq)) {
3324                 struct irte irte;
3325                 int ir_index;
3326                 u16 sub_handle;
3327 
3328                 ir_index = map_irq_to_irte_handle(irq, &sub_handle);
3329                 BUG_ON(ir_index == -1);
3330 
3331                 memset (&irte, 0, sizeof(irte));
3332 
3333                 irte.present = 1;
3334                 irte.dst_mode = apic->irq_dest_mode;
3335                 irte.trigger_mode = 0; /* edge */
3336                 irte.dlvry_mode = apic->irq_delivery_mode;
3337                 irte.vector = cfg->vector;
3338                 irte.dest_id = IRTE_DEST(dest);
3339 
3340                 /* Set source-id of interrupt request */
3341                 set_msi_sid(&irte, pdev);
3342 
3343                 modify_irte(irq, &irte);
3344 
3345                 msg->address_hi = MSI_ADDR_BASE_HI;
3346                 msg->data = sub_handle;
3347                 msg->address_lo = MSI_ADDR_BASE_LO | MSI_ADDR_IR_EXT_INT |
3348                                   MSI_ADDR_IR_SHV |
3349                                   MSI_ADDR_IR_INDEX1(ir_index) |
3350                                   MSI_ADDR_IR_INDEX2(ir_index);
3351         } else {
3352                 if (x2apic_enabled())
3353                         msg->address_hi = MSI_ADDR_BASE_HI |
3354                                           MSI_ADDR_EXT_DEST_ID(dest);
3355                 else
3356                         msg->address_hi = MSI_ADDR_BASE_HI;
3357 
3358                 msg->address_lo =
3359                         MSI_ADDR_BASE_LO |
3360                         ((apic->irq_dest_mode == 0) ?
3361                                 MSI_ADDR_DEST_MODE_PHYSICAL:
3362                                 MSI_ADDR_DEST_MODE_LOGICAL) |
3363                         ((apic->irq_delivery_mode != dest_LowestPrio) ?
3364                                 MSI_ADDR_REDIRECTION_CPU:
3365                                 MSI_ADDR_REDIRECTION_LOWPRI) |
3366                         MSI_ADDR_DEST_ID(dest);
3367 
3368                 msg->data =
3369                         MSI_DATA_TRIGGER_EDGE |
3370                         MSI_DATA_LEVEL_ASSERT |
3371                         ((apic->irq_delivery_mode != dest_LowestPrio) ?
3372                                 MSI_DATA_DELIVERY_FIXED:
3373                                 MSI_DATA_DELIVERY_LOWPRI) |
3374                         MSI_DATA_VECTOR(cfg->vector);
3375         }
3376         return err;
3377 }
3378 
3379 #ifdef CONFIG_SMP
3380 static int set_msi_irq_affinity(unsigned int irq, const struct cpumask *mask)
3381 {
3382         struct irq_desc *desc = irq_to_desc(irq);
3383         struct irq_cfg *cfg;
3384         struct msi_msg msg;
3385         unsigned int dest;
3386 
3387         dest = set_desc_affinity(desc, mask);
3388         if (dest == BAD_APICID)
3389                 return -1;
3390 
3391         cfg = desc->chip_data;
3392 
3393         read_msi_msg_desc(desc, &msg);
3394 
3395         msg.data &= ~MSI_DATA_VECTOR_MASK;
3396         msg.data |= MSI_DATA_VECTOR(cfg->vector);
3397         msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3398         msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3399 
3400         write_msi_msg_desc(desc, &msg);
3401 
3402         return 0;
3403 }
3404 #ifdef CONFIG_INTR_REMAP
3405 /*
3406  * Migrate the MSI irq to another cpumask. This migration is
3407  * done in the process context using interrupt-remapping hardware.
3408  */
3409 static int
3410 ir_set_msi_irq_affinity(unsigned int irq, const struct cpumask *mask)
3411 {
3412         struct irq_desc *desc = irq_to_desc(irq);
3413         struct irq_cfg *cfg = desc->chip_data;
3414         unsigned int dest;
3415         struct irte irte;
3416 
3417         if (get_irte(irq, &irte))
3418                 return -1;
3419 
3420         dest = set_desc_affinity(desc, mask);
3421         if (dest == BAD_APICID)
3422                 return -1;
3423 
3424         irte.vector = cfg->vector;
3425         irte.dest_id = IRTE_DEST(dest);
3426 
3427         /*
3428          * atomically update the IRTE with the new destination and vector.
3429          */
3430         modify_irte(irq, &irte);
3431 
3432         /*
3433          * After this point, all the interrupts will start arriving
3434          * at the new destination. So, time to cleanup the previous
3435          * vector allocation.
3436          */
3437         if (cfg->move_in_progress)
3438                 send_cleanup_vector(cfg);
3439 
3440         return 0;
3441 }
3442 
3443 #endif
3444 #endif /* CONFIG_SMP */
3445 
3446 /*
3447  * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
3448  * which implement the MSI or MSI-X Capability Structure.
3449  */
3450 static struct irq_chip msi_chip = {
3451         .name           = "PCI-MSI",
3452         .unmask         = unmask_msi_irq,
3453         .mask           = mask_msi_irq,
3454         .ack            = ack_apic_edge,
3455 #ifdef CONFIG_SMP
3456         .set_affinity   = set_msi_irq_affinity,
3457 #endif
3458         .retrigger      = ioapic_retrigger_irq,
3459 };
3460 
3461 static struct irq_chip msi_ir_chip = {
3462         .name           = "IR-PCI-MSI",
3463         .unmask         = unmask_msi_irq,
3464         .mask           = mask_msi_irq,
3465 #ifdef CONFIG_INTR_REMAP
3466         .ack            = ir_ack_apic_edge,
3467 #ifdef CONFIG_SMP
3468         .set_affinity   = ir_set_msi_irq_affinity,
3469 #endif
3470 #endif
3471         .retrigger      = ioapic_retrigger_irq,
3472 };
3473 
3474 /*
3475  * Map the PCI dev to the corresponding remapping hardware unit
3476  * and allocate 'nvec' consecutive interrupt-remapping table entries
3477  * in it.
3478  */
3479 static int msi_alloc_irte(struct pci_dev *dev, int irq, int nvec)
3480 {
3481         struct intel_iommu *iommu;
3482         int index;
3483 
3484         iommu = map_dev_to_ir(dev);
3485         if (!iommu) {
3486                 printk(KERN_ERR
3487                        "Unable to map PCI %s to iommu\n", pci_name(dev));
3488                 return -ENOENT;
3489         }
3490 
3491         index = alloc_irte(iommu, irq, nvec);
3492         if (index < 0) {
3493                 printk(KERN_ERR
3494                        "Unable to allocate %d IRTE for PCI %s\n", nvec,
3495                        pci_name(dev));
3496                 return -ENOSPC;
3497         }
3498         return index;
3499 }
3500 
3501 static int setup_msi_irq(struct pci_dev *dev, struct msi_desc *msidesc, int irq)
3502 {
3503         int ret;
3504         struct msi_msg msg;
3505 
3506         ret = msi_compose_msg(dev, irq, &msg);
3507         if (ret < 0)
3508                 return ret;
3509 
3510         set_irq_msi(irq, msidesc);
3511         write_msi_msg(irq, &msg);
3512 
3513         if (irq_remapped(irq)) {
3514                 struct irq_desc *desc = irq_to_desc(irq);
3515                 /*
3516                  * irq migration in process context
3517                  */
3518                 desc->status |= IRQ_MOVE_PCNTXT;
3519                 set_irq_chip_and_handler_name(irq, &msi_ir_chip, handle_edge_irq, "edge");
3520         } else
3521                 set_irq_chip_and_handler_name(irq, &msi_chip, handle_edge_irq, "edge");
3522 
3523         dev_printk(KERN_DEBUG, &dev->dev, "irq %d for MSI/MSI-X\n", irq);
3524 
3525         return 0;
3526 }
3527 
3528 int arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
3529 {
3530         unsigned int irq;
3531         int ret, sub_handle;
3532         struct msi_desc *msidesc;
3533         unsigned int irq_want;
3534         struct intel_iommu *iommu = NULL;
3535         int index = 0;
3536         int node;
3537 
3538         /* x86 doesn't support multiple MSI yet */
3539         if (type == PCI_CAP_ID_MSI && nvec > 1)
3540                 return 1;
3541 
3542         node = dev_to_node(&dev->dev);
3543         irq_want = nr_irqs_gsi;
3544         sub_handle = 0;
3545         list_for_each_entry(msidesc, &dev->msi_list, list) {
3546                 irq = create_irq_nr(irq_want, node);
3547                 if (irq == 0)
3548                         return -1;
3549                 irq_want = irq + 1;
3550                 if (!intr_remapping_enabled)
3551                         goto no_ir;
3552 
3553                 if (!sub_handle) {
3554                         /*
3555                          * allocate the consecutive block of IRTE's
3556                          * for 'nvec'
3557                          */
3558                         index = msi_alloc_irte(dev, irq, nvec);
3559                         if (index < 0) {
3560                                 ret = index;
3561                                 goto error;
3562                         }
3563                 } else {
3564                         iommu = map_dev_to_ir(dev);
3565                         if (!iommu) {
3566                                 ret = -ENOENT;
3567                                 goto error;
3568                         }
3569                         /*
3570                          * setup the mapping between the irq and the IRTE
3571                          * base index, the sub_handle pointing to the
3572                          * appropriate interrupt remap table entry.
3573                          */
3574                         set_irte_irq(irq, iommu, index, sub_handle);
3575                 }
3576 no_ir:
3577                 ret = setup_msi_irq(dev, msidesc, irq);
3578                 if (ret < 0)
3579                         goto error;
3580                 sub_handle++;
3581         }
3582         return 0;
3583 
3584 error:
3585         destroy_irq(irq);
3586         return ret;
3587 }
3588 
3589 void arch_teardown_msi_irq(unsigned int irq)
3590 {
3591         destroy_irq(irq);
3592 }
3593 
3594 #if defined (CONFIG_DMAR) || defined (CONFIG_INTR_REMAP)
3595 #ifdef CONFIG_SMP
3596 static int dmar_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         dmar_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         dmar_msi_write(irq, &msg);
3617 
3618         return 0;
3619 }
3620 
3621 #endif /* CONFIG_SMP */
3622 
3623 static struct irq_chip dmar_msi_type = {
3624         .name = "DMAR_MSI",
3625         .unmask = dmar_msi_unmask,
3626         .mask = dmar_msi_mask,
3627         .ack = ack_apic_edge,
3628 #ifdef CONFIG_SMP
3629         .set_affinity = dmar_msi_set_affinity,
3630 #endif
3631         .retrigger = ioapic_retrigger_irq,
3632 };
3633 
3634 int arch_setup_dmar_msi(unsigned int irq)
3635 {
3636         int ret;
3637         struct msi_msg msg;
3638 
3639         ret = msi_compose_msg(NULL, irq, &msg);
3640         if (ret < 0)
3641                 return ret;
3642         dmar_msi_write(irq, &msg);
3643         set_irq_chip_and_handler_name(irq, &dmar_msi_type, handle_edge_irq,
3644                 "edge");
3645         return 0;
3646 }
3647 #endif
3648 
3649 #ifdef CONFIG_HPET_TIMER
3650 
3651 #ifdef CONFIG_SMP
3652 static int hpet_msi_set_affinity(unsigned int irq, const struct cpumask *mask)
3653 {
3654         struct irq_desc *desc = irq_to_desc(irq);
3655         struct irq_cfg *cfg;
3656         struct msi_msg msg;
3657         unsigned int dest;
3658 
3659         dest = set_desc_affinity(desc, mask);
3660         if (dest == BAD_APICID)
3661                 return -1;
3662 
3663         cfg = desc->chip_data;
3664 
3665         hpet_msi_read(irq, &msg);
3666 
3667         msg.data &= ~MSI_DATA_VECTOR_MASK;
3668         msg.data |= MSI_DATA_VECTOR(cfg->vector);
3669         msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3670         msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3671 
3672         hpet_msi_write(irq, &msg);
3673 
3674         return 0;
3675 }
3676 
3677 #endif /* CONFIG_SMP */
3678 
3679 static struct irq_chip hpet_msi_type = {
3680         .name = "HPET_MSI",
3681         .unmask = hpet_msi_unmask,
3682         .mask = hpet_msi_mask,
3683         .ack = ack_apic_edge,
3684 #ifdef CONFIG_SMP
3685         .set_affinity = hpet_msi_set_affinity,
3686 #endif
3687         .retrigger = ioapic_retrigger_irq,
3688 };
3689 
3690 int arch_setup_hpet_msi(unsigned int irq)
3691 {
3692         int ret;
3693         struct msi_msg msg;
3694         struct irq_desc *desc = irq_to_desc(irq);
3695 
3696         ret = msi_compose_msg(NULL, irq, &msg);
3697         if (ret < 0)
3698                 return ret;
3699 
3700         hpet_msi_write(irq, &msg);
3701         desc->status |= IRQ_MOVE_PCNTXT;
3702         set_irq_chip_and_handler_name(irq, &hpet_msi_type, handle_edge_irq,
3703                 "edge");
3704 
3705         return 0;
3706 }
3707 #endif
3708 
3709 #endif /* CONFIG_PCI_MSI */
3710 /*
3711  * Hypertransport interrupt support
3712  */
3713 #ifdef CONFIG_HT_IRQ
3714 
3715 #ifdef CONFIG_SMP
3716 
3717 static void target_ht_irq(unsigned int irq, unsigned int dest, u8 vector)
3718 {
3719         struct ht_irq_msg msg;
3720         fetch_ht_irq_msg(irq, &msg);
3721 
3722         msg.address_lo &= ~(HT_IRQ_LOW_VECTOR_MASK | HT_IRQ_LOW_DEST_ID_MASK);
3723         msg.address_hi &= ~(HT_IRQ_HIGH_DEST_ID_MASK);
3724 
3725         msg.address_lo |= HT_IRQ_LOW_VECTOR(vector) | HT_IRQ_LOW_DEST_ID(dest);
3726         msg.address_hi |= HT_IRQ_HIGH_DEST_ID(dest);
3727 
3728         write_ht_irq_msg(irq, &msg);
3729 }
3730 
3731 static int set_ht_irq_affinity(unsigned int irq, const struct cpumask *mask)
3732 {
3733         struct irq_desc *desc = irq_to_desc(irq);
3734         struct irq_cfg *cfg;
3735         unsigned int dest;
3736 
3737         dest = set_desc_affinity(desc, mask);
3738         if (dest == BAD_APICID)
3739                 return -1;
3740 
3741         cfg = desc->chip_data;
3742 
3743         target_ht_irq(irq, dest, cfg->vector);
3744 
3745         return 0;
3746 }
3747 
3748 #endif
3749 
3750 static struct irq_chip ht_irq_chip = {
3751         .name           = "PCI-HT",
3752         .mask           = mask_ht_irq,
3753         .unmask         = unmask_ht_irq,
3754         .ack            = ack_apic_edge,
3755 #ifdef CONFIG_SMP
3756         .set_affinity   = set_ht_irq_affinity,
3757 #endif
3758         .retrigger      = ioapic_retrigger_irq,
3759 };
3760 
3761 int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev)
3762 {
3763         struct irq_cfg *cfg;
3764         int err;
3765 
3766         if (disable_apic)
3767                 return -ENXIO;
3768 
3769         cfg = irq_cfg(irq);
3770         err = assign_irq_vector(irq, cfg, apic->target_cpus());
3771         if (!err) {
3772                 struct ht_irq_msg msg;
3773                 unsigned dest;
3774 
3775                 dest = apic->cpu_mask_to_apicid_and(cfg->domain,
3776                                                     apic->target_cpus());
3777 
3778                 msg.address_hi = HT_IRQ_HIGH_DEST_ID(dest);
3779 
3780                 msg.address_lo =
3781                         HT_IRQ_LOW_BASE |
3782                         HT_IRQ_LOW_DEST_ID(dest) |
3783                         HT_IRQ_LOW_VECTOR(cfg->vector) |
3784                         ((apic->irq_dest_mode == 0) ?
3785                                 HT_IRQ_LOW_DM_PHYSICAL :
3786                                 HT_IRQ_LOW_DM_LOGICAL) |
3787                         HT_IRQ_LOW_RQEOI_EDGE |
3788                         ((apic->irq_delivery_mode != dest_LowestPrio) ?
3789                                 HT_IRQ_LOW_MT_FIXED :
3790                                 HT_IRQ_LOW_MT_ARBITRATED) |
3791                         HT_IRQ_LOW_IRQ_MASKED;
3792 
3793                 write_ht_irq_msg(irq, &msg);
3794 
3795                 set_irq_chip_and_handler_name(irq, &ht_irq_chip,
3796                                               handle_edge_irq, "edge");
3797 
3798                 dev_printk(KERN_DEBUG, &dev->dev, "irq %d for HT\n", irq);
3799         }
3800         return err;
3801 }
3802 #endif /* CONFIG_HT_IRQ */
3803 
3804 #ifdef CONFIG_X86_UV
3805 /*
3806  * Re-target the irq to the specified CPU and enable the specified MMR located
3807  * on the specified blade to allow the sending of MSIs to the specified CPU.
3808  */
3809 int arch_enable_uv_irq(char *irq_name, unsigned int irq, int cpu, int mmr_blade,
3810                        unsigned long mmr_offset)
3811 {
3812         const struct cpumask *eligible_cpu = cpumask_of(cpu);
3813         struct irq_cfg *cfg;
3814         int mmr_pnode;
3815         unsigned long mmr_value;
3816         struct uv_IO_APIC_route_entry *entry;
3817         unsigned long flags;
3818         int err;
3819 
3820         BUILD_BUG_ON(sizeof(struct uv_IO_APIC_route_entry) != sizeof(unsigned long));
3821 
3822         cfg = irq_cfg(irq);
3823 
3824         err = assign_irq_vector(irq, cfg, eligible_cpu);
3825         if (err != 0)
3826                 return err;
3827 
3828         spin_lock_irqsave(&vector_lock, flags);
3829         set_irq_chip_and_handler_name(irq, &uv_irq_chip, handle_percpu_irq,
3830                                       irq_name);
3831         spin_unlock_irqrestore(&vector_lock, flags);
3832 
3833         mmr_value = 0;
3834         entry = (struct uv_IO_APIC_route_entry *)&mmr_value;
3835         entry->vector           = cfg->vector;
3836         entry->delivery_mode    = apic->irq_delivery_mode;
3837         entry->dest_mode        = apic->irq_dest_mode;
3838         entry->polarity         = 0;
3839         entry->trigger          = 0;
3840         entry->mask             = 0;
3841         entry->dest             = apic->cpu_mask_to_apicid(eligible_cpu);
3842 
3843         mmr_pnode = uv_blade_to_pnode(mmr_blade);
3844         uv_write_global_mmr64(mmr_pnode, mmr_offset, mmr_value);
3845 
3846         if (cfg->move_in_progress)
3847                 send_cleanup_vector(cfg);
3848 
3849         return irq;
3850 }
3851 
3852 /*
3853  * Disable the specified MMR located on the specified blade so that MSIs are
3854  * longer allowed to be sent.
3855  */
3856 void arch_disable_uv_irq(int mmr_blade, unsigned long mmr_offset)
3857 {
3858         unsigned long mmr_value;
3859         struct uv_IO_APIC_route_entry *entry;
3860         int mmr_pnode;
3861 
3862         BUILD_BUG_ON(sizeof(struct uv_IO_APIC_route_entry) != sizeof(unsigned long));
3863 
3864         mmr_value = 0;
3865         entry = (struct uv_IO_APIC_route_entry *)&mmr_value;
3866         entry->mask = 1;
3867 
3868         mmr_pnode = uv_blade_to_pnode(mmr_blade);
3869         uv_write_global_mmr64(mmr_pnode, mmr_offset, mmr_value);
3870 }
3871 #endif /* CONFIG_X86_64 */
3872 
3873 int __init io_apic_get_redir_entries (int ioapic)
3874 {
3875         union IO_APIC_reg_01    reg_01;
3876         unsigned long flags;
3877 
3878         spin_lock_irqsave(&ioapic_lock, flags);
3879         reg_01.raw = io_apic_read(ioapic, 1);
3880         spin_unlock_irqrestore(&ioapic_lock, flags);
3881 
3882         return reg_01.bits.entries;
3883 }
3884 
3885 void __init probe_nr_irqs_gsi(void)
3886 {
3887         int nr = 0;
3888 
3889         nr = acpi_probe_gsi();
3890         if (nr > nr_irqs_gsi) {
3891                 nr_irqs_gsi = nr;
3892         } else {
3893                 /* for acpi=off or acpi is not compiled in */
3894                 int idx;
3895 
3896                 nr = 0;
3897                 for (idx = 0; idx < nr_ioapics; idx++)
3898                         nr += io_apic_get_redir_entries(idx) + 1;
3899 
3900                 if (nr > nr_irqs_gsi)
3901                         nr_irqs_gsi = nr;
3902         }
3903 
3904         printk(KERN_DEBUG "nr_irqs_gsi: %d\n", nr_irqs_gsi);
3905 }
3906 
3907 #ifdef CONFIG_SPARSE_IRQ
3908 int __init arch_probe_nr_irqs(void)
3909 {
3910         int nr;
3911 
3912         if (nr_irqs > (NR_VECTORS * nr_cpu_ids))
3913                 nr_irqs = NR_VECTORS * nr_cpu_ids;
3914 
3915         nr = nr_irqs_gsi + 8 * nr_cpu_ids;
3916 #if defined(CONFIG_PCI_MSI) || defined(CONFIG_HT_IRQ)
3917         /*
3918          * for MSI and HT dyn irq
3919          */
3920         nr += nr_irqs_gsi * 16;
3921 #endif
3922         if (nr < nr_irqs)
3923                 nr_irqs = nr;
3924 
3925         return 0;
3926 }
3927 #endif
3928 
3929 static int __io_apic_set_pci_routing(struct device *dev, int irq,
3930                                 struct io_apic_irq_attr *irq_attr)
3931 {
3932         struct irq_desc *desc;
3933         struct irq_cfg *cfg;
3934         int node;
3935         int ioapic, pin;
3936         int trigger, polarity;
3937 
3938         ioapic = irq_attr->ioapic;
3939         if (!IO_APIC_IRQ(irq)) {
3940                 apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
3941                         ioapic);
3942                 return -EINVAL;
3943         }
3944 
3945         if (dev)
3946                 node = dev_to_node(dev);
3947         else
3948                 node = cpu_to_node(boot_cpu_id);
3949 
3950         desc = irq_to_desc_alloc_node(irq, node);
3951         if (!desc) {
3952                 printk(KERN_INFO "can not get irq_desc %d\n", irq);
3953                 return 0;
3954         }
3955 
3956         pin = irq_attr->ioapic_pin;
3957         trigger = irq_attr->trigger;
3958         polarity = irq_attr->polarity;
3959 
3960         /*
3961          * IRQs < 16 are already in the irq_2_pin[] map
3962          */
3963         if (irq >= NR_IRQS_LEGACY) {
3964                 cfg = desc->chip_data;
3965                 add_pin_to_irq_node(cfg, node, ioapic, pin);
3966         }
3967 
3968         setup_IO_APIC_irq(ioapic, pin, irq, desc, trigger, polarity);
3969 
3970         return 0;
3971 }
3972 
3973 int io_apic_set_pci_routing(struct device *dev, int irq,
3974                                 struct io_apic_irq_attr *irq_attr)
3975 {
3976         int ioapic, pin;
3977         /*
3978          * Avoid pin reprogramming.  PRTs typically include entries
3979          * with redundant pin->gsi mappings (but unique PCI devices);
3980          * we only program the IOAPIC on the first.
3981          */
3982         ioapic = irq_attr->ioapic;
3983         pin = irq_attr->ioapic_pin;
3984         if (test_bit(pin, mp_ioapic_routing[ioapic].pin_programmed)) {
3985                 pr_debug("Pin %d-%d already programmed\n",
3986                          mp_ioapics[ioapic].apicid, pin);
3987                 return 0;
3988         }
3989         set_bit(pin, mp_ioapic_routing[ioapic].pin_programmed);
3990 
3991         return __io_apic_set_pci_routing(dev, irq, irq_attr);
3992 }
3993 
3994 /* --------------------------------------------------------------------------
3995                           ACPI-based IOAPIC Configuration
3996    -------------------------------------------------------------------------- */
3997 
3998 #ifdef CONFIG_ACPI
3999 
4000 #ifdef CONFIG_X86_32
4001 int __init io_apic_get_unique_id(int ioapic, int apic_id)
4002 {
4003         union IO_APIC_reg_00 reg_00;
4004         static physid_mask_t apic_id_map = PHYSID_MASK_NONE;
4005         physid_mask_t tmp;
4006         unsigned long flags;
4007         int i = 0;
4008 
4009         /*
4010          * The P4 platform supports up to 256 APIC IDs on two separate APIC
4011          * buses (one for LAPICs, one for IOAPICs), where predecessors only
4012          * supports up to 16 on one shared APIC bus.
4013          *
4014          * TBD: Expand LAPIC/IOAPIC support on P4-class systems to take full
4015          *      advantage of new APIC bus architecture.
4016          */
4017 
4018         if (physids_empty(apic_id_map))
4019                 apic_id_map = apic->ioapic_phys_id_map(phys_cpu_present_map);
4020 
4021         spin_lock_irqsave(&ioapic_lock, flags);
4022         reg_00.raw = io_apic_read(ioapic, 0);
4023         spin_unlock_irqrestore(&ioapic_lock, flags);
4024 
4025         if (apic_id >= get_physical_broadcast()) {
4026                 printk(KERN_WARNING "IOAPIC[%d]: Invalid apic_id %d, trying "
4027                         "%d\n", ioapic, apic_id, reg_00.bits.ID);
4028                 apic_id = reg_00.bits.ID;
4029         }
4030 
4031         /*
4032          * Every APIC in a system must have a unique ID or we get lots of nice
4033          * 'stuck on smp_invalidate_needed IPI wait' messages.
4034          */
4035         if (apic->check_apicid_used(apic_id_map, apic_id)) {
4036 
4037                 for (i = 0; i < get_physical_broadcast(); i++) {
4038                         if (!apic->check_apicid_used(apic_id_map, i))
4039                                 break;
4040                 }
4041 
4042                 if (i == get_physical_broadcast())
4043                         panic("Max apic_id exceeded!\n");
4044 
4045                 printk(KERN_WARNING "IOAPIC[%d]: apic_id %d already used, "
4046                         "trying %d\n", ioapic, apic_id, i);
4047 
4048                 apic_id = i;
4049         }
4050 
4051         tmp = apic->apicid_to_cpu_present(apic_id);
4052         physids_or(apic_id_map, apic_id_map, tmp);
4053 
4054         if (reg_00.bits.ID != apic_id) {
4055                 reg_00.bits.ID = apic_id;
4056 
4057                 spin_lock_irqsave(&ioapic_lock, flags);
4058                 io_apic_write(ioapic, 0, reg_00.raw);
4059                 reg_00.raw = io_apic_read(ioapic, 0);
4060                 spin_unlock_irqrestore(&ioapic_lock, flags);
4061 
4062                 /* Sanity check */
4063                 if (reg_00.bits.ID != apic_id) {
4064                         printk("IOAPIC[%d]: Unable to change apic_id!\n", ioapic);
4065                         return -1;
4066                 }
4067         }
4068 
4069         apic_printk(APIC_VERBOSE, KERN_INFO
4070                         "IOAPIC[%d]: Assigned apic_id %d\n", ioapic, apic_id);
4071 
4072         return apic_id;
4073 }
4074 #endif
4075 
4076 int __init io_apic_get_version(int ioapic)
4077 {
4078         union IO_APIC_reg_01    reg_01;
4079         unsigned long flags;
4080 
4081         spin_lock_irqsave(&ioapic_lock, flags);
4082         reg_01.raw = io_apic_read(ioapic, 1);
4083         spin_unlock_irqrestore(&ioapic_lock, flags);
4084 
4085         return reg_01.bits.version;
4086 }
4087 
4088 int acpi_get_override_irq(int bus_irq, int *trigger, int *polarity)
4089 {
4090         int i;
4091 
4092         if (skip_ioapic_setup)
4093                 return -1;
4094 
4095         for (i = 0; i < mp_irq_entries; i++)
4096                 if (mp_irqs[i].irqtype == mp_INT &&
4097                     mp_irqs[i].srcbusirq == bus_irq)
4098                         break;
4099         if (i >= mp_irq_entries)
4100                 return -1;
4101 
4102         *trigger = irq_trigger(i);
4103         *polarity = irq_polarity(i);
4104         return 0;
4105 }
4106 
4107 #endif /* CONFIG_ACPI */
4108 
4109 /*
4110  * This function currently is only a helper for the i386 smp boot process where
4111  * we need to reprogram the ioredtbls to cater for the cpus which have come online
4112  * so mask in all cases should simply be apic->target_cpus()
4113  */
4114 #ifdef CONFIG_SMP
4115 void __init setup_ioapic_dest(void)
4116 {
4117         int pin, ioapic = 0, irq, irq_entry;
4118         struct irq_desc *desc;
4119         const struct cpumask *mask;
4120 
4121         if (skip_ioapic_setup == 1)
4122                 return;
4123 
4124 #ifdef CONFIG_ACPI
4125         if (!acpi_disabled && acpi_ioapic) {
4126                 ioapic = mp_find_ioapic(0);
4127                 if (ioapic < 0)
4128                         ioapic = 0;
4129         }
4130 #endif
4131 
4132         for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
4133                 irq_entry = find_irq_entry(ioapic, pin, mp_INT);
4134                 if (irq_entry == -1)
4135                         continue;
4136                 irq = pin_2_irq(irq_entry, ioapic, pin);
4137 
4138                 desc = irq_to_desc(irq);
4139 
4140                 /*
4141                  * Honour affinities which have been set in early boot
4142                  */
4143                 if (desc->status &
4144                     (IRQ_NO_BALANCING | IRQ_AFFINITY_SET))
4145                         mask = desc->affinity;
4146                 else
4147                         mask = apic->target_cpus();
4148 
4149                 if (intr_remapping_enabled)
4150                         set_ir_ioapic_affinity_irq_desc(desc, mask);
4151                 else
4152                         set_ioapic_affinity_irq_desc(desc, mask);
4153         }
4154 
4155 }
4156 #endif
4157 
4158 #define IOAPIC_RESOURCE_NAME_SIZE 11
4159 
4160 static struct resource *ioapic_resources;
4161 
4162 static struct resource * __init ioapic_setup_resources(void)
4163 {
4164         unsigned long n;
4165         struct resource *res;
4166         char *mem;
4167         int i;
4168 
4169         if (nr_ioapics <= 0)
4170                 return NULL;
4171 
4172         n = IOAPIC_RESOURCE_NAME_SIZE + sizeof(struct resource);
4173         n *= nr_ioapics;
4174 
4175         mem = alloc_bootmem(n);
4176         res = (void *)mem;
4177 
4178         if (mem != NULL) {
4179                 mem += sizeof(struct resource) * nr_ioapics;
4180 
4181                 for (i = 0; i < nr_ioapics; i++) {
4182                         res[i].name = mem;
4183                         res[i].flags = IORESOURCE_MEM | IORESOURCE_BUSY;
4184                         sprintf(mem,  "IOAPIC %u", i);
4185                         mem += IOAPIC_RESOURCE_NAME_SIZE;
4186                 }
4187         }
4188 
4189         ioapic_resources = res;
4190 
4191         return res;
4192 }
4193 
4194 void __init ioapic_init_mappings(void)
4195 {
4196         unsigned long ioapic_phys, idx = FIX_IO_APIC_BASE_0;
4197         struct resource *ioapic_res;
4198         int i;
4199 
4200         ioapic_res = ioapic_setup_resources();
4201         for (i = 0; i < nr_ioapics; i++) {
4202                 if (smp_found_config) {
4203                         ioapic_phys = mp_ioapics[i].apicaddr;
4204 #ifdef CONFIG_X86_32
4205                         if (!ioapic_phys) {
4206                                 printk(KERN_ERR
4207                                        "WARNING: bogus zero IO-APIC "
4208                                        "address found in MPTABLE, "
4209                                        "disabling IO/APIC support!\n");
4210                                 smp_found_config = 0;
4211                                 skip_ioapic_setup = 1;
4212                                 goto fake_ioapic_page;
4213                         }
4214 #endif
4215                 } else {
4216 #ifdef CONFIG_X86_32
4217 fake_ioapic_page:
4218 #endif
4219                         ioapic_phys = (unsigned long)
4220                                 alloc_bootmem_pages(PAGE_SIZE);
4221                         ioapic_phys = __pa(ioapic_phys);
4222                 }
4223                 set_fixmap_nocache(idx, ioapic_phys);
4224                 apic_printk(APIC_VERBOSE,
4225                             "mapped IOAPIC to %08lx (%08lx)\n",
4226                             __fix_to_virt(idx), ioapic_phys);
4227                 idx++;
4228 
4229                 if (ioapic_res != NULL) {
4230                         ioapic_res->start = ioapic_phys;
4231                         ioapic_res->end = ioapic_phys + (4 * 1024) - 1;
4232                         ioapic_res++;
4233                 }
4234         }
4235 }
4236 
4237 void __init ioapic_insert_resources(void)
4238 {
4239         int i;
4240         struct resource *r = ioapic_resources;
4241 
4242         if (!r) {
4243                 if (nr_ioapics > 0)
4244                         printk(KERN_ERR
4245                                 "IO APIC resources couldn't be allocated.\n");
4246                 return;
4247         }
4248 
4249         for (i = 0; i < nr_ioapics; i++) {
4250                 insert_resource(&iomem_resource, r);
4251                 r++;
4252         }
4253 }
4254 
  This page was automatically generated by the LXR engine.