projects / deliverable / linux.git / blob
? search:


ae80638012dec8c93bf33c908b90dcbbbe8f509c
[deliverable/linux.git] / arch / x86 / kernel / io_apic.c
1 /*
2 * Intel IO-APIC support for multi-Pentium hosts.
3 *
4 * Copyright (C) 1997, 1998, 1999, 2000 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/desc.h>
50 #include <asm/proto.h>
51 #include <asm/acpi.h>
52 #include <asm/dma.h>
53 #include <asm/timer.h>
54 #include <asm/i8259.h>
55 #include <asm/nmi.h>
56 #include <asm/msidef.h>
57 #include <asm/hypertransport.h>
58 #include <asm/setup.h>
59 #include <asm/irq_remapping.h>
60 #include <asm/hpet.h>
61 #include <asm/uv/uv_hub.h>
62 #include <asm/uv/uv_irq.h>
63
64 #include <mach_ipi.h>
65 #include <mach_apic.h>
66 #include <mach_apicdef.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 mp_config_ioapic mp_ioapics[MAX_IO_APICS];
86 int nr_ioapics;
87
88 /* MP IRQ source entries */
89 struct mp_config_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 static int __init parse_noapic(char *str)
103 {
104 /* disable IO-APIC */
105 disable_ioapic_setup();
106 return 0;
107 }
108 early_param("noapic", parse_noapic);
109
110 struct irq_pin_list;
111
112 /*
113 * This is performance-critical, we want to do it O(1)
114 *
115 * the indexing order of this array favors 1:1 mappings
116 * between pins and IRQs.
117 */
118
119 struct irq_pin_list {
120 int apic, pin;
121 struct irq_pin_list *next;
122 };
123
124 static struct irq_pin_list *get_one_free_irq_2_pin(int cpu)
125 {
126 struct irq_pin_list *pin;
127 int node;
128
129 node = cpu_to_node(cpu);
130
131 pin = kzalloc_node(sizeof(*pin), GFP_ATOMIC, node);
132
133 return pin;
134 }
135
136 struct irq_cfg {
137 struct irq_pin_list *irq_2_pin;
138 cpumask_var_t domain;
139 cpumask_var_t old_domain;
140 unsigned move_cleanup_count;
141 u8 vector;
142 u8 move_in_progress : 1;
143 #ifdef CONFIG_NUMA_MIGRATE_IRQ_DESC
144 u8 move_desc_pending : 1;
145 #endif
146 };
147
148 /* irq_cfg is indexed by the sum of all RTEs in all I/O APICs. */
149 #ifdef CONFIG_SPARSE_IRQ
150 static struct irq_cfg irq_cfgx[] = {
151 #else
152 static struct irq_cfg irq_cfgx[NR_IRQS] = {
153 #endif
154 [0] = { .vector = IRQ0_VECTOR, },
155 [1] = { .vector = IRQ1_VECTOR, },
156 [2] = { .vector = IRQ2_VECTOR, },
157 [3] = { .vector = IRQ3_VECTOR, },
158 [4] = { .vector = IRQ4_VECTOR, },
159 [5] = { .vector = IRQ5_VECTOR, },
160 [6] = { .vector = IRQ6_VECTOR, },
161 [7] = { .vector = IRQ7_VECTOR, },
162 [8] = { .vector = IRQ8_VECTOR, },
163 [9] = { .vector = IRQ9_VECTOR, },
164 [10] = { .vector = IRQ10_VECTOR, },
165 [11] = { .vector = IRQ11_VECTOR, },
166 [12] = { .vector = IRQ12_VECTOR, },
167 [13] = { .vector = IRQ13_VECTOR, },
168 [14] = { .vector = IRQ14_VECTOR, },
169 [15] = { .vector = IRQ15_VECTOR, },
170 };
171
172 int __init arch_early_irq_init(void)
173 {
174 struct irq_cfg *cfg;
175 struct irq_desc *desc;
176 int count;
177 int i;
178
179 cfg = irq_cfgx;
180 count = ARRAY_SIZE(irq_cfgx);
181
182 for (i = 0; i < count; i++) {
183 desc = irq_to_desc(i);
184 desc->chip_data = &cfg[i];
185 alloc_bootmem_cpumask_var(&cfg[i].domain);
186 alloc_bootmem_cpumask_var(&cfg[i].old_domain);
187 if (i < NR_IRQS_LEGACY)
188 cpumask_setall(cfg[i].domain);
189 }
190
191 return 0;
192 }
193
194 #ifdef CONFIG_SPARSE_IRQ
195 static struct irq_cfg *irq_cfg(unsigned int irq)
196 {
197 struct irq_cfg *cfg = NULL;
198 struct irq_desc *desc;
199
200 desc = irq_to_desc(irq);
201 if (desc)
202 cfg = desc->chip_data;
203
204 return cfg;
205 }
206
207 static struct irq_cfg *get_one_free_irq_cfg(int cpu)
208 {
209 struct irq_cfg *cfg;
210 int node;
211
212 node = cpu_to_node(cpu);
213
214 cfg = kzalloc_node(sizeof(*cfg), GFP_ATOMIC, node);
215 if (cfg) {
216 if (!alloc_cpumask_var_node(&cfg->domain, GFP_ATOMIC, node)) {
217 kfree(cfg);
218 cfg = NULL;
219 } else if (!alloc_cpumask_var_node(&cfg->old_domain,
220 GFP_ATOMIC, node)) {
221 free_cpumask_var(cfg->domain);
222 kfree(cfg);
223 cfg = NULL;
224 } else {
225 cpumask_clear(cfg->domain);
226 cpumask_clear(cfg->old_domain);
227 }
228 }
229
230 return cfg;
231 }
232
233 int arch_init_chip_data(struct irq_desc *desc, int cpu)
234 {
235 struct irq_cfg *cfg;
236
237 cfg = desc->chip_data;
238 if (!cfg) {
239 desc->chip_data = get_one_free_irq_cfg(cpu);
240 if (!desc->chip_data) {
241 printk(KERN_ERR "can not alloc irq_cfg\n");
242 BUG_ON(1);
243 }
244 }
245
246 return 0;
247 }
248
249 #ifdef CONFIG_NUMA_MIGRATE_IRQ_DESC
250
251 static void
252 init_copy_irq_2_pin(struct irq_cfg *old_cfg, struct irq_cfg *cfg, int cpu)
253 {
254 struct irq_pin_list *old_entry, *head, *tail, *entry;
255
256 cfg->irq_2_pin = NULL;
257 old_entry = old_cfg->irq_2_pin;
258 if (!old_entry)
259 return;
260
261 entry = get_one_free_irq_2_pin(cpu);
262 if (!entry)
263 return;
264
265 entry->apic = old_entry->apic;
266 entry->pin = old_entry->pin;
267 head = entry;
268 tail = entry;
269 old_entry = old_entry->next;
270 while (old_entry) {
271 entry = get_one_free_irq_2_pin(cpu);
272 if (!entry) {
273 entry = head;
274 while (entry) {
275 head = entry->next;
276 kfree(entry);
277 entry = head;
278 }
279 /* still use the old one */
280 return;
281 }
282 entry->apic = old_entry->apic;
283 entry->pin = old_entry->pin;
284 tail->next = entry;
285 tail = entry;
286 old_entry = old_entry->next;
287 }
288
289 tail->next = NULL;
290 cfg->irq_2_pin = head;
291 }
292
293 static void free_irq_2_pin(struct irq_cfg *old_cfg, struct irq_cfg *cfg)
294 {
295 struct irq_pin_list *entry, *next;
296
297 if (old_cfg->irq_2_pin == cfg->irq_2_pin)
298 return;
299
300 entry = old_cfg->irq_2_pin;
301
302 while (entry) {
303 next = entry->next;
304 kfree(entry);
305 entry = next;
306 }
307 old_cfg->irq_2_pin = NULL;
308 }
309
310 void arch_init_copy_chip_data(struct irq_desc *old_desc,
311 struct irq_desc *desc, int cpu)
312 {
313 struct irq_cfg *cfg;
314 struct irq_cfg *old_cfg;
315
316 cfg = get_one_free_irq_cfg(cpu);
317
318 if (!cfg)
319 return;
320
321 desc->chip_data = cfg;
322
323 old_cfg = old_desc->chip_data;
324
325 memcpy(cfg, old_cfg, sizeof(struct irq_cfg));
326
327 init_copy_irq_2_pin(old_cfg, cfg, cpu);
328 }
329
330 static void free_irq_cfg(struct irq_cfg *old_cfg)
331 {
332 kfree(old_cfg);
333 }
334
335 void arch_free_chip_data(struct irq_desc *old_desc, struct irq_desc *desc)
336 {
337 struct irq_cfg *old_cfg, *cfg;
338
339 old_cfg = old_desc->chip_data;
340 cfg = desc->chip_data;
341
342 if (old_cfg == cfg)
343 return;
344
345 if (old_cfg) {
346 free_irq_2_pin(old_cfg, cfg);
347 free_irq_cfg(old_cfg);
348 old_desc->chip_data = NULL;
349 }
350 }
351
352 static void
353 set_extra_move_desc(struct irq_desc *desc, const struct cpumask *mask)
354 {
355 struct irq_cfg *cfg = desc->chip_data;
356
357 if (!cfg->move_in_progress) {
358 /* it means that domain is not changed */
359 if (!cpumask_intersects(desc->affinity, mask))
360 cfg->move_desc_pending = 1;
361 }
362 }
363 #endif
364
365 #else
366 static struct irq_cfg *irq_cfg(unsigned int irq)
367 {
368 return irq < nr_irqs ? irq_cfgx + irq : NULL;
369 }
370
371 #endif
372
373 #ifndef CONFIG_NUMA_MIGRATE_IRQ_DESC
374 static inline void
375 set_extra_move_desc(struct irq_desc *desc, const struct cpumask *mask)
376 {
377 }
378 #endif
379
380 struct io_apic {
381 unsigned int index;
382 unsigned int unused[3];
383 unsigned int data;
384 };
385
386 static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
387 {
388 return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
389 + (mp_ioapics[idx].mp_apicaddr & ~PAGE_MASK);
390 }
391
392 static inline unsigned int io_apic_read(unsigned int apic, unsigned int reg)
393 {
394 struct io_apic __iomem *io_apic = io_apic_base(apic);
395 writel(reg, &io_apic->index);
396 return readl(&io_apic->data);
397 }
398
399 static inline void io_apic_write(unsigned int apic, unsigned int reg, unsigned int value)
400 {
401 struct io_apic __iomem *io_apic = io_apic_base(apic);
402 writel(reg, &io_apic->index);
403 writel(value, &io_apic->data);
404 }
405
406 /*
407 * Re-write a value: to be used for read-modify-write
408 * cycles where the read already set up the index register.
409 *
410 * Older SiS APIC requires we rewrite the index register
411 */
412 static inline void io_apic_modify(unsigned int apic, unsigned int reg, unsigned int value)
413 {
414 struct io_apic __iomem *io_apic = io_apic_base(apic);
415
416 if (sis_apic_bug)
417 writel(reg, &io_apic->index);
418 writel(value, &io_apic->data);
419 }
420
421 static bool io_apic_level_ack_pending(struct irq_cfg *cfg)
422 {
423 struct irq_pin_list *entry;
424 unsigned long flags;
425
426 spin_lock_irqsave(&ioapic_lock, flags);
427 entry = cfg->irq_2_pin;
428 for (;;) {
429 unsigned int reg;
430 int pin;
431
432 if (!entry)
433 break;
434 pin = entry->pin;
435 reg = io_apic_read(entry->apic, 0x10 + pin*2);
436 /* Is the remote IRR bit set? */
437 if (reg & IO_APIC_REDIR_REMOTE_IRR) {
438 spin_unlock_irqrestore(&ioapic_lock, flags);
439 return true;
440 }
441 if (!entry->next)
442 break;
443 entry = entry->next;
444 }
445 spin_unlock_irqrestore(&ioapic_lock, flags);
446
447 return false;
448 }
449
450 union entry_union {
451 struct { u32 w1, w2; };
452 struct IO_APIC_route_entry entry;
453 };
454
455 static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
456 {
457 union entry_union eu;
458 unsigned long flags;
459 spin_lock_irqsave(&ioapic_lock, flags);
460 eu.w1 = io_apic_read(apic, 0x10 + 2 * pin);
461 eu.w2 = io_apic_read(apic, 0x11 + 2 * pin);
462 spin_unlock_irqrestore(&ioapic_lock, flags);
463 return eu.entry;
464 }
465
466 /*
467 * When we write a new IO APIC routing entry, we need to write the high
468 * word first! If the mask bit in the low word is clear, we will enable
469 * the interrupt, and we need to make sure the entry is fully populated
470 * before that happens.
471 */
472 static void
473 __ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
474 {
475 union entry_union eu;
476 eu.entry = e;
477 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
478 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
479 }
480
481 static void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
482 {
483 unsigned long flags;
484 spin_lock_irqsave(&ioapic_lock, flags);
485 __ioapic_write_entry(apic, pin, e);
486 spin_unlock_irqrestore(&ioapic_lock, flags);
487 }
488
489 /*
490 * When we mask an IO APIC routing entry, we need to write the low
491 * word first, in order to set the mask bit before we change the
492 * high bits!
493 */
494 static void ioapic_mask_entry(int apic, int pin)
495 {
496 unsigned long flags;
497 union entry_union eu = { .entry.mask = 1 };
498
499 spin_lock_irqsave(&ioapic_lock, flags);
500 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
501 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
502 spin_unlock_irqrestore(&ioapic_lock, flags);
503 }
504
505 #ifdef CONFIG_SMP
506 static void send_cleanup_vector(struct irq_cfg *cfg)
507 {
508 cpumask_var_t cleanup_mask;
509
510 if (unlikely(!alloc_cpumask_var(&cleanup_mask, GFP_ATOMIC))) {
511 unsigned int i;
512 cfg->move_cleanup_count = 0;
513 for_each_cpu_and(i, cfg->old_domain, cpu_online_mask)
514 cfg->move_cleanup_count++;
515 for_each_cpu_and(i, cfg->old_domain, cpu_online_mask)
516 send_IPI_mask(cpumask_of(i), IRQ_MOVE_CLEANUP_VECTOR);
517 } else {
518 cpumask_and(cleanup_mask, cfg->old_domain, cpu_online_mask);
519 cfg->move_cleanup_count = cpumask_weight(cleanup_mask);
520 send_IPI_mask(cleanup_mask, IRQ_MOVE_CLEANUP_VECTOR);
521 free_cpumask_var(cleanup_mask);
522 }
523 cfg->move_in_progress = 0;
524 }
525
526 static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, struct irq_cfg *cfg)
527 {
528 int apic, pin;
529 struct irq_pin_list *entry;
530 u8 vector = cfg->vector;
531
532 entry = cfg->irq_2_pin;
533 for (;;) {
534 unsigned int reg;
535
536 if (!entry)
537 break;
538
539 apic = entry->apic;
540 pin = entry->pin;
541 #ifdef CONFIG_INTR_REMAP
542 /*
543 * With interrupt-remapping, destination information comes
544 * from interrupt-remapping table entry.
545 */
546 if (!irq_remapped(irq))
547 io_apic_write(apic, 0x11 + pin*2, dest);
548 #else
549 io_apic_write(apic, 0x11 + pin*2, dest);
550 #endif
551 reg = io_apic_read(apic, 0x10 + pin*2);
552 reg &= ~IO_APIC_REDIR_VECTOR_MASK;
553 reg |= vector;
554 io_apic_modify(apic, 0x10 + pin*2, reg);
555 if (!entry->next)
556 break;
557 entry = entry->next;
558 }
559 }
560
561 static int
562 assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask);
563
564 /*
565 * Either sets desc->affinity to a valid value, and returns cpu_mask_to_apicid
566 * of that, or returns BAD_APICID and leaves desc->affinity untouched.
567 */
568 static unsigned int
569 set_desc_affinity(struct irq_desc *desc, const struct cpumask *mask)
570 {
571 struct irq_cfg *cfg;
572 unsigned int irq;
573
574 if (!cpumask_intersects(mask, cpu_online_mask))
575 return BAD_APICID;
576
577 irq = desc->irq;
578 cfg = desc->chip_data;
579 if (assign_irq_vector(irq, cfg, mask))
580 return BAD_APICID;
581
582 cpumask_and(desc->affinity, cfg->domain, mask);
583 set_extra_move_desc(desc, mask);
584 return cpu_mask_to_apicid_and(desc->affinity, cpu_online_mask);
585 }
586
587 static void
588 set_ioapic_affinity_irq_desc(struct irq_desc *desc, const struct cpumask *mask)
589 {
590 struct irq_cfg *cfg;
591 unsigned long flags;
592 unsigned int dest;
593 unsigned int irq;
594
595 irq = desc->irq;
596 cfg = desc->chip_data;
597
598 spin_lock_irqsave(&ioapic_lock, flags);
599 dest = set_desc_affinity(desc, mask);
600 if (dest != BAD_APICID) {
601 /* Only the high 8 bits are valid. */
602 dest = SET_APIC_LOGICAL_ID(dest);
603 __target_IO_APIC_irq(irq, dest, cfg);
604 }
605 spin_unlock_irqrestore(&ioapic_lock, flags);
606 }
607
608 static void
609 set_ioapic_affinity_irq(unsigned int irq, const struct cpumask *mask)
610 {
611 struct irq_desc *desc;
612
613 desc = irq_to_desc(irq);
614
615 set_ioapic_affinity_irq_desc(desc, mask);
616 }
617 #endif /* CONFIG_SMP */
618
619 /*
620 * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
621 * shared ISA-space IRQs, so we have to support them. We are super
622 * fast in the common case, and fast for shared ISA-space IRQs.
623 */
624 static void add_pin_to_irq_cpu(struct irq_cfg *cfg, int cpu, int apic, int pin)
625 {
626 struct irq_pin_list *entry;
627
628 entry = cfg->irq_2_pin;
629 if (!entry) {
630 entry = get_one_free_irq_2_pin(cpu);
631 if (!entry) {
632 printk(KERN_ERR "can not alloc irq_2_pin to add %d - %d\n",
633 apic, pin);
634 return;
635 }
636 cfg->irq_2_pin = entry;
637 entry->apic = apic;
638 entry->pin = pin;
639 return;
640 }
641
642 while (entry->next) {
643 /* not again, please */
644 if (entry->apic == apic && entry->pin == pin)
645 return;
646
647 entry = entry->next;
648 }
649
650 entry->next = get_one_free_irq_2_pin(cpu);
651 entry = entry->next;
652 entry->apic = apic;
653 entry->pin = pin;
654 }
655
656 /*
657 * Reroute an IRQ to a different pin.
658 */
659 static void __init replace_pin_at_irq_cpu(struct irq_cfg *cfg, int cpu,
660 int oldapic, int oldpin,
661 int newapic, int newpin)
662 {
663 struct irq_pin_list *entry = cfg->irq_2_pin;
664 int replaced = 0;
665
666 while (entry) {
667 if (entry->apic == oldapic && entry->pin == oldpin) {
668 entry->apic = newapic;
669 entry->pin = newpin;
670 replaced = 1;
671 /* every one is different, right? */
672 break;
673 }
674 entry = entry->next;
675 }
676
677 /* why? call replace before add? */
678 if (!replaced)
679 add_pin_to_irq_cpu(cfg, cpu, newapic, newpin);
680 }
681
682 static inline void io_apic_modify_irq(struct irq_cfg *cfg,
683 int mask_and, int mask_or,
684 void (*final)(struct irq_pin_list *entry))
685 {
686 int pin;
687 struct irq_pin_list *entry;
688
689 for (entry = cfg->irq_2_pin; entry != NULL; entry = entry->next) {
690 unsigned int reg;
691 pin = entry->pin;
692 reg = io_apic_read(entry->apic, 0x10 + pin * 2);
693 reg &= mask_and;
694 reg |= mask_or;
695 io_apic_modify(entry->apic, 0x10 + pin * 2, reg);
696 if (final)
697 final(entry);
698 }
699 }
700
701 static void __unmask_IO_APIC_irq(struct irq_cfg *cfg)
702 {
703 io_apic_modify_irq(cfg, ~IO_APIC_REDIR_MASKED, 0, NULL);
704 }
705
706 #ifdef CONFIG_X86_64
707 static void io_apic_sync(struct irq_pin_list *entry)
708 {
709 /*
710 * Synchronize the IO-APIC and the CPU by doing
711 * a dummy read from the IO-APIC
712 */
713 struct io_apic __iomem *io_apic;
714 io_apic = io_apic_base(entry->apic);
715 readl(&io_apic->data);
716 }
717
718 static void __mask_IO_APIC_irq(struct irq_cfg *cfg)
719 {
720 io_apic_modify_irq(cfg, ~0, IO_APIC_REDIR_MASKED, &io_apic_sync);
721 }
722 #else /* CONFIG_X86_32 */
723 static void __mask_IO_APIC_irq(struct irq_cfg *cfg)
724 {
725 io_apic_modify_irq(cfg, ~0, IO_APIC_REDIR_MASKED, NULL);
726 }
727
728 static void __mask_and_edge_IO_APIC_irq(struct irq_cfg *cfg)
729 {
730 io_apic_modify_irq(cfg, ~IO_APIC_REDIR_LEVEL_TRIGGER,
731 IO_APIC_REDIR_MASKED, NULL);
732 }
733
734 static void __unmask_and_level_IO_APIC_irq(struct irq_cfg *cfg)
735 {
736 io_apic_modify_irq(cfg, ~IO_APIC_REDIR_MASKED,
737 IO_APIC_REDIR_LEVEL_TRIGGER, NULL);
738 }
739 #endif /* CONFIG_X86_32 */
740
741 static void mask_IO_APIC_irq_desc(struct irq_desc *desc)
742 {
743 struct irq_cfg *cfg = desc->chip_data;
744 unsigned long flags;
745
746 BUG_ON(!cfg);
747
748 spin_lock_irqsave(&ioapic_lock, flags);
749 __mask_IO_APIC_irq(cfg);
750 spin_unlock_irqrestore(&ioapic_lock, flags);
751 }
752
753 static void unmask_IO_APIC_irq_desc(struct irq_desc *desc)
754 {
755 struct irq_cfg *cfg = desc->chip_data;
756 unsigned long flags;
757
758 spin_lock_irqsave(&ioapic_lock, flags);
759 __unmask_IO_APIC_irq(cfg);
760 spin_unlock_irqrestore(&ioapic_lock, flags);
761 }
762
763 static void mask_IO_APIC_irq(unsigned int irq)
764 {
765 struct irq_desc *desc = irq_to_desc(irq);
766
767 mask_IO_APIC_irq_desc(desc);
768 }
769 static void unmask_IO_APIC_irq(unsigned int irq)
770 {
771 struct irq_desc *desc = irq_to_desc(irq);
772
773 unmask_IO_APIC_irq_desc(desc);
774 }
775
776 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
777 {
778 struct IO_APIC_route_entry entry;
779
780 /* Check delivery_mode to be sure we're not clearing an SMI pin */
781 entry = ioapic_read_entry(apic, pin);
782 if (entry.delivery_mode == dest_SMI)
783 return;
784 /*
785 * Disable it in the IO-APIC irq-routing table:
786 */
787 ioapic_mask_entry(apic, pin);
788 }
789
790 static void clear_IO_APIC (void)
791 {
792 int apic, pin;
793
794 for (apic = 0; apic < nr_ioapics; apic++)
795 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
796 clear_IO_APIC_pin(apic, pin);
797 }
798
799 #if !defined(CONFIG_SMP) && defined(CONFIG_X86_32)
800 void send_IPI_self(int vector)
801 {
802 unsigned int cfg;
803
804 /*
805 * Wait for idle.
806 */
807 apic_wait_icr_idle();
808 cfg = APIC_DM_FIXED | APIC_DEST_SELF | vector | APIC_DEST_LOGICAL;
809 /*
810 * Send the IPI. The write to APIC_ICR fires this off.
811 */
812 apic_write(APIC_ICR, cfg);
813 }
814 #endif /* !CONFIG_SMP && CONFIG_X86_32*/
815
816 #ifdef CONFIG_X86_32
817 /*
818 * support for broken MP BIOSs, enables hand-redirection of PIRQ0-7 to
819 * specific CPU-side IRQs.
820 */
821
822 #define MAX_PIRQS 8
823 static int pirq_entries [MAX_PIRQS];
824 static int pirqs_enabled;
825
826 static int __init ioapic_pirq_setup(char *str)
827 {
828 int i, max;
829 int ints[MAX_PIRQS+1];
830
831 get_options(str, ARRAY_SIZE(ints), ints);
832
833 for (i = 0; i < MAX_PIRQS; i++)
834 pirq_entries[i] = -1;
835
836 pirqs_enabled = 1;
837 apic_printk(APIC_VERBOSE, KERN_INFO
838 "PIRQ redirection, working around broken MP-BIOS.\n");
839 max = MAX_PIRQS;
840 if (ints[0] < MAX_PIRQS)
841 max = ints[0];
842
843 for (i = 0; i < max; i++) {
844 apic_printk(APIC_VERBOSE, KERN_DEBUG
845 "... PIRQ%d -> IRQ %d\n", i, ints[i+1]);
846 /*
847 * PIRQs are mapped upside down, usually.
848 */
849 pirq_entries[MAX_PIRQS-i-1] = ints[i+1];
850 }
851 return 1;
852 }
853
854 __setup("pirq=", ioapic_pirq_setup);
855 #endif /* CONFIG_X86_32 */
856
857 #ifdef CONFIG_INTR_REMAP
858 /* I/O APIC RTE contents at the OS boot up */
859 static struct IO_APIC_route_entry *early_ioapic_entries[MAX_IO_APICS];
860
861 /*
862 * Saves and masks all the unmasked IO-APIC RTE's
863 */
864 int save_mask_IO_APIC_setup(void)
865 {
866 union IO_APIC_reg_01 reg_01;
867 unsigned long flags;
868 int apic, pin;
869
870 /*
871 * The number of IO-APIC IRQ registers (== #pins):
872 */
873 for (apic = 0; apic < nr_ioapics; apic++) {
874 spin_lock_irqsave(&ioapic_lock, flags);
875 reg_01.raw = io_apic_read(apic, 1);
876 spin_unlock_irqrestore(&ioapic_lock, flags);
877 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
878 }
879
880 for (apic = 0; apic < nr_ioapics; apic++) {
881 early_ioapic_entries[apic] =
882 kzalloc(sizeof(struct IO_APIC_route_entry) *
883 nr_ioapic_registers[apic], GFP_KERNEL);
884 if (!early_ioapic_entries[apic])
885 goto nomem;
886 }
887
888 for (apic = 0; apic < nr_ioapics; apic++)
889 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
890 struct IO_APIC_route_entry entry;
891
892 entry = early_ioapic_entries[apic][pin] =
893 ioapic_read_entry(apic, pin);
894 if (!entry.mask) {
895 entry.mask = 1;
896 ioapic_write_entry(apic, pin, entry);
897 }
898 }
899
900 return 0;
901
902 nomem:
903 while (apic >= 0)
904 kfree(early_ioapic_entries[apic--]);
905 memset(early_ioapic_entries, 0,
906 ARRAY_SIZE(early_ioapic_entries));
907
908 return -ENOMEM;
909 }
910
911 void restore_IO_APIC_setup(void)
912 {
913 int apic, pin;
914
915 for (apic = 0; apic < nr_ioapics; apic++) {
916 if (!early_ioapic_entries[apic])
917 break;
918 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
919 ioapic_write_entry(apic, pin,
920 early_ioapic_entries[apic][pin]);
921 kfree(early_ioapic_entries[apic]);
922 early_ioapic_entries[apic] = NULL;
923 }
924 }
925
926 void reinit_intr_remapped_IO_APIC(int intr_remapping)
927 {
928 /*
929 * for now plain restore of previous settings.
930 * TBD: In the case of OS enabling interrupt-remapping,
931 * IO-APIC RTE's need to be setup to point to interrupt-remapping
932 * table entries. for now, do a plain restore, and wait for
933 * the setup_IO_APIC_irqs() to do proper initialization.
934 */
935 restore_IO_APIC_setup();
936 }
937 #endif
938
939 /*
940 * Find the IRQ entry number of a certain pin.
941 */
942 static int find_irq_entry(int apic, int pin, int type)
943 {
944 int i;
945
946 for (i = 0; i < mp_irq_entries; i++)
947 if (mp_irqs[i].mp_irqtype == type &&
948 (mp_irqs[i].mp_dstapic == mp_ioapics[apic].mp_apicid ||
949 mp_irqs[i].mp_dstapic == MP_APIC_ALL) &&
950 mp_irqs[i].mp_dstirq == pin)
951 return i;
952
953 return -1;
954 }
955
956 /*
957 * Find the pin to which IRQ[irq] (ISA) is connected
958 */
959 static int __init find_isa_irq_pin(int irq, int type)
960 {
961 int i;
962
963 for (i = 0; i < mp_irq_entries; i++) {
964 int lbus = mp_irqs[i].mp_srcbus;
965
966 if (test_bit(lbus, mp_bus_not_pci) &&
967 (mp_irqs[i].mp_irqtype == type) &&
968 (mp_irqs[i].mp_srcbusirq == irq))
969
970 return mp_irqs[i].mp_dstirq;
971 }
972 return -1;
973 }
974
975 static int __init find_isa_irq_apic(int irq, int type)
976 {
977 int i;
978
979 for (i = 0; i < mp_irq_entries; i++) {
980 int lbus = mp_irqs[i].mp_srcbus;
981
982 if (test_bit(lbus, mp_bus_not_pci) &&
983 (mp_irqs[i].mp_irqtype == type) &&
984 (mp_irqs[i].mp_srcbusirq == irq))
985 break;
986 }
987 if (i < mp_irq_entries) {
988 int apic;
989 for(apic = 0; apic < nr_ioapics; apic++) {
990 if (mp_ioapics[apic].mp_apicid == mp_irqs[i].mp_dstapic)
991 return apic;
992 }
993 }
994
995 return -1;
996 }
997
998 /*
999 * Find a specific PCI IRQ entry.
1000 * Not an __init, possibly needed by modules
1001 */
1002 static int pin_2_irq(int idx, int apic, int pin);
1003
1004 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
1005 {
1006 int apic, i, best_guess = -1;
1007
1008 apic_printk(APIC_DEBUG, "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
1009 bus, slot, pin);
1010 if (test_bit(bus, mp_bus_not_pci)) {
1011 apic_printk(APIC_VERBOSE, "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
1012 return -1;
1013 }
1014 for (i = 0; i < mp_irq_entries; i++) {
1015 int lbus = mp_irqs[i].mp_srcbus;
1016
1017 for (apic = 0; apic < nr_ioapics; apic++)
1018 if (mp_ioapics[apic].mp_apicid == mp_irqs[i].mp_dstapic ||
1019 mp_irqs[i].mp_dstapic == MP_APIC_ALL)
1020 break;
1021
1022 if (!test_bit(lbus, mp_bus_not_pci) &&
1023 !mp_irqs[i].mp_irqtype &&
1024 (bus == lbus) &&
1025 (slot == ((mp_irqs[i].mp_srcbusirq >> 2) & 0x1f))) {
1026 int irq = pin_2_irq(i,apic,mp_irqs[i].mp_dstirq);
1027
1028 if (!(apic || IO_APIC_IRQ(irq)))
1029 continue;
1030
1031 if (pin == (mp_irqs[i].mp_srcbusirq & 3))
1032 return irq;
1033 /*
1034 * Use the first all-but-pin matching entry as a
1035 * best-guess fuzzy result for broken mptables.
1036 */
1037 if (best_guess < 0)
1038 best_guess = irq;
1039 }
1040 }
1041 return best_guess;
1042 }
1043
1044 EXPORT_SYMBOL(IO_APIC_get_PCI_irq_vector);
1045
1046 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
1047 /*
1048 * EISA Edge/Level control register, ELCR
1049 */
1050 static int EISA_ELCR(unsigned int irq)
1051 {
1052 if (irq < NR_IRQS_LEGACY) {
1053 unsigned int port = 0x4d0 + (irq >> 3);
1054 return (inb(port) >> (irq & 7)) & 1;
1055 }
1056 apic_printk(APIC_VERBOSE, KERN_INFO
1057 "Broken MPtable reports ISA irq %d\n", irq);
1058 return 0;
1059 }
1060
1061 #endif
1062
1063 /* ISA interrupts are always polarity zero edge triggered,
1064 * when listed as conforming in the MP table. */
1065
1066 #define default_ISA_trigger(idx) (0)
1067 #define default_ISA_polarity(idx) (0)
1068
1069 /* EISA interrupts are always polarity zero and can be edge or level
1070 * trigger depending on the ELCR value. If an interrupt is listed as
1071 * EISA conforming in the MP table, that means its trigger type must
1072 * be read in from the ELCR */
1073
1074 #define default_EISA_trigger(idx) (EISA_ELCR(mp_irqs[idx].mp_srcbusirq))
1075 #define default_EISA_polarity(idx) default_ISA_polarity(idx)
1076
1077 /* PCI interrupts are always polarity one level triggered,
1078 * when listed as conforming in the MP table. */
1079
1080 #define default_PCI_trigger(idx) (1)
1081 #define default_PCI_polarity(idx) (1)
1082
1083 /* MCA interrupts are always polarity zero level triggered,
1084 * when listed as conforming in the MP table. */
1085
1086 #define default_MCA_trigger(idx) (1)
1087 #define default_MCA_polarity(idx) default_ISA_polarity(idx)
1088
1089 static int MPBIOS_polarity(int idx)
1090 {
1091 int bus = mp_irqs[idx].mp_srcbus;
1092 int polarity;
1093
1094 /*
1095 * Determine IRQ line polarity (high active or low active):
1096 */
1097 switch (mp_irqs[idx].mp_irqflag & 3)
1098 {
1099 case 0: /* conforms, ie. bus-type dependent polarity */
1100 if (test_bit(bus, mp_bus_not_pci))
1101 polarity = default_ISA_polarity(idx);
1102 else
1103 polarity = default_PCI_polarity(idx);
1104 break;
1105 case 1: /* high active */
1106 {
1107 polarity = 0;
1108 break;
1109 }
1110 case 2: /* reserved */
1111 {
1112 printk(KERN_WARNING "broken BIOS!!\n");
1113 polarity = 1;
1114 break;
1115 }
1116 case 3: /* low active */
1117 {
1118 polarity = 1;
1119 break;
1120 }
1121 default: /* invalid */
1122 {
1123 printk(KERN_WARNING "broken BIOS!!\n");
1124 polarity = 1;
1125 break;
1126 }
1127 }
1128 return polarity;
1129 }
1130
1131 static int MPBIOS_trigger(int idx)
1132 {
1133 int bus = mp_irqs[idx].mp_srcbus;
1134 int trigger;
1135
1136 /*
1137 * Determine IRQ trigger mode (edge or level sensitive):
1138 */
1139 switch ((mp_irqs[idx].mp_irqflag>>2) & 3)
1140 {
1141 case 0: /* conforms, ie. bus-type dependent */
1142 if (test_bit(bus, mp_bus_not_pci))
1143 trigger = default_ISA_trigger(idx);
1144 else
1145 trigger = default_PCI_trigger(idx);
1146 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
1147 switch (mp_bus_id_to_type[bus]) {
1148 case MP_BUS_ISA: /* ISA pin */
1149 {
1150 /* set before the switch */
1151 break;
1152 }
1153 case MP_BUS_EISA: /* EISA pin */
1154 {
1155 trigger = default_EISA_trigger(idx);
1156 break;
1157 }
1158 case MP_BUS_PCI: /* PCI pin */
1159 {
1160 /* set before the switch */
1161 break;
1162 }
1163 case MP_BUS_MCA: /* MCA pin */
1164 {
1165 trigger = default_MCA_trigger(idx);
1166 break;
1167 }
1168 default:
1169 {
1170 printk(KERN_WARNING "broken BIOS!!\n");
1171 trigger = 1;
1172 break;
1173 }
1174 }
1175 #endif
1176 break;
1177 case 1: /* edge */
1178 {
1179 trigger = 0;
1180 break;
1181 }
1182 case 2: /* reserved */
1183 {
1184 printk(KERN_WARNING "broken BIOS!!\n");
1185 trigger = 1;
1186 break;
1187 }
1188 case 3: /* level */
1189 {
1190 trigger = 1;
1191 break;
1192 }
1193 default: /* invalid */
1194 {
1195 printk(KERN_WARNING "broken BIOS!!\n");
1196 trigger = 0;
1197 break;
1198 }
1199 }
1200 return trigger;
1201 }
1202
1203 static inline int irq_polarity(int idx)
1204 {
1205 return MPBIOS_polarity(idx);
1206 }
1207
1208 static inline int irq_trigger(int idx)
1209 {
1210 return MPBIOS_trigger(idx);
1211 }
1212
1213 int (*ioapic_renumber_irq)(int ioapic, int irq);
1214 static int pin_2_irq(int idx, int apic, int pin)
1215 {
1216 int irq, i;
1217 int bus = mp_irqs[idx].mp_srcbus;
1218
1219 /*
1220 * Debugging check, we are in big trouble if this message pops up!
1221 */
1222 if (mp_irqs[idx].mp_dstirq != pin)
1223 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
1224
1225 if (test_bit(bus, mp_bus_not_pci)) {
1226 irq = mp_irqs[idx].mp_srcbusirq;
1227 } else {
1228 /*
1229 * PCI IRQs are mapped in order
1230 */
1231 i = irq = 0;
1232 while (i < apic)
1233 irq += nr_ioapic_registers[i++];
1234 irq += pin;
1235 /*
1236 * For MPS mode, so far only needed by ES7000 platform
1237 */
1238 if (ioapic_renumber_irq)
1239 irq = ioapic_renumber_irq(apic, irq);
1240 }
1241
1242 #ifdef CONFIG_X86_32
1243 /*
1244 * PCI IRQ command line redirection. Yes, limits are hardcoded.
1245 */
1246 if ((pin >= 16) && (pin <= 23)) {
1247 if (pirq_entries[pin-16] != -1) {
1248 if (!pirq_entries[pin-16]) {
1249 apic_printk(APIC_VERBOSE, KERN_DEBUG
1250 "disabling PIRQ%d\n", pin-16);
1251 } else {
1252 irq = pirq_entries[pin-16];
1253 apic_printk(APIC_VERBOSE, KERN_DEBUG
1254 "using PIRQ%d -> IRQ %d\n",
1255 pin-16, irq);
1256 }
1257 }
1258 }
1259 #endif
1260
1261 return irq;
1262 }
1263
1264 void lock_vector_lock(void)
1265 {
1266 /* Used to the online set of cpus does not change
1267 * during assign_irq_vector.
1268 */
1269 spin_lock(&vector_lock);
1270 }
1271
1272 void unlock_vector_lock(void)
1273 {
1274 spin_unlock(&vector_lock);
1275 }
1276
1277 static int
1278 __assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask)
1279 {
1280 /*
1281 * NOTE! The local APIC isn't very good at handling
1282 * multiple interrupts at the same interrupt level.
1283 * As the interrupt level is determined by taking the
1284 * vector number and shifting that right by 4, we
1285 * want to spread these out a bit so that they don't
1286 * all fall in the same interrupt level.
1287 *
1288 * Also, we've got to be careful not to trash gate
1289 * 0x80, because int 0x80 is hm, kind of importantish. ;)
1290 */
1291 static int current_vector = FIRST_DEVICE_VECTOR, current_offset = 0;
1292 unsigned int old_vector;
1293 int cpu, err;
1294 cpumask_var_t tmp_mask;
1295
1296 if ((cfg->move_in_progress) || cfg->move_cleanup_count)
1297 return -EBUSY;
1298
1299 if (!alloc_cpumask_var(&tmp_mask, GFP_ATOMIC))
1300 return -ENOMEM;
1301
1302 old_vector = cfg->vector;
1303 if (old_vector) {
1304 cpumask_and(tmp_mask, mask, cpu_online_mask);
1305 cpumask_and(tmp_mask, cfg->domain, tmp_mask);
1306 if (!cpumask_empty(tmp_mask)) {
1307 free_cpumask_var(tmp_mask);
1308 return 0;
1309 }
1310 }
1311
1312 /* Only try and allocate irqs on cpus that are present */
1313 err = -ENOSPC;
1314 for_each_cpu_and(cpu, mask, cpu_online_mask) {
1315 int new_cpu;
1316 int vector, offset;
1317
1318 vector_allocation_domain(cpu, tmp_mask);
1319
1320 vector = current_vector;
1321 offset = current_offset;
1322 next:
1323 vector += 8;
1324 if (vector >= first_system_vector) {
1325 /* If out of vectors on large boxen, must share them. */
1326 offset = (offset + 1) % 8;
1327 vector = FIRST_DEVICE_VECTOR + offset;
1328 }
1329 if (unlikely(current_vector == vector))
1330 continue;
1331
1332 if (test_bit(vector, used_vectors))
1333 goto next;
1334
1335 for_each_cpu_and(new_cpu, tmp_mask, cpu_online_mask)
1336 if (per_cpu(vector_irq, new_cpu)[vector] != -1)
1337 goto next;
1338 /* Found one! */
1339 current_vector = vector;
1340 current_offset = offset;
1341 if (old_vector) {
1342 cfg->move_in_progress = 1;
1343 cpumask_copy(cfg->old_domain, cfg->domain);
1344 }
1345 for_each_cpu_and(new_cpu, tmp_mask, cpu_online_mask)
1346 per_cpu(vector_irq, new_cpu)[vector] = irq;
1347 cfg->vector = vector;
1348 cpumask_copy(cfg->domain, tmp_mask);
1349 err = 0;
1350 break;
1351 }
1352 free_cpumask_var(tmp_mask);
1353 return err;
1354 }
1355
1356 static int
1357 assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask)
1358 {
1359 int err;
1360 unsigned long flags;
1361
1362 spin_lock_irqsave(&vector_lock, flags);
1363 err = __assign_irq_vector(irq, cfg, mask);
1364 spin_unlock_irqrestore(&vector_lock, flags);
1365 return err;
1366 }
1367
1368 static void __clear_irq_vector(int irq, struct irq_cfg *cfg)
1369 {
1370 int cpu, vector;
1371
1372 BUG_ON(!cfg->vector);
1373
1374 vector = cfg->vector;
1375 for_each_cpu_and(cpu, cfg->domain, cpu_online_mask)
1376 per_cpu(vector_irq, cpu)[vector] = -1;
1377
1378 cfg->vector = 0;
1379 cpumask_clear(cfg->domain);
1380
1381 if (likely(!cfg->move_in_progress))
1382 return;
1383 for_each_cpu_and(cpu, cfg->old_domain, cpu_online_mask) {
1384 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS;
1385 vector++) {
1386 if (per_cpu(vector_irq, cpu)[vector] != irq)
1387 continue;
1388 per_cpu(vector_irq, cpu)[vector] = -1;
1389 break;
1390 }
1391 }
1392 cfg->move_in_progress = 0;
1393 }
1394
1395 void __setup_vector_irq(int cpu)
1396 {
1397 /* Initialize vector_irq on a new cpu */
1398 /* This function must be called with vector_lock held */
1399 int irq, vector;
1400 struct irq_cfg *cfg;
1401 struct irq_desc *desc;
1402
1403 /* Mark the inuse vectors */
1404 for_each_irq_desc(irq, desc) {
1405 cfg = desc->chip_data;
1406 if (!cpumask_test_cpu(cpu, cfg->domain))
1407 continue;
1408 vector = cfg->vector;
1409 per_cpu(vector_irq, cpu)[vector] = irq;
1410 }
1411 /* Mark the free vectors */
1412 for (vector = 0; vector < NR_VECTORS; ++vector) {
1413 irq = per_cpu(vector_irq, cpu)[vector];
1414 if (irq < 0)
1415 continue;
1416
1417 cfg = irq_cfg(irq);
1418 if (!cpumask_test_cpu(cpu, cfg->domain))
1419 per_cpu(vector_irq, cpu)[vector] = -1;
1420 }
1421 }
1422
1423 static struct irq_chip ioapic_chip;
1424 #ifdef CONFIG_INTR_REMAP
1425 static struct irq_chip ir_ioapic_chip;
1426 #endif
1427
1428 #define IOAPIC_AUTO -1
1429 #define IOAPIC_EDGE 0
1430 #define IOAPIC_LEVEL 1
1431
1432 #ifdef CONFIG_X86_32
1433 static inline int IO_APIC_irq_trigger(int irq)
1434 {
1435 int apic, idx, pin;
1436
1437 for (apic = 0; apic < nr_ioapics; apic++) {
1438 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1439 idx = find_irq_entry(apic, pin, mp_INT);
1440 if ((idx != -1) && (irq == pin_2_irq(idx, apic, pin)))
1441 return irq_trigger(idx);
1442 }
1443 }
1444 /*
1445 * nonexistent IRQs are edge default
1446 */
1447 return 0;
1448 }
1449 #else
1450 static inline int IO_APIC_irq_trigger(int irq)
1451 {
1452 return 1;
1453 }
1454 #endif
1455
1456 static void ioapic_register_intr(int irq, struct irq_desc *desc, unsigned long trigger)
1457 {
1458
1459 if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1460 trigger == IOAPIC_LEVEL)
1461 desc->status |= IRQ_LEVEL;
1462 else
1463 desc->status &= ~IRQ_LEVEL;
1464
1465 #ifdef CONFIG_INTR_REMAP
1466 if (irq_remapped(irq)) {
1467 desc->status |= IRQ_MOVE_PCNTXT;
1468 if (trigger)
1469 set_irq_chip_and_handler_name(irq, &ir_ioapic_chip,
1470 handle_fasteoi_irq,
1471 "fasteoi");
1472 else
1473 set_irq_chip_and_handler_name(irq, &ir_ioapic_chip,
1474 handle_edge_irq, "edge");
1475 return;
1476 }
1477 #endif
1478 if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1479 trigger == IOAPIC_LEVEL)
1480 set_irq_chip_and_handler_name(irq, &ioapic_chip,
1481 handle_fasteoi_irq,
1482 "fasteoi");
1483 else
1484 set_irq_chip_and_handler_name(irq, &ioapic_chip,
1485 handle_edge_irq, "edge");
1486 }
1487
1488 static int setup_ioapic_entry(int apic, int irq,
1489 struct IO_APIC_route_entry *entry,
1490 unsigned int destination, int trigger,
1491 int polarity, int vector)
1492 {
1493 /*
1494 * add it to the IO-APIC irq-routing table:
1495 */
1496 memset(entry,0,sizeof(*entry));
1497
1498 #ifdef CONFIG_INTR_REMAP
1499 if (intr_remapping_enabled) {
1500 struct intel_iommu *iommu = map_ioapic_to_ir(apic);
1501 struct irte irte;
1502 struct IR_IO_APIC_route_entry *ir_entry =
1503 (struct IR_IO_APIC_route_entry *) entry;
1504 int index;
1505
1506 if (!iommu)
1507 panic("No mapping iommu for ioapic %d\n", apic);
1508
1509 index = alloc_irte(iommu, irq, 1);
1510 if (index < 0)
1511 panic("Failed to allocate IRTE for ioapic %d\n", apic);
1512
1513 memset(&irte, 0, sizeof(irte));
1514
1515 irte.present = 1;
1516 irte.dst_mode = INT_DEST_MODE;
1517 irte.trigger_mode = trigger;
1518 irte.dlvry_mode = INT_DELIVERY_MODE;
1519 irte.vector = vector;
1520 irte.dest_id = IRTE_DEST(destination);
1521
1522 modify_irte(irq, &irte);
1523
1524 ir_entry->index2 = (index >> 15) & 0x1;
1525 ir_entry->zero = 0;
1526 ir_entry->format = 1;
1527 ir_entry->index = (index & 0x7fff);
1528 } else
1529 #endif
1530 {
1531 entry->delivery_mode = INT_DELIVERY_MODE;
1532 entry->dest_mode = INT_DEST_MODE;
1533 entry->dest = destination;
1534 }
1535
1536 entry->mask = 0; /* enable IRQ */
1537 entry->trigger = trigger;
1538 entry->polarity = polarity;
1539 entry->vector = vector;
1540
1541 /* Mask level triggered irqs.
1542 * Use IRQ_DELAYED_DISABLE for edge triggered irqs.
1543 */
1544 if (trigger)
1545 entry->mask = 1;
1546 return 0;
1547 }
1548
1549 static void setup_IO_APIC_irq(int apic, int pin, unsigned int irq, struct irq_desc *desc,
1550 int trigger, int polarity)
1551 {
1552 struct irq_cfg *cfg;
1553 struct IO_APIC_route_entry entry;
1554 unsigned int dest;
1555
1556 if (!IO_APIC_IRQ(irq))
1557 return;
1558
1559 cfg = desc->chip_data;
1560
1561 if (assign_irq_vector(irq, cfg, TARGET_CPUS))
1562 return;
1563
1564 dest = cpu_mask_to_apicid_and(cfg->domain, TARGET_CPUS);
1565
1566 apic_printk(APIC_VERBOSE,KERN_DEBUG
1567 "IOAPIC[%d]: Set routing entry (%d-%d -> 0x%x -> "
1568 "IRQ %d Mode:%i Active:%i)\n",
1569 apic, mp_ioapics[apic].mp_apicid, pin, cfg->vector,
1570 irq, trigger, polarity);
1571
1572
1573 if (setup_ioapic_entry(mp_ioapics[apic].mp_apicid, irq, &entry,
1574 dest, trigger, polarity, cfg->vector)) {
1575 printk("Failed to setup ioapic entry for ioapic %d, pin %d\n",
1576 mp_ioapics[apic].mp_apicid, pin);
1577 __clear_irq_vector(irq, cfg);
1578 return;
1579 }
1580
1581 ioapic_register_intr(irq, desc, trigger);
1582 if (irq < NR_IRQS_LEGACY)
1583 disable_8259A_irq(irq);
1584
1585 ioapic_write_entry(apic, pin, entry);
1586 }
1587
1588 static void __init setup_IO_APIC_irqs(void)
1589 {
1590 int apic, pin, idx, irq;
1591 int notcon = 0;
1592 struct irq_desc *desc;
1593 struct irq_cfg *cfg;
1594 int cpu = boot_cpu_id;
1595
1596 apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
1597
1598 for (apic = 0; apic < nr_ioapics; apic++) {
1599 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1600
1601 idx = find_irq_entry(apic, pin, mp_INT);
1602 if (idx == -1) {
1603 if (!notcon) {
1604 notcon = 1;
1605 apic_printk(APIC_VERBOSE,
1606 KERN_DEBUG " %d-%d",
1607 mp_ioapics[apic].mp_apicid,
1608 pin);
1609 } else
1610 apic_printk(APIC_VERBOSE, " %d-%d",
1611 mp_ioapics[apic].mp_apicid,
1612 pin);
1613 continue;
1614 }
1615 if (notcon) {
1616 apic_printk(APIC_VERBOSE,
1617 " (apicid-pin) not connected\n");
1618 notcon = 0;
1619 }
1620
1621 irq = pin_2_irq(idx, apic, pin);
1622 #ifdef CONFIG_X86_32
1623 if (multi_timer_check(apic, irq))
1624 continue;
1625 #endif
1626 desc = irq_to_desc_alloc_cpu(irq, cpu);
1627 if (!desc) {
1628 printk(KERN_INFO "can not get irq_desc for %d\n", irq);
1629 continue;
1630 }
1631 cfg = desc->chip_data;
1632 add_pin_to_irq_cpu(cfg, cpu, apic, pin);
1633
1634 setup_IO_APIC_irq(apic, pin, irq, desc,
1635 irq_trigger(idx), irq_polarity(idx));
1636 }
1637 }
1638
1639 if (notcon)
1640 apic_printk(APIC_VERBOSE,
1641 " (apicid-pin) not connected\n");
1642 }
1643
1644 /*
1645 * Set up the timer pin, possibly with the 8259A-master behind.
1646 */
1647 static void __init setup_timer_IRQ0_pin(unsigned int apic, unsigned int pin,
1648 int vector)
1649 {
1650 struct IO_APIC_route_entry entry;
1651
1652 #ifdef CONFIG_INTR_REMAP
1653 if (intr_remapping_enabled)
1654 return;
1655 #endif
1656
1657 memset(&entry, 0, sizeof(entry));
1658
1659 /*
1660 * We use logical delivery to get the timer IRQ
1661 * to the first CPU.
1662 */
1663 entry.dest_mode = INT_DEST_MODE;
1664 entry.mask = 1; /* mask IRQ now */
1665 entry.dest = cpu_mask_to_apicid(TARGET_CPUS);
1666 entry.delivery_mode = INT_DELIVERY_MODE;
1667 entry.polarity = 0;
1668 entry.trigger = 0;
1669 entry.vector = vector;
1670
1671 /*
1672 * The timer IRQ doesn't have to know that behind the
1673 * scene we may have a 8259A-master in AEOI mode ...
1674 */
1675 set_irq_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq, "edge");
1676
1677 /*
1678 * Add it to the IO-APIC irq-routing table:
1679 */
1680 ioapic_write_entry(apic, pin, entry);
1681 }
1682
1683
1684 __apicdebuginit(void) print_IO_APIC(void)
1685 {
1686 int apic, i;
1687 union IO_APIC_reg_00 reg_00;
1688 union IO_APIC_reg_01 reg_01;
1689 union IO_APIC_reg_02 reg_02;
1690 union IO_APIC_reg_03 reg_03;
1691 unsigned long flags;
1692 struct irq_cfg *cfg;
1693 struct irq_desc *desc;
1694 unsigned int irq;
1695
1696 if (apic_verbosity == APIC_QUIET)
1697 return;
1698
1699 printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
1700 for (i = 0; i < nr_ioapics; i++)
1701 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
1702 mp_ioapics[i].mp_apicid, nr_ioapic_registers[i]);
1703
1704 /*
1705 * We are a bit conservative about what we expect. We have to
1706 * know about every hardware change ASAP.
1707 */
1708 printk(KERN_INFO "testing the IO APIC.......................\n");
1709
1710 for (apic = 0; apic < nr_ioapics; apic++) {
1711
1712 spin_lock_irqsave(&ioapic_lock, flags);
1713 reg_00.raw = io_apic_read(apic, 0);
1714 reg_01.raw = io_apic_read(apic, 1);
1715 if (reg_01.bits.version >= 0x10)
1716 reg_02.raw = io_apic_read(apic, 2);
1717 if (reg_01.bits.version >= 0x20)
1718 reg_03.raw = io_apic_read(apic, 3);
1719 spin_unlock_irqrestore(&ioapic_lock, flags);
1720
1721 printk("\n");
1722 printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].mp_apicid);
1723 printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
1724 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID);
1725 printk(KERN_DEBUG "....... : Delivery Type: %X\n", reg_00.bits.delivery_type);
1726 printk(KERN_DEBUG "....... : LTS : %X\n", reg_00.bits.LTS);
1727
1728 printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)&reg_01);
1729 printk(KERN_DEBUG "....... : max redirection entries: %04X\n", reg_01.bits.entries);
1730
1731 printk(KERN_DEBUG "....... : PRQ implemented: %X\n", reg_01.bits.PRQ);
1732 printk(KERN_DEBUG "....... : IO APIC version: %04X\n", reg_01.bits.version);
1733
1734 /*
1735 * Some Intel chipsets with IO APIC VERSION of 0x1? don't have reg_02,
1736 * but the value of reg_02 is read as the previous read register
1737 * value, so ignore it if reg_02 == reg_01.
1738 */
1739 if (reg_01.bits.version >= 0x10 && reg_02.raw != reg_01.raw) {
1740 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
1741 printk(KERN_DEBUG "....... : arbitration: %02X\n", reg_02.bits.arbitration);
1742 }
1743
1744 /*
1745 * Some Intel chipsets with IO APIC VERSION of 0x2? don't have reg_02
1746 * or reg_03, but the value of reg_0[23] is read as the previous read
1747 * register value, so ignore it if reg_03 == reg_0[12].
1748 */
1749 if (reg_01.bits.version >= 0x20 && reg_03.raw != reg_02.raw &&
1750 reg_03.raw != reg_01.raw) {
1751 printk(KERN_DEBUG ".... register #03: %08X\n", reg_03.raw);
1752 printk(KERN_DEBUG "....... : Boot DT : %X\n", reg_03.bits.boot_DT);
1753 }
1754
1755 printk(KERN_DEBUG ".... IRQ redirection table:\n");
1756
1757 printk(KERN_DEBUG " NR Dst Mask Trig IRR Pol"
1758 " Stat Dmod Deli Vect: \n");
1759
1760 for (i = 0; i <= reg_01.bits.entries; i++) {
1761 struct IO_APIC_route_entry entry;
1762
1763 entry = ioapic_read_entry(apic, i);
1764
1765 printk(KERN_DEBUG " %02x %03X ",
1766 i,
1767 entry.dest
1768 );
1769
1770 printk("%1d %1d %1d %1d %1d %1d %1d %02X\n",
1771 entry.mask,
1772 entry.trigger,
1773 entry.irr,
1774 entry.polarity,
1775 entry.delivery_status,
1776 entry.dest_mode,
1777 entry.delivery_mode,
1778 entry.vector
1779 );
1780 }
1781 }
1782 printk(KERN_DEBUG "IRQ to pin mappings:\n");
1783 for_each_irq_desc(irq, desc) {
1784 struct irq_pin_list *entry;
1785
1786 cfg = desc->chip_data;
1787 entry = cfg->irq_2_pin;
1788 if (!entry)
1789 continue;
1790 printk(KERN_DEBUG "IRQ%d ", irq);
1791 for (;;) {
1792 printk("-> %d:%d", entry->apic, entry->pin);
1793 if (!entry->next)
1794 break;
1795 entry = entry->next;
1796 }
1797 printk("\n");
1798 }
1799
1800 printk(KERN_INFO ".................................... done.\n");
1801
1802 return;
1803 }
1804
1805 __apicdebuginit(void) print_APIC_bitfield(int base)
1806 {
1807 unsigned int v;
1808 int i, j;
1809
1810 if (apic_verbosity == APIC_QUIET)
1811 return;
1812
1813 printk(KERN_DEBUG "0123456789abcdef0123456789abcdef\n" KERN_DEBUG);
1814 for (i = 0; i < 8; i++) {
1815 v = apic_read(base + i*0x10);
1816 for (j = 0; j < 32; j++) {
1817 if (v & (1<<j))
1818 printk("1");
1819 else
1820 printk("0");
1821 }
1822 printk("\n");
1823 }
1824 }
1825
1826 __apicdebuginit(void) print_local_APIC(void *dummy)
1827 {
1828 unsigned int v, ver, maxlvt;
1829 u64 icr;
1830
1831 if (apic_verbosity == APIC_QUIET)
1832 return;
1833
1834 printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1835 smp_processor_id(), hard_smp_processor_id());
1836 v = apic_read(APIC_ID);
1837 printk(KERN_INFO "... APIC ID: %08x (%01x)\n", v, read_apic_id());
1838 v = apic_read(APIC_LVR);
1839 printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1840 ver = GET_APIC_VERSION(v);
1841 maxlvt = lapic_get_maxlvt();
1842
1843 v = apic_read(APIC_TASKPRI);
1844 printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1845
1846 if (APIC_INTEGRATED(ver)) { /* !82489DX */
1847 if (!APIC_XAPIC(ver)) {
1848 v = apic_read(APIC_ARBPRI);
1849 printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
1850 v & APIC_ARBPRI_MASK);
1851 }
1852 v = apic_read(APIC_PROCPRI);
1853 printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1854 }
1855
1856 /*
1857 * Remote read supported only in the 82489DX and local APIC for
1858 * Pentium processors.
1859 */
1860 if (!APIC_INTEGRATED(ver) || maxlvt == 3) {
1861 v = apic_read(APIC_RRR);
1862 printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1863 }
1864
1865 v = apic_read(APIC_LDR);
1866 printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1867 if (!x2apic_enabled()) {
1868 v = apic_read(APIC_DFR);
1869 printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1870 }
1871 v = apic_read(APIC_SPIV);
1872 printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1873
1874 printk(KERN_DEBUG "... APIC ISR field:\n");
1875 print_APIC_bitfield(APIC_ISR);
1876 printk(KERN_DEBUG "... APIC TMR field:\n");
1877 print_APIC_bitfield(APIC_TMR);
1878 printk(KERN_DEBUG "... APIC IRR field:\n");
1879 print_APIC_bitfield(APIC_IRR);
1880
1881 if (APIC_INTEGRATED(ver)) { /* !82489DX */
1882 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
1883 apic_write(APIC_ESR, 0);
1884
1885 v = apic_read(APIC_ESR);
1886 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1887 }
1888
1889 icr = apic_icr_read();
1890 printk(KERN_DEBUG "... APIC ICR: %08x\n", (u32)icr);
1891 printk(KERN_DEBUG "... APIC ICR2: %08x\n", (u32)(icr >> 32));
1892
1893 v = apic_read(APIC_LVTT);
1894 printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1895
1896 if (maxlvt > 3) { /* PC is LVT#4. */
1897 v = apic_read(APIC_LVTPC);
1898 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1899 }
1900 v = apic_read(APIC_LVT0);
1901 printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1902 v = apic_read(APIC_LVT1);
1903 printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1904
1905 if (maxlvt > 2) { /* ERR is LVT#3. */
1906 v = apic_read(APIC_LVTERR);
1907 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1908 }
1909
1910 v = apic_read(APIC_TMICT);
1911 printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1912 v = apic_read(APIC_TMCCT);
1913 printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1914 v = apic_read(APIC_TDCR);
1915 printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1916 printk("\n");
1917 }
1918
1919 __apicdebuginit(void) print_all_local_APICs(void)
1920 {
1921 int cpu;
1922
1923 preempt_disable();
1924 for_each_online_cpu(cpu)
1925 smp_call_function_single(cpu, print_local_APIC, NULL, 1);
1926 preempt_enable();
1927 }
1928
1929 __apicdebuginit(void) print_PIC(void)
1930 {
1931 unsigned int v;
1932 unsigned long flags;
1933
1934 if (apic_verbosity == APIC_QUIET)
1935 return;
1936
1937 printk(KERN_DEBUG "\nprinting PIC contents\n");
1938
1939 spin_lock_irqsave(&i8259A_lock, flags);
1940
1941 v = inb(0xa1) << 8 | inb(0x21);
1942 printk(KERN_DEBUG "... PIC IMR: %04x\n", v);
1943
1944 v = inb(0xa0) << 8 | inb(0x20);
1945 printk(KERN_DEBUG "... PIC IRR: %04x\n", v);
1946
1947 outb(0x0b,0xa0);
1948 outb(0x0b,0x20);
1949 v = inb(0xa0) << 8 | inb(0x20);
1950 outb(0x0a,0xa0);
1951 outb(0x0a,0x20);
1952
1953 spin_unlock_irqrestore(&i8259A_lock, flags);
1954
1955 printk(KERN_DEBUG "... PIC ISR: %04x\n", v);
1956
1957 v = inb(0x4d1) << 8 | inb(0x4d0);
1958 printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1959 }
1960
1961 __apicdebuginit(int) print_all_ICs(void)
1962 {
1963 print_PIC();
1964 print_all_local_APICs();
1965 print_IO_APIC();
1966
1967 return 0;
1968 }
1969
1970 fs_initcall(print_all_ICs);
1971
1972
1973 /* Where if anywhere is the i8259 connect in external int mode */
1974 static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
1975
1976 void __init enable_IO_APIC(void)
1977 {
1978 union IO_APIC_reg_01 reg_01;
1979 int i8259_apic, i8259_pin;
1980 int apic;
1981 unsigned long flags;
1982
1983 #ifdef CONFIG_X86_32
1984 int i;
1985 if (!pirqs_enabled)
1986 for (i = 0; i < MAX_PIRQS; i++)
1987 pirq_entries[i] = -1;
1988 #endif
1989
1990 /*
1991 * The number of IO-APIC IRQ registers (== #pins):
1992 */
1993 for (apic = 0; apic < nr_ioapics; apic++) {
1994 spin_lock_irqsave(&ioapic_lock, flags);
1995 reg_01.raw = io_apic_read(apic, 1);
1996 spin_unlock_irqrestore(&ioapic_lock, flags);
1997 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
1998 }
1999 for(apic = 0; apic < nr_ioapics; apic++) {
2000 int pin;
2001 /* See if any of the pins is in ExtINT mode */
2002 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
2003 struct IO_APIC_route_entry entry;
2004 entry = ioapic_read_entry(apic, pin);
2005
2006 /* If the interrupt line is enabled and in ExtInt mode
2007 * I have found the pin where the i8259 is connected.
2008 */
2009 if ((entry.mask == 0) && (entry.delivery_mode == dest_ExtINT)) {
2010 ioapic_i8259.apic = apic;
2011 ioapic_i8259.pin = pin;
2012 goto found_i8259;
2013 }
2014 }
2015 }
2016 found_i8259:
2017 /* Look to see what if the MP table has reported the ExtINT */
2018 /* If we could not find the appropriate pin by looking at the ioapic
2019 * the i8259 probably is not connected the ioapic but give the
2020 * mptable a chance anyway.
2021 */
2022 i8259_pin = find_isa_irq_pin(0, mp_ExtINT);
2023 i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
2024 /* Trust the MP table if nothing is setup in the hardware */
2025 if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
2026 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
2027 ioapic_i8259.pin = i8259_pin;
2028 ioapic_i8259.apic = i8259_apic;
2029 }
2030 /* Complain if the MP table and the hardware disagree */
2031 if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
2032 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
2033 {
2034 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
2035 }
2036
2037 /*
2038 * Do not trust the IO-APIC being empty at bootup
2039 */
2040 clear_IO_APIC();
2041 }
2042
2043 /*
2044 * Not an __init, needed by the reboot code
2045 */
2046 void disable_IO_APIC(void)
2047 {
2048 /*
2049 * Clear the IO-APIC before rebooting:
2050 */
2051 clear_IO_APIC();
2052
2053 /*
2054 * If the i8259 is routed through an IOAPIC
2055 * Put that IOAPIC in virtual wire mode
2056 * so legacy interrupts can be delivered.
2057 */
2058 if (ioapic_i8259.pin != -1) {
2059 struct IO_APIC_route_entry entry;
2060
2061 memset(&entry, 0, sizeof(entry));
2062 entry.mask = 0; /* Enabled */
2063 entry.trigger = 0; /* Edge */
2064 entry.irr = 0;
2065 entry.polarity = 0; /* High */
2066 entry.delivery_status = 0;
2067 entry.dest_mode = 0; /* Physical */
2068 entry.delivery_mode = dest_ExtINT; /* ExtInt */
2069 entry.vector = 0;
2070 entry.dest = read_apic_id();
2071
2072 /*
2073 * Add it to the IO-APIC irq-routing table:
2074 */
2075 ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
2076 }
2077
2078 disconnect_bsp_APIC(ioapic_i8259.pin != -1);
2079 }
2080
2081 #ifdef CONFIG_X86_32
2082 /*
2083 * function to set the IO-APIC physical IDs based on the
2084 * values stored in the MPC table.
2085 *
2086 * by Matt Domsch <Matt_Domsch@dell.com> Tue Dec 21 12:25:05 CST 1999
2087 */
2088
2089 static void __init setup_ioapic_ids_from_mpc(void)
2090 {
2091 union IO_APIC_reg_00 reg_00;
2092 physid_mask_t phys_id_present_map;
2093 int apic;
2094 int i;
2095 unsigned char old_id;
2096 unsigned long flags;
2097
2098 if (x86_quirks->setup_ioapic_ids && x86_quirks->setup_ioapic_ids())
2099 return;
2100
2101 /*
2102 * Don't check I/O APIC IDs for xAPIC systems. They have
2103 * no meaning without the serial APIC bus.
2104 */
2105 if (!(boot_cpu_data.x86_vendor == X86_VENDOR_INTEL)
2106 || APIC_XAPIC(apic_version[boot_cpu_physical_apicid]))
2107 return;
2108 /*
2109 * This is broken; anything with a real cpu count has to
2110 * circumvent this idiocy regardless.
2111 */
2112 phys_id_present_map = ioapic_phys_id_map(phys_cpu_present_map);
2113
2114 /*
2115 * Set the IOAPIC ID to the value stored in the MPC table.
2116 */
2117 for (apic = 0; apic < nr_ioapics; apic++) {
2118
2119 /* Read the register 0 value */
2120 spin_lock_irqsave(&ioapic_lock, flags);
2121 reg_00.raw = io_apic_read(apic, 0);
2122 spin_unlock_irqrestore(&ioapic_lock, flags);
2123
2124 old_id = mp_ioapics[apic].mp_apicid;
2125
2126 if (mp_ioapics[apic].mp_apicid >= get_physical_broadcast()) {
2127 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID is %d in the MPC table!...\n",
2128 apic, mp_ioapics[apic].mp_apicid);
2129 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
2130 reg_00.bits.ID);
2131 mp_ioapics[apic].mp_apicid = reg_00.bits.ID;
2132 }
2133
2134 /*
2135 * Sanity check, is the ID really free? Every APIC in a
2136 * system must have a unique ID or we get lots of nice
2137 * 'stuck on smp_invalidate_needed IPI wait' messages.
2138 */
2139 if (check_apicid_used(phys_id_present_map,
2140 mp_ioapics[apic].mp_apicid)) {
2141 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID %d is already used!...\n",
2142 apic, mp_ioapics[apic].mp_apicid);
2143 for (i = 0; i < get_physical_broadcast(); i++)
2144 if (!physid_isset(i, phys_id_present_map))
2145 break;
2146 if (i >= get_physical_broadcast())
2147 panic("Max APIC ID exceeded!\n");
2148 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
2149 i);
2150 physid_set(i, phys_id_present_map);
2151 mp_ioapics[apic].mp_apicid = i;
2152 } else {
2153 physid_mask_t tmp;
2154 tmp = apicid_to_cpu_present(mp_ioapics[apic].mp_apicid);
2155 apic_printk(APIC_VERBOSE, "Setting %d in the "
2156 "phys_id_present_map\n",
2157 mp_ioapics[apic].mp_apicid);
2158 physids_or(phys_id_present_map, phys_id_present_map, tmp);
2159 }
2160
2161
2162 /*
2163 * We need to adjust the IRQ routing table
2164 * if the ID changed.
2165 */
2166 if (old_id != mp_ioapics[apic].mp_apicid)
2167 for (i = 0; i < mp_irq_entries; i++)
2168 if (mp_irqs[i].mp_dstapic == old_id)
2169 mp_irqs[i].mp_dstapic
2170 = mp_ioapics[apic].mp_apicid;
2171
2172 /*
2173 * Read the right value from the MPC table and
2174 * write it into the ID register.
2175 */
2176 apic_printk(APIC_VERBOSE, KERN_INFO
2177 "...changing IO-APIC physical APIC ID to %d ...",
2178 mp_ioapics[apic].mp_apicid);
2179
2180 reg_00.bits.ID = mp_ioapics[apic].mp_apicid;
2181 spin_lock_irqsave(&ioapic_lock, flags);
2182 io_apic_write(apic, 0, reg_00.raw);
2183 spin_unlock_irqrestore(&ioapic_lock, flags);
2184
2185 /*
2186 * Sanity check
2187 */
2188 spin_lock_irqsave(&ioapic_lock, flags);
2189 reg_00.raw = io_apic_read(apic, 0);
2190 spin_unlock_irqrestore(&ioapic_lock, flags);
2191 if (reg_00.bits.ID != mp_ioapics[apic].mp_apicid)
2192 printk("could not set ID!\n");
2193 else
2194 apic_printk(APIC_VERBOSE, " ok.\n");
2195 }
2196 }
2197 #endif
2198
2199 int no_timer_check __initdata;
2200
2201 static int __init notimercheck(char *s)
2202 {
2203 no_timer_check = 1;
2204 return 1;
2205 }
2206 __setup("no_timer_check", notimercheck);
2207
2208 /*
2209 * There is a nasty bug in some older SMP boards, their mptable lies
2210 * about the timer IRQ. We do the following to work around the situation:
2211 *
2212 * - timer IRQ defaults to IO-APIC IRQ
2213 * - if this function detects that timer IRQs are defunct, then we fall
2214 * back to ISA timer IRQs
2215 */
2216 static int __init timer_irq_works(void)
2217 {
2218 unsigned long t1 = jiffies;
2219 unsigned long flags;
2220
2221 if (no_timer_check)
2222 return 1;
2223
2224 local_save_flags(flags);
2225 local_irq_enable();
2226 /* Let ten ticks pass... */
2227 mdelay((10 * 1000) / HZ);
2228 local_irq_restore(flags);
2229
2230 /*
2231 * Expect a few ticks at least, to be sure some possible
2232 * glue logic does not lock up after one or two first
2233 * ticks in a non-ExtINT mode. Also the local APIC
2234 * might have cached one ExtINT interrupt. Finally, at
2235 * least one tick may be lost due to delays.
2236 */
2237
2238 /* jiffies wrap? */
2239 if (time_after(jiffies, t1 + 4))
2240 return 1;
2241 return 0;
2242 }
2243
2244 /*
2245 * In the SMP+IOAPIC case it might happen that there are an unspecified
2246 * number of pending IRQ events unhandled. These cases are very rare,
2247 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
2248 * better to do it this way as thus we do not have to be aware of
2249 * 'pending' interrupts in the IRQ path, except at this point.
2250 */
2251 /*
2252 * Edge triggered needs to resend any interrupt
2253 * that was delayed but this is now handled in the device
2254 * independent code.
2255 */
2256
2257 /*
2258 * Starting up a edge-triggered IO-APIC interrupt is
2259 * nasty - we need to make sure that we get the edge.
2260 * If it is already asserted for some reason, we need
2261 * return 1 to indicate that is was pending.
2262 *
2263 * This is not complete - we should be able to fake
2264 * an edge even if it isn't on the 8259A...
2265 */
2266
2267 static unsigned int startup_ioapic_irq(unsigned int irq)
2268 {
2269 int was_pending = 0;
2270 unsigned long flags;
2271 struct irq_cfg *cfg;
2272
2273 spin_lock_irqsave(&ioapic_lock, flags);
2274 if (irq < NR_IRQS_LEGACY) {
2275 disable_8259A_irq(irq);
2276 if (i8259A_irq_pending(irq))
2277 was_pending = 1;
2278 }
2279 cfg = irq_cfg(irq);
2280 __unmask_IO_APIC_irq(cfg);
2281 spin_unlock_irqrestore(&ioapic_lock, flags);
2282
2283 return was_pending;
2284 }
2285
2286 #ifdef CONFIG_X86_64
2287 static int ioapic_retrigger_irq(unsigned int irq)
2288 {
2289
2290 struct irq_cfg *cfg = irq_cfg(irq);
2291 unsigned long flags;
2292
2293 spin_lock_irqsave(&vector_lock, flags);
2294 send_IPI_mask(cpumask_of(cpumask_first(cfg->domain)), cfg->vector);
2295 spin_unlock_irqrestore(&vector_lock, flags);
2296
2297 return 1;
2298 }
2299 #else
2300 static int ioapic_retrigger_irq(unsigned int irq)
2301 {
2302 send_IPI_self(irq_cfg(irq)->vector);
2303
2304 return 1;
2305 }
2306 #endif
2307
2308 /*
2309 * Level and edge triggered IO-APIC interrupts need different handling,
2310 * so we use two separate IRQ descriptors. Edge triggered IRQs can be
2311 * handled with the level-triggered descriptor, but that one has slightly
2312 * more overhead. Level-triggered interrupts cannot be handled with the
2313 * edge-triggered handler, without risking IRQ storms and other ugly
2314 * races.
2315 */
2316
2317 #ifdef CONFIG_SMP
2318
2319 #ifdef CONFIG_INTR_REMAP
2320 static void ir_irq_migration(struct work_struct *work);
2321
2322 static DECLARE_DELAYED_WORK(ir_migration_work, ir_irq_migration);
2323
2324 /*
2325 * Migrate the IO-APIC irq in the presence of intr-remapping.
2326 *
2327 * For edge triggered, irq migration is a simple atomic update(of vector
2328 * and cpu destination) of IRTE and flush the hardware cache.
2329 *
2330 * For level triggered, we need to modify the io-apic RTE aswell with the update
2331 * vector information, along with modifying IRTE with vector and destination.
2332 * So irq migration for level triggered is little bit more complex compared to
2333 * edge triggered migration. But the good news is, we use the same algorithm
2334 * for level triggered migration as we have today, only difference being,
2335 * we now initiate the irq migration from process context instead of the
2336 * interrupt context.
2337 *
2338 * In future, when we do a directed EOI (combined with cpu EOI broadcast
2339 * suppression) to the IO-APIC, level triggered irq migration will also be
2340 * as simple as edge triggered migration and we can do the irq migration
2341 * with a simple atomic update to IO-APIC RTE.
2342 */
2343 static void
2344 migrate_ioapic_irq_desc(struct irq_desc *desc, const struct cpumask *mask)
2345 {
2346 struct irq_cfg *cfg;
2347 struct irte irte;
2348 int modify_ioapic_rte;
2349 unsigned int dest;
2350 unsigned long flags;
2351 unsigned int irq;
2352
2353 if (!cpumask_intersects(mask, cpu_online_mask))
2354 return;
2355
2356 irq = desc->irq;
2357 if (get_irte(irq, &irte))
2358 return;
2359
2360 cfg = desc->chip_data;
2361 if (assign_irq_vector(irq, cfg, mask))
2362 return;
2363
2364 set_extra_move_desc(desc, mask);
2365
2366 dest = cpu_mask_to_apicid_and(cfg->domain, mask);
2367
2368 modify_ioapic_rte = desc->status & IRQ_LEVEL;
2369 if (modify_ioapic_rte) {
2370 spin_lock_irqsave(&ioapic_lock, flags);
2371 __target_IO_APIC_irq(irq, dest, cfg);
2372 spin_unlock_irqrestore(&ioapic_lock, flags);
2373 }
2374
2375 irte.vector = cfg->vector;
2376 irte.dest_id = IRTE_DEST(dest);
2377
2378 /*
2379 * Modified the IRTE and flushes the Interrupt entry cache.
2380 */
2381 modify_irte(irq, &irte);
2382
2383 if (cfg->move_in_progress)
2384 send_cleanup_vector(cfg);
2385
2386 cpumask_copy(desc->affinity, mask);
2387 }
2388
2389 static int migrate_irq_remapped_level_desc(struct irq_desc *desc)
2390 {
2391 int ret = -1;
2392 struct irq_cfg *cfg = desc->chip_data;
2393
2394 mask_IO_APIC_irq_desc(desc);
2395
2396 if (io_apic_level_ack_pending(cfg)) {
2397 /*
2398 * Interrupt in progress. Migrating irq now will change the
2399 * vector information in the IO-APIC RTE and that will confuse
2400 * the EOI broadcast performed by cpu.
2401 * So, delay the irq migration to the next instance.
2402 */
2403 schedule_delayed_work(&ir_migration_work, 1);
2404 goto unmask;
2405 }
2406
2407 /* everthing is clear. we have right of way */
2408 migrate_ioapic_irq_desc(desc, desc->pending_mask);
2409
2410 ret = 0;
2411 desc->status &= ~IRQ_MOVE_PENDING;
2412 cpumask_clear(desc->pending_mask);
2413
2414 unmask:
2415 unmask_IO_APIC_irq_desc(desc);
2416
2417 return ret;
2418 }
2419
2420 static void ir_irq_migration(struct work_struct *work)
2421 {
2422 unsigned int irq;
2423 struct irq_desc *desc;
2424
2425 for_each_irq_desc(irq, desc) {
2426 if (desc->status & IRQ_MOVE_PENDING) {
2427 unsigned long flags;
2428
2429 spin_lock_irqsave(&desc->lock, flags);
2430 if (!desc->chip->set_affinity ||
2431 !(desc->status & IRQ_MOVE_PENDING)) {
2432 desc->status &= ~IRQ_MOVE_PENDING;
2433 spin_unlock_irqrestore(&desc->lock, flags);
2434 continue;
2435 }
2436
2437 desc->chip->set_affinity(irq, desc->pending_mask);
2438 spin_unlock_irqrestore(&desc->lock, flags);
2439 }
2440 }
2441 }
2442
2443 /*
2444 * Migrates the IRQ destination in the process context.
2445 */
2446 static void set_ir_ioapic_affinity_irq_desc(struct irq_desc *desc,
2447 const struct cpumask *mask)
2448 {
2449 if (desc->status & IRQ_LEVEL) {
2450 desc->status |= IRQ_MOVE_PENDING;
2451 cpumask_copy(desc->pending_mask, mask);
2452 migrate_irq_remapped_level_desc(desc);
2453 return;
2454 }
2455
2456 migrate_ioapic_irq_desc(desc, mask);
2457 }
2458 static void set_ir_ioapic_affinity_irq(unsigned int irq,
2459 const struct cpumask *mask)
2460 {
2461 struct irq_desc *desc = irq_to_desc(irq);
2462
2463 set_ir_ioapic_affinity_irq_desc(desc, mask);
2464 }
2465 #endif
2466
2467 asmlinkage void smp_irq_move_cleanup_interrupt(void)
2468 {
2469 unsigned vector, me;
2470
2471 ack_APIC_irq();
2472 exit_idle();
2473 irq_enter();
2474
2475 me = smp_processor_id();
2476 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
2477 unsigned int irq;
2478 struct irq_desc *desc;
2479 struct irq_cfg *cfg;
2480 irq = __get_cpu_var(vector_irq)[vector];
2481
2482 if (irq == -1)
2483 continue;
2484
2485 desc = irq_to_desc(irq);
2486 if (!desc)
2487 continue;
2488
2489 cfg = irq_cfg(irq);
2490 spin_lock(&desc->lock);
2491 if (!cfg->move_cleanup_count)
2492 goto unlock;
2493
2494 if (vector == cfg->vector && cpumask_test_cpu(me, cfg->domain))
2495 goto unlock;
2496
2497 __get_cpu_var(vector_irq)[vector] = -1;
2498 cfg->move_cleanup_count--;
2499 unlock:
2500 spin_unlock(&desc->lock);
2501 }
2502
2503 irq_exit();
2504 }
2505
2506 static void irq_complete_move(struct irq_desc **descp)
2507 {
2508 struct irq_desc *desc = *descp;
2509 struct irq_cfg *cfg = desc->chip_data;
2510 unsigned vector, me;
2511
2512 if (likely(!cfg->move_in_progress)) {
2513 #ifdef CONFIG_NUMA_MIGRATE_IRQ_DESC
2514 if (likely(!cfg->move_desc_pending))
2515 return;
2516
2517 /* domain has not changed, but affinity did */
2518 me = smp_processor_id();
2519 if (cpumask_test_cpu(me, desc->affinity)) {
2520 *descp = desc = move_irq_desc(desc, me);
2521 /* get the new one */
2522 cfg = desc->chip_data;
2523 cfg->move_desc_pending = 0;
2524 }
2525 #endif
2526 return;
2527 }
2528
2529 vector = ~get_irq_regs()->orig_ax;
2530 me = smp_processor_id();
2531 #ifdef CONFIG_NUMA_MIGRATE_IRQ_DESC
2532 *descp = desc = move_irq_desc(desc, me);
2533 /* get the new one */
2534 cfg = desc->chip_data;
2535 #endif
2536
2537 if (vector == cfg->vector && cpumask_test_cpu(me, cfg->domain))
2538 send_cleanup_vector(cfg);
2539 }
2540 #else
2541 static inline void irq_complete_move(struct irq_desc **descp) {}
2542 #endif
2543
2544 #ifdef CONFIG_INTR_REMAP
2545 static void ack_x2apic_level(unsigned int irq)
2546 {
2547 ack_x2APIC_irq();
2548 }
2549
2550 static void ack_x2apic_edge(unsigned int irq)
2551 {
2552 ack_x2APIC_irq();
2553 }
2554
2555 #endif
2556
2557 static void ack_apic_edge(unsigned int irq)
2558 {
2559 struct irq_desc *desc = irq_to_desc(irq);
2560
2561 irq_complete_move(&desc);
2562 move_native_irq(irq);
2563 ack_APIC_irq();
2564 }
2565
2566 atomic_t irq_mis_count;
2567
2568 static void ack_apic_level(unsigned int irq)
2569 {
2570 struct irq_desc *desc = irq_to_desc(irq);
2571
2572 #ifdef CONFIG_X86_32
2573 unsigned long v;
2574 int i;
2575 #endif
2576 struct irq_cfg *cfg;
2577 int do_unmask_irq = 0;
2578
2579 irq_complete_move(&desc);
2580 #ifdef CONFIG_GENERIC_PENDING_IRQ
2581 /* If we are moving the irq we need to mask it */
2582 if (unlikely(desc->status & IRQ_MOVE_PENDING)) {
2583 do_unmask_irq = 1;
2584 mask_IO_APIC_irq_desc(desc);
2585 }
2586 #endif
2587
2588 #ifdef CONFIG_X86_32
2589 /*
2590 * It appears there is an erratum which affects at least version 0x11
2591 * of I/O APIC (that's the 82093AA and cores integrated into various
2592 * chipsets). Under certain conditions a level-triggered interrupt is
2593 * erroneously delivered as edge-triggered one but the respective IRR
2594 * bit gets set nevertheless. As a result the I/O unit expects an EOI
2595 * message but it will never arrive and further interrupts are blocked
2596 * from the source. The exact reason is so far unknown, but the
2597 * phenomenon was observed when two consecutive interrupt requests
2598 * from a given source get delivered to the same CPU and the source is
2599 * temporarily disabled in between.
2600 *
2601 * A workaround is to simulate an EOI message manually. We achieve it
2602 * by setting the trigger mode to edge and then to level when the edge
2603 * trigger mode gets detected in the TMR of a local APIC for a
2604 * level-triggered interrupt. We mask the source for the time of the
2605 * operation to prevent an edge-triggered interrupt escaping meanwhile.
2606 * The idea is from Manfred Spraul. --macro
2607 */
2608 cfg = desc->chip_data;
2609 i = cfg->vector;
2610
2611 v = apic_read(APIC_TMR + ((i & ~0x1f) >> 1));
2612 #endif
2613
2614 /*
2615 * We must acknowledge the irq before we move it or the acknowledge will
2616 * not propagate properly.
2617 */
2618 ack_APIC_irq();
2619
2620 /* Now we can move and renable the irq */
2621 if (unlikely(do_unmask_irq)) {
2622 /* Only migrate the irq if the ack has been received.
2623 *
2624 * On rare occasions the broadcast level triggered ack gets
2625 * delayed going to ioapics, and if we reprogram the
2626 * vector while Remote IRR is still set the irq will never
2627 * fire again.
2628 *
2629 * To prevent this scenario we read the Remote IRR bit
2630 * of the ioapic. This has two effects.
2631 * - On any sane system the read of the ioapic will
2632 * flush writes (and acks) going to the ioapic from
2633 * this cpu.
2634 * - We get to see if the ACK has actually been delivered.
2635 *
2636 * Based on failed experiments of reprogramming the
2637 * ioapic entry from outside of irq context starting
2638 * with masking the ioapic entry and then polling until
2639 * Remote IRR was clear before reprogramming the
2640 * ioapic I don't trust the Remote IRR bit to be
2641 * completey accurate.
2642 *
2643 * However there appears to be no other way to plug
2644 * this race, so if the Remote IRR bit is not
2645 * accurate and is causing problems then it is a hardware bug
2646 * and you can go talk to the chipset vendor about it.
2647 */
2648 cfg = desc->chip_data;
2649 if (!io_apic_level_ack_pending(cfg))
2650 move_masked_irq(irq);
2651 unmask_IO_APIC_irq_desc(desc);
2652 }
2653
2654 #ifdef CONFIG_X86_32
2655 if (!(v & (1 << (i & 0x1f)))) {
2656 atomic_inc(&irq_mis_count);
2657 spin_lock(&ioapic_lock);
2658 __mask_and_edge_IO_APIC_irq(cfg);
2659 __unmask_and_level_IO_APIC_irq(cfg);
2660 spin_unlock(&ioapic_lock);
2661 }
2662 #endif
2663 }
2664
2665 static struct irq_chip ioapic_chip __read_mostly = {
2666 .name = "IO-APIC",
2667 .startup = startup_ioapic_irq,
2668 .mask = mask_IO_APIC_irq,
2669 .unmask = unmask_IO_APIC_irq,
2670 .ack = ack_apic_edge,
2671 .eoi = ack_apic_level,
2672 #ifdef CONFIG_SMP
2673 .set_affinity = set_ioapic_affinity_irq,
2674 #endif
2675 .retrigger = ioapic_retrigger_irq,
2676 };
2677
2678 #ifdef CONFIG_INTR_REMAP
2679 static struct irq_chip ir_ioapic_chip __read_mostly = {
2680 .name = "IR-IO-APIC",
2681 .startup = startup_ioapic_irq,
2682 .mask = mask_IO_APIC_irq,
2683 .unmask = unmask_IO_APIC_irq,
2684 .ack = ack_x2apic_edge,
2685 .eoi = ack_x2apic_level,
2686 #ifdef CONFIG_SMP
2687 .set_affinity = set_ir_ioapic_affinity_irq,
2688 #endif
2689 .retrigger = ioapic_retrigger_irq,
2690 };
2691 #endif
2692
2693 static inline void init_IO_APIC_traps(void)
2694 {
2695 int irq;
2696 struct irq_desc *desc;
2697 struct irq_cfg *cfg;
2698
2699 /*
2700 * NOTE! The local APIC isn't very good at handling
2701 * multiple interrupts at the same interrupt level.
2702 * As the interrupt level is determined by taking the
2703 * vector number and shifting that right by 4, we
2704 * want to spread these out a bit so that they don't
2705 * all fall in the same interrupt level.
2706 *
2707 * Also, we've got to be careful not to trash gate
2708 * 0x80, because int 0x80 is hm, kind of importantish. ;)
2709 */
2710 for_each_irq_desc(irq, desc) {
2711 cfg = desc->chip_data;
2712 if (IO_APIC_IRQ(irq) && cfg && !cfg->vector) {
2713 /*
2714 * Hmm.. We don't have an entry for this,
2715 * so default to an old-fashioned 8259
2716 * interrupt if we can..
2717 */
2718 if (irq < NR_IRQS_LEGACY)
2719 make_8259A_irq(irq);
2720 else
2721 /* Strange. Oh, well.. */
2722 desc->chip = &no_irq_chip;
2723 }
2724 }
2725 }
2726
2727 /*
2728 * The local APIC irq-chip implementation:
2729 */
2730
2731 static void mask_lapic_irq(unsigned int irq)
2732 {
2733 unsigned long v;
2734
2735 v = apic_read(APIC_LVT0);
2736 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
2737 }
2738
2739 static void unmask_lapic_irq(unsigned int irq)
2740 {
2741 unsigned long v;
2742
2743 v = apic_read(APIC_LVT0);
2744 apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
2745 }
2746
2747 static void ack_lapic_irq(unsigned int irq)
2748 {
2749 ack_APIC_irq();
2750 }
2751
2752 static struct irq_chip lapic_chip __read_mostly = {
2753 .name = "local-APIC",
2754 .mask = mask_lapic_irq,
2755 .unmask = unmask_lapic_irq,
2756 .ack = ack_lapic_irq,
2757 };
2758
2759 static void lapic_register_intr(int irq, struct irq_desc *desc)
2760 {
2761 desc->status &= ~IRQ_LEVEL;
2762 set_irq_chip_and_handler_name(irq, &lapic_chip, handle_edge_irq,
2763 "edge");
2764 }
2765
2766 static void __init setup_nmi(void)
2767 {
2768 /*
2769 * Dirty trick to enable the NMI watchdog ...
2770 * We put the 8259A master into AEOI mode and
2771 * unmask on all local APICs LVT0 as NMI.
2772 *
2773 * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
2774 * is from Maciej W. Rozycki - so we do not have to EOI from
2775 * the NMI handler or the timer interrupt.
2776 */
2777 apic_printk(APIC_VERBOSE, KERN_INFO "activating NMI Watchdog ...");
2778
2779 enable_NMI_through_LVT0();
2780
2781 apic_printk(APIC_VERBOSE, " done.\n");
2782 }
2783
2784 /*
2785 * This looks a bit hackish but it's about the only one way of sending
2786 * a few INTA cycles to 8259As and any associated glue logic. ICR does
2787 * not support the ExtINT mode, unfortunately. We need to send these
2788 * cycles as some i82489DX-based boards have glue logic that keeps the
2789 * 8259A interrupt line asserted until INTA. --macro
2790 */
2791 static inline void __init unlock_ExtINT_logic(void)
2792 {
2793 int apic, pin, i;
2794 struct IO_APIC_route_entry entry0, entry1;
2795 unsigned char save_control, save_freq_select;
2796
2797 pin = find_isa_irq_pin(8, mp_INT);
2798 if (pin == -1) {
2799 WARN_ON_ONCE(1);
2800 return;
2801 }
2802 apic = find_isa_irq_apic(8, mp_INT);
2803 if (apic == -1) {
2804 WARN_ON_ONCE(1);
2805 return;
2806 }
2807
2808 entry0 = ioapic_read_entry(apic, pin);
2809 clear_IO_APIC_pin(apic, pin);
2810
2811 memset(&entry1, 0, sizeof(entry1));
2812
2813 entry1.dest_mode = 0; /* physical delivery */
2814 entry1.mask = 0; /* unmask IRQ now */
2815 entry1.dest = hard_smp_processor_id();
2816 entry1.delivery_mode = dest_ExtINT;
2817 entry1.polarity = entry0.polarity;
2818 entry1.trigger = 0;
2819 entry1.vector = 0;
2820
2821 ioapic_write_entry(apic, pin, entry1);
2822
2823 save_control = CMOS_READ(RTC_CONTROL);
2824 save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
2825 CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
2826 RTC_FREQ_SELECT);
2827 CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
2828
2829 i = 100;
2830 while (i-- > 0) {
2831 mdelay(10);
2832 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
2833 i -= 10;
2834 }
2835
2836 CMOS_WRITE(save_control, RTC_CONTROL);
2837 CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
2838 clear_IO_APIC_pin(apic, pin);
2839
2840 ioapic_write_entry(apic, pin, entry0);
2841 }
2842
2843 static int disable_timer_pin_1 __initdata;
2844 /* Actually the next is obsolete, but keep it for paranoid reasons -AK */
2845 static int __init disable_timer_pin_setup(char *arg)
2846 {
2847 disable_timer_pin_1 = 1;
2848 return 0;
2849 }
2850 early_param("disable_timer_pin_1", disable_timer_pin_setup);
2851
2852 int timer_through_8259 __initdata;
2853
2854 /*
2855 * This code may look a bit paranoid, but it's supposed to cooperate with
2856 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ
2857 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast
2858 * fanatically on his truly buggy board.
2859 *
2860 * FIXME: really need to revamp this for all platforms.
2861 */
2862 static inline void __init check_timer(void)
2863 {
2864 struct irq_desc *desc = irq_to_desc(0);
2865 struct irq_cfg *cfg = desc->chip_data;
2866 int cpu = boot_cpu_id;
2867 int apic1, pin1, apic2, pin2;
2868 unsigned long flags;
2869 unsigned int ver;
2870 int no_pin1 = 0;
2871
2872 local_irq_save(flags);
2873
2874 ver = apic_read(APIC_LVR);
2875 ver = GET_APIC_VERSION(ver);
2876
2877 /*
2878 * get/set the timer IRQ vector:
2879 */
2880 disable_8259A_irq(0);
2881 assign_irq_vector(0, cfg, TARGET_CPUS);
2882
2883 /*
2884 * As IRQ0 is to be enabled in the 8259A, the virtual
2885 * wire has to be disabled in the local APIC. Also
2886 * timer interrupts need to be acknowledged manually in
2887 * the 8259A for the i82489DX when using the NMI
2888 * watchdog as that APIC treats NMIs as level-triggered.
2889 * The AEOI mode will finish them in the 8259A
2890 * automatically.
2891 */
2892 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
2893 init_8259A(1);
2894 #ifdef CONFIG_X86_32
2895 timer_ack = (nmi_watchdog == NMI_IO_APIC && !APIC_INTEGRATED(ver));
2896 #endif
2897
2898 pin1 = find_isa_irq_pin(0, mp_INT);
2899 apic1 = find_isa_irq_apic(0, mp_INT);
2900 pin2 = ioapic_i8259.pin;
2901 apic2 = ioapic_i8259.apic;
2902
2903 apic_printk(APIC_QUIET, KERN_INFO "..TIMER: vector=0x%02X "
2904 "apic1=%d pin1=%d apic2=%d pin2=%d\n",
2905 cfg->vector, apic1, pin1, apic2, pin2);
2906
2907 /*
2908 * Some BIOS writers are clueless and report the ExtINTA
2909 * I/O APIC input from the cascaded 8259A as the timer
2910 * interrupt input. So just in case, if only one pin
2911 * was found above, try it both directly and through the
2912 * 8259A.
2913 */
2914 if (pin1 == -1) {
2915 #ifdef CONFIG_INTR_REMAP
2916 if (intr_remapping_enabled)
2917 panic("BIOS bug: timer not connected to IO-APIC");
2918 #endif
2919 pin1 = pin2;
2920 apic1 = apic2;
2921 no_pin1 = 1;
2922 } else if (pin2 == -1) {
2923 pin2 = pin1;
2924 apic2 = apic1;
2925 }
2926
2927 if (pin1 != -1) {
2928 /*
2929 * Ok, does IRQ0 through the IOAPIC work?
2930 */
2931 if (no_pin1) {
2932 add_pin_to_irq_cpu(cfg, cpu, apic1, pin1);
2933 setup_timer_IRQ0_pin(apic1, pin1, cfg->vector);
2934 }
2935 unmask_IO_APIC_irq_desc(desc);
2936 if (timer_irq_works()) {
2937 if (nmi_watchdog == NMI_IO_APIC) {
2938 setup_nmi();
2939 enable_8259A_irq(0);
2940 }
2941 if (disable_timer_pin_1 > 0)
2942 clear_IO_APIC_pin(0, pin1);
2943 goto out;
2944 }
2945 #ifdef CONFIG_INTR_REMAP
2946 if (intr_remapping_enabled)
2947 panic("timer doesn't work through Interrupt-remapped IO-APIC");
2948 #endif
2949 clear_IO_APIC_pin(apic1, pin1);
2950 if (!no_pin1)
2951 apic_printk(APIC_QUIET, KERN_ERR "..MP-BIOS bug: "
2952 "8254 timer not connected to IO-APIC\n");
2953
2954 apic_printk(APIC_QUIET, KERN_INFO "...trying to set up timer "
2955 "(IRQ0) through the 8259A ...\n");
2956 apic_printk(APIC_QUIET, KERN_INFO
2957 "..... (found apic %d pin %d) ...\n", apic2, pin2);
2958 /*
2959 * legacy devices should be connected to IO APIC #0
2960 */
2961 replace_pin_at_irq_cpu(cfg, cpu, apic1, pin1, apic2, pin2);
2962 setup_timer_IRQ0_pin(apic2, pin2, cfg->vector);
2963 unmask_IO_APIC_irq_desc(desc);
2964 enable_8259A_irq(0);
2965 if (timer_irq_works()) {
2966 apic_printk(APIC_QUIET, KERN_INFO "....... works.\n");
2967 timer_through_8259 = 1;
2968 if (nmi_watchdog == NMI_IO_APIC) {
2969 disable_8259A_irq(0);
2970 setup_nmi();
2971 enable_8259A_irq(0);
2972 }
2973 goto out;
2974 }
2975 /*
2976 * Cleanup, just in case ...
2977 */
2978 disable_8259A_irq(0);
2979 clear_IO_APIC_pin(apic2, pin2);
2980 apic_printk(APIC_QUIET, KERN_INFO "....... failed.\n");
2981 }
2982
2983 if (nmi_watchdog == NMI_IO_APIC) {
2984 apic_printk(APIC_QUIET, KERN_WARNING "timer doesn't work "
2985 "through the IO-APIC - disabling NMI Watchdog!\n");
2986 nmi_watchdog = NMI_NONE;
2987 }
2988 #ifdef CONFIG_X86_32
2989 timer_ack = 0;
2990 #endif
2991
2992 apic_printk(APIC_QUIET, KERN_INFO
2993 "...trying to set up timer as Virtual Wire IRQ...\n");
2994
2995 lapic_register_intr(0, desc);
2996 apic_write(APIC_LVT0, APIC_DM_FIXED | cfg->vector); /* Fixed mode */
2997 enable_8259A_irq(0);
2998
2999 if (timer_irq_works()) {
3000 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
3001 goto out;
3002 }
3003 disable_8259A_irq(0);
3004 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | cfg->vector);
3005 apic_printk(APIC_QUIET, KERN_INFO "..... failed.\n");
3006
3007 apic_printk(APIC_QUIET, KERN_INFO
3008 "...trying to set up timer as ExtINT IRQ...\n");
3009
3010 init_8259A(0);
3011 make_8259A_irq(0);
3012 apic_write(APIC_LVT0, APIC_DM_EXTINT);
3013
3014 unlock_ExtINT_logic();
3015
3016 if (timer_irq_works()) {
3017 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
3018 goto out;
3019 }
3020 apic_printk(APIC_QUIET, KERN_INFO "..... failed :(.\n");
3021 panic("IO-APIC + timer doesn't work! Boot with apic=debug and send a "
3022 "report. Then try booting with the 'noapic' option.\n");
3023 out:
3024 local_irq_restore(flags);
3025 }
3026
3027 /*
3028 * Traditionally ISA IRQ2 is the cascade IRQ, and is not available
3029 * to devices. However there may be an I/O APIC pin available for
3030 * this interrupt regardless. The pin may be left unconnected, but
3031 * typically it will be reused as an ExtINT cascade interrupt for
3032 * the master 8259A. In the MPS case such a pin will normally be
3033 * reported as an ExtINT interrupt in the MP table. With ACPI
3034 * there is no provision for ExtINT interrupts, and in the absence
3035 * of an override it would be treated as an ordinary ISA I/O APIC
3036 * interrupt, that is edge-triggered and unmasked by default. We
3037 * used to do this, but it caused problems on some systems because
3038 * of the NMI watchdog and sometimes IRQ0 of the 8254 timer using
3039 * the same ExtINT cascade interrupt to drive the local APIC of the
3040 * bootstrap processor. Therefore we refrain from routing IRQ2 to
3041 * the I/O APIC in all cases now. No actual device should request
3042 * it anyway. --macro
3043 */
3044 #define PIC_IRQS (1 << PIC_CASCADE_IR)
3045
3046 void __init setup_IO_APIC(void)
3047 {
3048
3049 #ifdef CONFIG_X86_32
3050 enable_IO_APIC();
3051 #else
3052 /*
3053 * calling enable_IO_APIC() is moved to setup_local_APIC for BP
3054 */
3055 #endif
3056
3057 io_apic_irqs = ~PIC_IRQS;
3058
3059 apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
3060 /*
3061 * Set up IO-APIC IRQ routing.
3062 */
3063 #ifdef CONFIG_X86_32
3064 if (!acpi_ioapic)
3065 setup_ioapic_ids_from_mpc();
3066 #endif
3067 sync_Arb_IDs();
3068 setup_IO_APIC_irqs();
3069 init_IO_APIC_traps();
3070 check_timer();
3071 }
3072
3073 /*
3074 * Called after all the initialization is done. If we didnt find any
3075 * APIC bugs then we can allow the modify fast path
3076 */
3077
3078 static int __init io_apic_bug_finalize(void)
3079 {
3080 if (sis_apic_bug == -1)
3081 sis_apic_bug = 0;
3082 return 0;
3083 }
3084
3085 late_initcall(io_apic_bug_finalize);
3086
3087 struct sysfs_ioapic_data {
3088 struct sys_device dev;
3089 struct IO_APIC_route_entry entry[0];
3090 };
3091 static struct sysfs_ioapic_data * mp_ioapic_data[MAX_IO_APICS];
3092
3093 static int ioapic_suspend(struct sys_device *dev, pm_message_t state)
3094 {
3095 struct IO_APIC_route_entry *entry;
3096 struct sysfs_ioapic_data *data;
3097 int i;
3098
3099 data = container_of(dev, struct sysfs_ioapic_data, dev);
3100 entry = data->entry;
3101 for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ )
3102 *entry = ioapic_read_entry(dev->id, i);
3103
3104 return 0;
3105 }
3106
3107 static int ioapic_resume(struct sys_device *dev)
3108 {
3109 struct IO_APIC_route_entry *entry;
3110 struct sysfs_ioapic_data *data;
3111 unsigned long flags;
3112 union IO_APIC_reg_00 reg_00;
3113 int i;
3114
3115 data = container_of(dev, struct sysfs_ioapic_data, dev);
3116 entry = data->entry;
3117
3118 spin_lock_irqsave(&ioapic_lock, flags);
3119 reg_00.raw = io_apic_read(dev->id, 0);
3120 if (reg_00.bits.ID != mp_ioapics[dev->id].mp_apicid) {
3121 reg_00.bits.ID = mp_ioapics[dev->id].mp_apicid;
3122 io_apic_write(dev->id, 0, reg_00.raw);
3123 }
3124 spin_unlock_irqrestore(&ioapic_lock, flags);
3125 for (i = 0; i < nr_ioapic_registers[dev->id]; i++)
3126 ioapic_write_entry(dev->id, i, entry[i]);
3127
3128 return 0;
3129 }
3130
3131 static struct sysdev_class ioapic_sysdev_class = {
3132 .name = "ioapic",
3133 .suspend = ioapic_suspend,
3134 .resume = ioapic_resume,
3135 };
3136
3137 static int __init ioapic_init_sysfs(void)
3138 {
3139 struct sys_device * dev;
3140 int i, size, error;
3141
3142 error = sysdev_class_register(&ioapic_sysdev_class);
3143 if (error)
3144 return error;
3145
3146 for (i = 0; i < nr_ioapics; i++ ) {
3147 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
3148 * sizeof(struct IO_APIC_route_entry);
3149 mp_ioapic_data[i] = kzalloc(size, GFP_KERNEL);
3150 if (!mp_ioapic_data[i]) {
3151 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
3152 continue;
3153 }
3154 dev = &mp_ioapic_data[i]->dev;
3155 dev->id = i;
3156 dev->cls = &ioapic_sysdev_class;
3157 error = sysdev_register(dev);
3158 if (error) {
3159 kfree(mp_ioapic_data[i]);
3160 mp_ioapic_data[i] = NULL;
3161 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
3162 continue;
3163 }
3164 }
3165
3166 return 0;
3167 }
3168
3169 device_initcall(ioapic_init_sysfs);
3170
3171 /*
3172 * Dynamic irq allocate and deallocation
3173 */
3174 unsigned int create_irq_nr(unsigned int irq_want)
3175 {
3176 /* Allocate an unused irq */
3177 unsigned int irq;
3178 unsigned int new;
3179 unsigned long flags;
3180 struct irq_cfg *cfg_new = NULL;
3181 int cpu = boot_cpu_id;
3182 struct irq_desc *desc_new = NULL;
3183
3184 irq = 0;
3185 spin_lock_irqsave(&vector_lock, flags);
3186 for (new = irq_want; new < nr_irqs; new++) {
3187 if (platform_legacy_irq(new))
3188 continue;
3189
3190 desc_new = irq_to_desc_alloc_cpu(new, cpu);
3191 if (!desc_new) {
3192 printk(KERN_INFO "can not get irq_desc for %d\n", new);
3193 continue;
3194 }
3195 cfg_new = desc_new->chip_data;
3196
3197 if (cfg_new->vector != 0)
3198 continue;
3199 if (__assign_irq_vector(new, cfg_new, TARGET_CPUS) == 0)
3200 irq = new;
3201 break;
3202 }
3203 spin_unlock_irqrestore(&vector_lock, flags);
3204
3205 if (irq > 0) {
3206 dynamic_irq_init(irq);
3207 /* restore it, in case dynamic_irq_init clear it */
3208 if (desc_new)
3209 desc_new->chip_data = cfg_new;
3210 }
3211 return irq;
3212 }
3213
3214 static int nr_irqs_gsi = NR_IRQS_LEGACY;
3215 int create_irq(void)
3216 {
3217 unsigned int irq_want;
3218 int irq;
3219
3220 irq_want = nr_irqs_gsi;
3221 irq = create_irq_nr(irq_want);
3222
3223 if (irq == 0)
3224 irq = -1;
3225
3226 return irq;
3227 }
3228
3229 void destroy_irq(unsigned int irq)
3230 {
3231 unsigned long flags;
3232 struct irq_cfg *cfg;
3233 struct irq_desc *desc;
3234
3235 /* store it, in case dynamic_irq_cleanup clear it */
3236 desc = irq_to_desc(irq);
3237 cfg = desc->chip_data;
3238 dynamic_irq_cleanup(irq);
3239 /* connect back irq_cfg */
3240 if (desc)
3241 desc->chip_data = cfg;
3242
3243 #ifdef CONFIG_INTR_REMAP
3244 free_irte(irq);
3245 #endif
3246 spin_lock_irqsave(&vector_lock, flags);
3247 __clear_irq_vector(irq, cfg);
3248 spin_unlock_irqrestore(&vector_lock, flags);
3249 }
3250
3251 /*
3252 * MSI message composition
3253 */
3254 #ifdef CONFIG_PCI_MSI
3255 static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq, struct msi_msg *msg)
3256 {
3257 struct irq_cfg *cfg;
3258 int err;
3259 unsigned dest;
3260
3261 cfg = irq_cfg(irq);
3262 err = assign_irq_vector(irq, cfg, TARGET_CPUS);
3263 if (err)
3264 return err;
3265
3266 dest = cpu_mask_to_apicid_and(cfg->domain, TARGET_CPUS);
3267
3268 #ifdef CONFIG_INTR_REMAP
3269 if (irq_remapped(irq)) {
3270 struct irte irte;
3271 int ir_index;
3272 u16 sub_handle;
3273
3274 ir_index = map_irq_to_irte_handle(irq, &sub_handle);
3275 BUG_ON(ir_index == -1);
3276
3277 memset (&irte, 0, sizeof(irte));
3278
3279 irte.present = 1;
3280 irte.dst_mode = INT_DEST_MODE;
3281 irte.trigger_mode = 0; /* edge */
3282 irte.dlvry_mode = INT_DELIVERY_MODE;
3283 irte.vector = cfg->vector;
3284 irte.dest_id = IRTE_DEST(dest);
3285
3286 modify_irte(irq, &irte);
3287
3288 msg->address_hi = MSI_ADDR_BASE_HI;
3289 msg->data = sub_handle;
3290 msg->address_lo = MSI_ADDR_BASE_LO | MSI_ADDR_IR_EXT_INT |
3291 MSI_ADDR_IR_SHV |
3292 MSI_ADDR_IR_INDEX1(ir_index) |
3293 MSI_ADDR_IR_INDEX2(ir_index);
3294 } else
3295 #endif
3296 {
3297 msg->address_hi = MSI_ADDR_BASE_HI;
3298 msg->address_lo =
3299 MSI_ADDR_BASE_LO |
3300 ((INT_DEST_MODE == 0) ?
3301 MSI_ADDR_DEST_MODE_PHYSICAL:
3302 MSI_ADDR_DEST_MODE_LOGICAL) |
3303 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
3304 MSI_ADDR_REDIRECTION_CPU:
3305 MSI_ADDR_REDIRECTION_LOWPRI) |
3306 MSI_ADDR_DEST_ID(dest);
3307
3308 msg->data =
3309 MSI_DATA_TRIGGER_EDGE |
3310 MSI_DATA_LEVEL_ASSERT |
3311 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
3312 MSI_DATA_DELIVERY_FIXED:
3313 MSI_DATA_DELIVERY_LOWPRI) |
3314 MSI_DATA_VECTOR(cfg->vector);
3315 }
3316 return err;
3317 }
3318
3319 #ifdef CONFIG_SMP
3320 static void set_msi_irq_affinity(unsigned int irq, const struct cpumask *mask)
3321 {
3322 struct irq_desc *desc = irq_to_desc(irq);
3323 struct irq_cfg *cfg;
3324 struct msi_msg msg;
3325 unsigned int dest;
3326
3327 dest = set_desc_affinity(desc, mask);
3328 if (dest == BAD_APICID)
3329 return;
3330
3331 cfg = desc->chip_data;
3332
3333 read_msi_msg_desc(desc, &msg);
3334
3335 msg.data &= ~MSI_DATA_VECTOR_MASK;
3336 msg.data |= MSI_DATA_VECTOR(cfg->vector);
3337 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3338 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3339
3340 write_msi_msg_desc(desc, &msg);
3341 }
3342 #ifdef CONFIG_INTR_REMAP
3343 /*
3344 * Migrate the MSI irq to another cpumask. This migration is
3345 * done in the process context using interrupt-remapping hardware.
3346 */
3347 static void
3348 ir_set_msi_irq_affinity(unsigned int irq, const struct cpumask *mask)
3349 {
3350 struct irq_desc *desc = irq_to_desc(irq);
3351 struct irq_cfg *cfg = desc->chip_data;
3352 unsigned int dest;
3353 struct irte irte;
3354
3355 if (get_irte(irq, &irte))
3356 return;
3357
3358 dest = set_desc_affinity(desc, mask);
3359 if (dest == BAD_APICID)
3360 return;
3361
3362 irte.vector = cfg->vector;
3363 irte.dest_id = IRTE_DEST(dest);
3364
3365 /*
3366 * atomically update the IRTE with the new destination and vector.
3367 */
3368 modify_irte(irq, &irte);
3369
3370 /*
3371 * After this point, all the interrupts will start arriving
3372 * at the new destination. So, time to cleanup the previous
3373 * vector allocation.
3374 */
3375 if (cfg->move_in_progress)
3376 send_cleanup_vector(cfg);
3377 }
3378
3379 #endif
3380 #endif /* CONFIG_SMP */
3381
3382 /*
3383 * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
3384 * which implement the MSI or MSI-X Capability Structure.
3385 */
3386 static struct irq_chip msi_chip = {
3387 .name = "PCI-MSI",
3388 .unmask = unmask_msi_irq,
3389 .mask = mask_msi_irq,
3390 .ack = ack_apic_edge,
3391 #ifdef CONFIG_SMP
3392 .set_affinity = set_msi_irq_affinity,
3393 #endif
3394 .retrigger = ioapic_retrigger_irq,
3395 };
3396
3397 #ifdef CONFIG_INTR_REMAP
3398 static struct irq_chip msi_ir_chip = {
3399 .name = "IR-PCI-MSI",
3400 .unmask = unmask_msi_irq,
3401 .mask = mask_msi_irq,
3402 .ack = ack_x2apic_edge,
3403 #ifdef CONFIG_SMP
3404 .set_affinity = ir_set_msi_irq_affinity,
3405 #endif
3406 .retrigger = ioapic_retrigger_irq,
3407 };
3408
3409 /*
3410 * Map the PCI dev to the corresponding remapping hardware unit
3411 * and allocate 'nvec' consecutive interrupt-remapping table entries
3412 * in it.
3413 */
3414 static int msi_alloc_irte(struct pci_dev *dev, int irq, int nvec)
3415 {
3416 struct intel_iommu *iommu;
3417 int index;
3418
3419 iommu = map_dev_to_ir(dev);
3420 if (!iommu) {
3421 printk(KERN_ERR
3422 "Unable to map PCI %s to iommu\n", pci_name(dev));
3423 return -ENOENT;
3424 }
3425
3426 index = alloc_irte(iommu, irq, nvec);
3427 if (index < 0) {
3428 printk(KERN_ERR
3429 "Unable to allocate %d IRTE for PCI %s\n", nvec,
3430 pci_name(dev));
3431 return -ENOSPC;
3432 }
3433 return index;
3434 }
3435 #endif
3436
3437 static int setup_msi_irq(struct pci_dev *dev, struct msi_desc *msidesc, int irq)
3438 {
3439 int ret;
3440 struct msi_msg msg;
3441
3442 ret = msi_compose_msg(dev, irq, &msg);
3443 if (ret < 0)
3444 return ret;
3445
3446 set_irq_msi(irq, msidesc);
3447 write_msi_msg(irq, &msg);
3448
3449 #ifdef CONFIG_INTR_REMAP
3450 if (irq_remapped(irq)) {
3451 struct irq_desc *desc = irq_to_desc(irq);
3452 /*
3453 * irq migration in process context
3454 */
3455 desc->status |= IRQ_MOVE_PCNTXT;
3456 set_irq_chip_and_handler_name(irq, &msi_ir_chip, handle_edge_irq, "edge");
3457 } else
3458 #endif
3459 set_irq_chip_and_handler_name(irq, &msi_chip, handle_edge_irq, "edge");
3460
3461 dev_printk(KERN_DEBUG, &dev->dev, "irq %d for MSI/MSI-X\n", irq);
3462
3463 return 0;
3464 }
3465
3466 int arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *msidesc)
3467 {
3468 unsigned int irq;
3469 int ret;
3470 unsigned int irq_want;
3471
3472 irq_want = nr_irqs_gsi;
3473 irq = create_irq_nr(irq_want);
3474 if (irq == 0)
3475 return -1;
3476
3477 #ifdef CONFIG_INTR_REMAP
3478 if (!intr_remapping_enabled)
3479 goto no_ir;
3480
3481 ret = msi_alloc_irte(dev, irq, 1);
3482 if (ret < 0)
3483 goto error;
3484 no_ir:
3485 #endif
3486 ret = setup_msi_irq(dev, msidesc, irq);
3487 if (ret < 0) {
3488 destroy_irq(irq);
3489 return ret;
3490 }
3491 return 0;
3492
3493 #ifdef CONFIG_INTR_REMAP
3494 error:
3495 destroy_irq(irq);
3496 return ret;
3497 #endif
3498 }
3499
3500 int arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
3501 {
3502 unsigned int irq;
3503 int ret, sub_handle;
3504 struct msi_desc *msidesc;
3505 unsigned int irq_want;
3506
3507 #ifdef CONFIG_INTR_REMAP
3508 struct intel_iommu *iommu = 0;
3509 int index = 0;
3510 #endif
3511
3512 irq_want = nr_irqs_gsi;
3513 sub_handle = 0;
3514 list_for_each_entry(msidesc, &dev->msi_list, list) {
3515 irq = create_irq_nr(irq_want);
3516 irq_want++;
3517 if (irq == 0)
3518 return -1;
3519 #ifdef CONFIG_INTR_REMAP
3520 if (!intr_remapping_enabled)
3521 goto no_ir;
3522
3523 if (!sub_handle) {
3524 /*
3525 * allocate the consecutive block of IRTE's
3526 * for 'nvec'
3527 */
3528 index = msi_alloc_irte(dev, irq, nvec);
3529 if (index < 0) {
3530 ret = index;
3531 goto error;
3532 }
3533 } else {
3534 iommu = map_dev_to_ir(dev);
3535 if (!iommu) {
3536 ret = -ENOENT;
3537 goto error;
3538 }
3539 /*
3540 * setup the mapping between the irq and the IRTE
3541 * base index, the sub_handle pointing to the
3542 * appropriate interrupt remap table entry.
3543 */
3544 set_irte_irq(irq, iommu, index, sub_handle);
3545 }
3546 no_ir:
3547 #endif
3548 ret = setup_msi_irq(dev, msidesc, irq);
3549 if (ret < 0)
3550 goto error;
3551 sub_handle++;
3552 }
3553 return 0;
3554
3555 error:
3556 destroy_irq(irq);
3557 return ret;
3558 }
3559
3560 void arch_teardown_msi_irq(unsigned int irq)
3561 {
3562 destroy_irq(irq);
3563 }
3564
3565 #ifdef CONFIG_DMAR
3566 #ifdef CONFIG_SMP
3567 static void dmar_msi_set_affinity(unsigned int irq, const struct cpumask *mask)
3568 {
3569 struct irq_desc *desc = irq_to_desc(irq);
3570 struct irq_cfg *cfg;
3571 struct msi_msg msg;
3572 unsigned int dest;
3573
3574 dest = set_desc_affinity(desc, mask);
3575 if (dest == BAD_APICID)
3576 return;
3577
3578 cfg = desc->chip_data;
3579
3580 dmar_msi_read(irq, &msg);
3581
3582 msg.data &= ~MSI_DATA_VECTOR_MASK;
3583 msg.data |= MSI_DATA_VECTOR(cfg->vector);
3584 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3585 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3586
3587 dmar_msi_write(irq, &msg);
3588 }
3589
3590 #endif /* CONFIG_SMP */
3591
3592 struct irq_chip dmar_msi_type = {
3593 .name = "DMAR_MSI",
3594 .unmask = dmar_msi_unmask,
3595 .mask = dmar_msi_mask,
3596 .ack = ack_apic_edge,
3597 #ifdef CONFIG_SMP
3598 .set_affinity = dmar_msi_set_affinity,
3599 #endif
3600 .retrigger = ioapic_retrigger_irq,
3601 };
3602
3603 int arch_setup_dmar_msi(unsigned int irq)
3604 {
3605 int ret;
3606 struct msi_msg msg;
3607
3608 ret = msi_compose_msg(NULL, irq, &msg);
3609 if (ret < 0)
3610 return ret;
3611 dmar_msi_write(irq, &msg);
3612 set_irq_chip_and_handler_name(irq, &dmar_msi_type, handle_edge_irq,
3613 "edge");
3614 return 0;
3615 }
3616 #endif
3617
3618 #ifdef CONFIG_HPET_TIMER
3619
3620 #ifdef CONFIG_SMP
3621 static void hpet_msi_set_affinity(unsigned int irq, const struct cpumask *mask)
3622 {
3623 struct irq_desc *desc = irq_to_desc(irq);
3624 struct irq_cfg *cfg;
3625 struct msi_msg msg;
3626 unsigned int dest;
3627
3628 dest = set_desc_affinity(desc, mask);
3629 if (dest == BAD_APICID)
3630 return;
3631
3632 cfg = desc->chip_data;
3633
3634 hpet_msi_read(irq, &msg);
3635
3636 msg.data &= ~MSI_DATA_VECTOR_MASK;
3637 msg.data |= MSI_DATA_VECTOR(cfg->vector);
3638 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3639 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3640
3641 hpet_msi_write(irq, &msg);
3642 }
3643
3644 #endif /* CONFIG_SMP */
3645
3646 struct irq_chip hpet_msi_type = {
3647 .name = "HPET_MSI",
3648 .unmask = hpet_msi_unmask,
3649 .mask = hpet_msi_mask,
3650 .ack = ack_apic_edge,
3651 #ifdef CONFIG_SMP
3652 .set_affinity = hpet_msi_set_affinity,
3653 #endif
3654 .retrigger = ioapic_retrigger_irq,
3655 };
3656
3657 int arch_setup_hpet_msi(unsigned int irq)
3658 {
3659 int ret;
3660 struct msi_msg msg;
3661
3662 ret = msi_compose_msg(NULL, irq, &msg);
3663 if (ret < 0)
3664 return ret;
3665
3666 hpet_msi_write(irq, &msg);
3667 set_irq_chip_and_handler_name(irq, &hpet_msi_type, handle_edge_irq,
3668 "edge");
3669
3670 return 0;
3671 }
3672 #endif
3673
3674 #endif /* CONFIG_PCI_MSI */
3675 /*
3676 * Hypertransport interrupt support
3677 */
3678 #ifdef CONFIG_HT_IRQ
3679
3680 #ifdef CONFIG_SMP
3681
3682 static void target_ht_irq(unsigned int irq, unsigned int dest, u8 vector)
3683 {
3684 struct ht_irq_msg msg;
3685 fetch_ht_irq_msg(irq, &msg);
3686
3687 msg.address_lo &= ~(HT_IRQ_LOW_VECTOR_MASK | HT_IRQ_LOW_DEST_ID_MASK);
3688 msg.address_hi &= ~(HT_IRQ_HIGH_DEST_ID_MASK);
3689
3690 msg.address_lo |= HT_IRQ_LOW_VECTOR(vector) | HT_IRQ_LOW_DEST_ID(dest);
3691 msg.address_hi |= HT_IRQ_HIGH_DEST_ID(dest);
3692
3693 write_ht_irq_msg(irq, &msg);
3694 }
3695
3696 static void set_ht_irq_affinity(unsigned int irq, const struct cpumask *mask)
3697 {
3698 struct irq_desc *desc = irq_to_desc(irq);
3699 struct irq_cfg *cfg;
3700 unsigned int dest;
3701
3702 dest = set_desc_affinity(desc, mask);
3703 if (dest == BAD_APICID)
3704 return;
3705
3706 cfg = desc->chip_data;
3707
3708 target_ht_irq(irq, dest, cfg->vector);
3709 }
3710
3711 #endif
3712
3713 static struct irq_chip ht_irq_chip = {
3714 .name = "PCI-HT",
3715 .mask = mask_ht_irq,
3716 .unmask = unmask_ht_irq,
3717 .ack = ack_apic_edge,
3718 #ifdef CONFIG_SMP
3719 .set_affinity = set_ht_irq_affinity,
3720 #endif
3721 .retrigger = ioapic_retrigger_irq,
3722 };
3723
3724 int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev)
3725 {
3726 struct irq_cfg *cfg;
3727 int err;
3728
3729 cfg = irq_cfg(irq);
3730 err = assign_irq_vector(irq, cfg, TARGET_CPUS);
3731 if (!err) {
3732 struct ht_irq_msg msg;
3733 unsigned dest;
3734
3735 dest = cpu_mask_to_apicid_and(cfg->domain, TARGET_CPUS);
3736
3737 msg.address_hi = HT_IRQ_HIGH_DEST_ID(dest);
3738
3739 msg.address_lo =
3740 HT_IRQ_LOW_BASE |
3741 HT_IRQ_LOW_DEST_ID(dest) |
3742 HT_IRQ_LOW_VECTOR(cfg->vector) |
3743 ((INT_DEST_MODE == 0) ?
3744 HT_IRQ_LOW_DM_PHYSICAL :
3745 HT_IRQ_LOW_DM_LOGICAL) |
3746 HT_IRQ_LOW_RQEOI_EDGE |
3747 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
3748 HT_IRQ_LOW_MT_FIXED :
3749 HT_IRQ_LOW_MT_ARBITRATED) |
3750 HT_IRQ_LOW_IRQ_MASKED;
3751
3752 write_ht_irq_msg(irq, &msg);
3753
3754 set_irq_chip_and_handler_name(irq, &ht_irq_chip,
3755 handle_edge_irq, "edge");
3756
3757 dev_printk(KERN_DEBUG, &dev->dev, "irq %d for HT\n", irq);
3758 }
3759 return err;
3760 }
3761 #endif /* CONFIG_HT_IRQ */
3762
3763 #ifdef CONFIG_X86_64
3764 /*
3765 * Re-target the irq to the specified CPU and enable the specified MMR located
3766 * on the specified blade to allow the sending of MSIs to the specified CPU.
3767 */
3768 int arch_enable_uv_irq(char *irq_name, unsigned int irq, int cpu, int mmr_blade,
3769 unsigned long mmr_offset)
3770 {
3771 const struct cpumask *eligible_cpu = cpumask_of(cpu);
3772 struct irq_cfg *cfg;
3773 int mmr_pnode;
3774 unsigned long mmr_value;
3775 struct uv_IO_APIC_route_entry *entry;
3776 unsigned long flags;
3777 int err;
3778
3779 cfg = irq_cfg(irq);
3780
3781 err = assign_irq_vector(irq, cfg, eligible_cpu);
3782 if (err != 0)
3783 return err;
3784
3785 spin_lock_irqsave(&vector_lock, flags);
3786 set_irq_chip_and_handler_name(irq, &uv_irq_chip, handle_percpu_irq,
3787 irq_name);
3788 spin_unlock_irqrestore(&vector_lock, flags);
3789
3790 mmr_value = 0;
3791 entry = (struct uv_IO_APIC_route_entry *)&mmr_value;
3792 BUG_ON(sizeof(struct uv_IO_APIC_route_entry) != sizeof(unsigned long));
3793
3794 entry->vector = cfg->vector;
3795 entry->delivery_mode = INT_DELIVERY_MODE;
3796 entry->dest_mode = INT_DEST_MODE;
3797 entry->polarity = 0;
3798 entry->trigger = 0;
3799 entry->mask = 0;
3800 entry->dest = cpu_mask_to_apicid(eligible_cpu);
3801
3802 mmr_pnode = uv_blade_to_pnode(mmr_blade);
3803 uv_write_global_mmr64(mmr_pnode, mmr_offset, mmr_value);
3804
3805 return irq;
3806 }
3807
3808 /*
3809 * Disable the specified MMR located on the specified blade so that MSIs are
3810 * longer allowed to be sent.
3811 */
3812 void arch_disable_uv_irq(int mmr_blade, unsigned long mmr_offset)
3813 {
3814 unsigned long mmr_value;
3815 struct uv_IO_APIC_route_entry *entry;
3816 int mmr_pnode;
3817
3818 mmr_value = 0;
3819 entry = (struct uv_IO_APIC_route_entry *)&mmr_value;
3820 BUG_ON(sizeof(struct uv_IO_APIC_route_entry) != sizeof(unsigned long));
3821
3822 entry->mask = 1;
3823
3824 mmr_pnode = uv_blade_to_pnode(mmr_blade);
3825 uv_write_global_mmr64(mmr_pnode, mmr_offset, mmr_value);
3826 }
3827 #endif /* CONFIG_X86_64 */
3828
3829 int __init io_apic_get_redir_entries (int ioapic)
3830 {
3831 union IO_APIC_reg_01 reg_01;
3832 unsigned long flags;
3833
3834 spin_lock_irqsave(&ioapic_lock, flags);
3835 reg_01.raw = io_apic_read(ioapic, 1);
3836 spin_unlock_irqrestore(&ioapic_lock, flags);
3837
3838 return reg_01.bits.entries;
3839 }
3840
3841 void __init probe_nr_irqs_gsi(void)
3842 {
3843 int idx;
3844 int nr = 0;
3845
3846 for (idx = 0; idx < nr_ioapics; idx++)
3847 nr += io_apic_get_redir_entries(idx) + 1;
3848
3849 if (nr > nr_irqs_gsi)
3850 nr_irqs_gsi = nr;
3851 }
3852
3853 /* --------------------------------------------------------------------------
3854 ACPI-based IOAPIC Configuration
3855 -------------------------------------------------------------------------- */
3856
3857 #ifdef CONFIG_ACPI
3858
3859 #ifdef CONFIG_X86_32
3860 int __init io_apic_get_unique_id(int ioapic, int apic_id)
3861 {
3862 union IO_APIC_reg_00 reg_00;
3863 static physid_mask_t apic_id_map = PHYSID_MASK_NONE;
3864 physid_mask_t tmp;
3865 unsigned long flags;
3866 int i = 0;
3867
3868 /*
3869 * The P4 platform supports up to 256 APIC IDs on two separate APIC
3870 * buses (one for LAPICs, one for IOAPICs), where predecessors only
3871 * supports up to 16 on one shared APIC bus.
3872 *
3873 * TBD: Expand LAPIC/IOAPIC support on P4-class systems to take full
3874 * advantage of new APIC bus architecture.
3875 */
3876
3877 if (physids_empty(apic_id_map))
3878 apic_id_map = ioapic_phys_id_map(phys_cpu_present_map);
3879
3880 spin_lock_irqsave(&ioapic_lock, flags);
3881 reg_00.raw = io_apic_read(ioapic, 0);
3882 spin_unlock_irqrestore(&ioapic_lock, flags);
3883
3884 if (apic_id >= get_physical_broadcast()) {
3885 printk(KERN_WARNING "IOAPIC[%d]: Invalid apic_id %d, trying "
3886 "%d\n", ioapic, apic_id, reg_00.bits.ID);
3887 apic_id = reg_00.bits.ID;
3888 }
3889
3890 /*
3891 * Every APIC in a system must have a unique ID or we get lots of nice
3892 * 'stuck on smp_invalidate_needed IPI wait' messages.
3893 */
3894 if (check_apicid_used(apic_id_map, apic_id)) {
3895
3896 for (i = 0; i < get_physical_broadcast(); i++) {
3897 if (!check_apicid_used(apic_id_map, i))
3898 break;
3899 }
3900
3901 if (i == get_physical_broadcast())
3902 panic("Max apic_id exceeded!\n");
3903
3904 printk(KERN_WARNING "IOAPIC[%d]: apic_id %d already used, "
3905 "trying %d\n", ioapic, apic_id, i);
3906
3907 apic_id = i;
3908 }
3909
3910 tmp = apicid_to_cpu_present(apic_id);
3911 physids_or(apic_id_map, apic_id_map, tmp);
3912
3913 if (reg_00.bits.ID != apic_id) {
3914 reg_00.bits.ID = apic_id;
3915
3916 spin_lock_irqsave(&ioapic_lock, flags);
3917 io_apic_write(ioapic, 0, reg_00.raw);
3918 reg_00.raw = io_apic_read(ioapic, 0);
3919 spin_unlock_irqrestore(&ioapic_lock, flags);
3920
3921 /* Sanity check */
3922 if (reg_00.bits.ID != apic_id) {
3923 printk("IOAPIC[%d]: Unable to change apic_id!\n", ioapic);
3924 return -1;
3925 }
3926 }
3927
3928 apic_printk(APIC_VERBOSE, KERN_INFO
3929 "IOAPIC[%d]: Assigned apic_id %d\n", ioapic, apic_id);
3930
3931 return apic_id;
3932 }
3933
3934 int __init io_apic_get_version(int ioapic)
3935 {
3936 union IO_APIC_reg_01 reg_01;
3937 unsigned long flags;
3938
3939 spin_lock_irqsave(&ioapic_lock, flags);
3940 reg_01.raw = io_apic_read(ioapic, 1);
3941 spin_unlock_irqrestore(&ioapic_lock, flags);
3942
3943 return reg_01.bits.version;
3944 }
3945 #endif
3946
3947 int io_apic_set_pci_routing (int ioapic, int pin, int irq, int triggering, int polarity)
3948 {
3949 struct irq_desc *desc;
3950 struct irq_cfg *cfg;
3951 int cpu = boot_cpu_id;
3952
3953 if (!IO_APIC_IRQ(irq)) {
3954 apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
3955 ioapic);
3956 return -EINVAL;
3957 }
3958
3959 desc = irq_to_desc_alloc_cpu(irq, cpu);
3960 if (!desc) {
3961 printk(KERN_INFO "can not get irq_desc %d\n", irq);
3962 return 0;
3963 }
3964
3965 /*
3966 * IRQs < 16 are already in the irq_2_pin[] map
3967 */
3968 if (irq >= NR_IRQS_LEGACY) {
3969 cfg = desc->chip_data;
3970 add_pin_to_irq_cpu(cfg, cpu, ioapic, pin);
3971 }
3972
3973 setup_IO_APIC_irq(ioapic, pin, irq, desc, triggering, polarity);
3974
3975 return 0;
3976 }
3977
3978
3979 int acpi_get_override_irq(int bus_irq, int *trigger, int *polarity)
3980 {
3981 int i;
3982
3983 if (skip_ioapic_setup)
3984 return -1;
3985
3986 for (i = 0; i < mp_irq_entries; i++)
3987 if (mp_irqs[i].mp_irqtype == mp_INT &&
3988 mp_irqs[i].mp_srcbusirq == bus_irq)
3989 break;
3990 if (i >= mp_irq_entries)
3991 return -1;
3992
3993 *trigger = irq_trigger(i);
3994 *polarity = irq_polarity(i);
3995 return 0;
3996 }
3997
3998 #endif /* CONFIG_ACPI */
3999
4000 /*
4001 * This function currently is only a helper for the i386 smp boot process where
4002 * we need to reprogram the ioredtbls to cater for the cpus which have come online
4003 * so mask in all cases should simply be TARGET_CPUS
4004 */
4005 #ifdef CONFIG_SMP
4006 void __init setup_ioapic_dest(void)
4007 {
4008 int pin, ioapic, irq, irq_entry;
4009 struct irq_desc *desc;
4010 struct irq_cfg *cfg;
4011 const struct cpumask *mask;
4012
4013 if (skip_ioapic_setup == 1)
4014 return;
4015
4016 for (ioapic = 0; ioapic < nr_ioapics; ioapic++) {
4017 for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
4018 irq_entry = find_irq_entry(ioapic, pin, mp_INT);
4019 if (irq_entry == -1)
4020 continue;
4021 irq = pin_2_irq(irq_entry, ioapic, pin);
4022
4023 /* setup_IO_APIC_irqs could fail to get vector for some device
4024 * when you have too many devices, because at that time only boot
4025 * cpu is online.
4026 */
4027 desc = irq_to_desc(irq);
4028 cfg = desc->chip_data;
4029 if (!cfg->vector) {
4030 setup_IO_APIC_irq(ioapic, pin, irq, desc,
4031 irq_trigger(irq_entry),
4032 irq_polarity(irq_entry));
4033 continue;
4034
4035 }
4036
4037 /*
4038 * Honour affinities which have been set in early boot
4039 */
4040 if (desc->status &
4041 (IRQ_NO_BALANCING | IRQ_AFFINITY_SET))
4042 mask = desc->affinity;
4043 else
4044 mask = TARGET_CPUS;
4045
4046 #ifdef CONFIG_INTR_REMAP
4047 if (intr_remapping_enabled)
4048 set_ir_ioapic_affinity_irq_desc(desc, mask);
4049 else
4050 #endif
4051 set_ioapic_affinity_irq_desc(desc, mask);
4052 }
4053
4054 }
4055 }
4056 #endif
4057
4058 #define IOAPIC_RESOURCE_NAME_SIZE 11
4059
4060 static struct resource *ioapic_resources;
4061
4062 static struct resource * __init ioapic_setup_resources(void)
4063 {
4064 unsigned long n;
4065 struct resource *res;
4066 char *mem;
4067 int i;
4068
4069 if (nr_ioapics <= 0)
4070 return NULL;
4071
4072 n = IOAPIC_RESOURCE_NAME_SIZE + sizeof(struct resource);
4073 n *= nr_ioapics;
4074
4075 mem = alloc_bootmem(n);
4076 res = (void *)mem;
4077
4078 if (mem != NULL) {
4079 mem += sizeof(struct resource) * nr_ioapics;
4080
4081 for (i = 0; i < nr_ioapics; i++) {
4082 res[i].name = mem;
4083 res[i].flags = IORESOURCE_MEM | IORESOURCE_BUSY;
4084 sprintf(mem, "IOAPIC %u", i);
4085 mem += IOAPIC_RESOURCE_NAME_SIZE;
4086 }
4087 }
4088
4089 ioapic_resources = res;
4090
4091 return res;
4092 }
4093
4094 void __init ioapic_init_mappings(void)
4095 {
4096 unsigned long ioapic_phys, idx = FIX_IO_APIC_BASE_0;
4097 struct resource *ioapic_res;
4098 int i;
4099
4100 ioapic_res = ioapic_setup_resources();
4101 for (i = 0; i < nr_ioapics; i++) {
4102 if (smp_found_config) {
4103 ioapic_phys = mp_ioapics[i].mp_apicaddr;
4104 #ifdef CONFIG_X86_32
4105 if (!ioapic_phys) {
4106 printk(KERN_ERR
4107 "WARNING: bogus zero IO-APIC "
4108 "address found in MPTABLE, "
4109 "disabling IO/APIC support!\n");
4110 smp_found_config = 0;
4111 skip_ioapic_setup = 1;
4112 goto fake_ioapic_page;
4113 }
4114 #endif
4115 } else {
4116 #ifdef CONFIG_X86_32
4117 fake_ioapic_page:
4118 #endif
4119 ioapic_phys = (unsigned long)
4120 alloc_bootmem_pages(PAGE_SIZE);
4121 ioapic_phys = __pa(ioapic_phys);
4122 }
4123 set_fixmap_nocache(idx, ioapic_phys);
4124 apic_printk(APIC_VERBOSE,
4125 "mapped IOAPIC to %08lx (%08lx)\n",
4126 __fix_to_virt(idx), ioapic_phys);
4127 idx++;
4128
4129 if (ioapic_res != NULL) {
4130 ioapic_res->start = ioapic_phys;
4131 ioapic_res->end = ioapic_phys + (4 * 1024) - 1;
4132 ioapic_res++;
4133 }
4134 }
4135 }
4136
4137 static int __init ioapic_insert_resources(void)
4138 {
4139 int i;
4140 struct resource *r = ioapic_resources;
4141
4142 if (!r) {
4143 printk(KERN_ERR
4144 "IO APIC resources could be not be allocated.\n");
4145 return -1;
4146 }
4147
4148 for (i = 0; i < nr_ioapics; i++) {
4149 insert_resource(&iomem_resource, r);
4150 r++;
4151 }
4152
4153 return 0;
4154 }
4155
4156 /* Insert the IO APIC resources after PCI initialization has occured to handle
4157 * IO APICS that are mapped in on a BAR in PCI space. */
4158 late_initcall(ioapic_insert_resources);
Linux kernel - Deliverables
This page took 0.113164 seconds and 4 git commands to generate.