]> pilppa.org Git - linux-2.6-omap-h63xx.git/blob - arch/x86_64/kernel/io_apic.c
Merge branch 'devel' of master.kernel.org:/home/rmk/linux-2.6-arm
[linux-2.6-omap-h63xx.git] / arch / x86_64 / 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/acpi.h>
31 #include <linux/sysdev.h>
32 #include <linux/msi.h>
33 #include <linux/htirq.h>
34 #ifdef CONFIG_ACPI
35 #include <acpi/acpi_bus.h>
36 #endif
37
38 #include <asm/idle.h>
39 #include <asm/io.h>
40 #include <asm/smp.h>
41 #include <asm/desc.h>
42 #include <asm/proto.h>
43 #include <asm/mach_apic.h>
44 #include <asm/acpi.h>
45 #include <asm/dma.h>
46 #include <asm/nmi.h>
47 #include <asm/msidef.h>
48 #include <asm/hypertransport.h>
49
50 struct irq_cfg {
51         cpumask_t domain;
52         cpumask_t old_domain;
53         unsigned move_cleanup_count;
54         u8 vector;
55         u8 move_in_progress : 1;
56 };
57
58 /* irq_cfg is indexed by the sum of all RTEs in all I/O APICs. */
59 struct irq_cfg irq_cfg[NR_IRQS] __read_mostly = {
60         [0]  = { .domain = CPU_MASK_ALL, .vector = IRQ0_VECTOR,  },
61         [1]  = { .domain = CPU_MASK_ALL, .vector = IRQ1_VECTOR,  },
62         [2]  = { .domain = CPU_MASK_ALL, .vector = IRQ2_VECTOR,  },
63         [3]  = { .domain = CPU_MASK_ALL, .vector = IRQ3_VECTOR,  },
64         [4]  = { .domain = CPU_MASK_ALL, .vector = IRQ4_VECTOR,  },
65         [5]  = { .domain = CPU_MASK_ALL, .vector = IRQ5_VECTOR,  },
66         [6]  = { .domain = CPU_MASK_ALL, .vector = IRQ6_VECTOR,  },
67         [7]  = { .domain = CPU_MASK_ALL, .vector = IRQ7_VECTOR,  },
68         [8]  = { .domain = CPU_MASK_ALL, .vector = IRQ8_VECTOR,  },
69         [9]  = { .domain = CPU_MASK_ALL, .vector = IRQ9_VECTOR,  },
70         [10] = { .domain = CPU_MASK_ALL, .vector = IRQ10_VECTOR, },
71         [11] = { .domain = CPU_MASK_ALL, .vector = IRQ11_VECTOR, },
72         [12] = { .domain = CPU_MASK_ALL, .vector = IRQ12_VECTOR, },
73         [13] = { .domain = CPU_MASK_ALL, .vector = IRQ13_VECTOR, },
74         [14] = { .domain = CPU_MASK_ALL, .vector = IRQ14_VECTOR, },
75         [15] = { .domain = CPU_MASK_ALL, .vector = IRQ15_VECTOR, },
76 };
77
78 static int assign_irq_vector(int irq, cpumask_t mask);
79
80 #define __apicdebuginit  __init
81
82 int sis_apic_bug; /* not actually supported, dummy for compile */
83
84 static int no_timer_check;
85
86 static int disable_timer_pin_1 __initdata;
87
88 int timer_over_8254 __initdata = 1;
89
90 /* Where if anywhere is the i8259 connect in external int mode */
91 static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
92
93 static DEFINE_SPINLOCK(ioapic_lock);
94 DEFINE_SPINLOCK(vector_lock);
95
96 /*
97  * # of IRQ routing registers
98  */
99 int nr_ioapic_registers[MAX_IO_APICS];
100
101 /*
102  * Rough estimation of how many shared IRQs there are, can
103  * be changed anytime.
104  */
105 #define MAX_PLUS_SHARED_IRQS NR_IRQS
106 #define PIN_MAP_SIZE (MAX_PLUS_SHARED_IRQS + NR_IRQS)
107
108 /*
109  * This is performance-critical, we want to do it O(1)
110  *
111  * the indexing order of this array favors 1:1 mappings
112  * between pins and IRQs.
113  */
114
115 static struct irq_pin_list {
116         short apic, pin, next;
117 } irq_2_pin[PIN_MAP_SIZE];
118
119 struct io_apic {
120         unsigned int index;
121         unsigned int unused[3];
122         unsigned int data;
123 };
124
125 static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
126 {
127         return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
128                 + (mp_ioapics[idx].mpc_apicaddr & ~PAGE_MASK);
129 }
130
131 static inline unsigned int io_apic_read(unsigned int apic, unsigned int reg)
132 {
133         struct io_apic __iomem *io_apic = io_apic_base(apic);
134         writel(reg, &io_apic->index);
135         return readl(&io_apic->data);
136 }
137
138 static inline void io_apic_write(unsigned int apic, unsigned int reg, unsigned int value)
139 {
140         struct io_apic __iomem *io_apic = io_apic_base(apic);
141         writel(reg, &io_apic->index);
142         writel(value, &io_apic->data);
143 }
144
145 /*
146  * Re-write a value: to be used for read-modify-write
147  * cycles where the read already set up the index register.
148  */
149 static inline void io_apic_modify(unsigned int apic, unsigned int value)
150 {
151         struct io_apic __iomem *io_apic = io_apic_base(apic);
152         writel(value, &io_apic->data);
153 }
154
155 static int io_apic_level_ack_pending(unsigned int irq)
156 {
157         struct irq_pin_list *entry;
158         unsigned long flags;
159         int pending = 0;
160
161         spin_lock_irqsave(&ioapic_lock, flags);
162         entry = irq_2_pin + irq;
163         for (;;) {
164                 unsigned int reg;
165                 int pin;
166
167                 pin = entry->pin;
168                 if (pin == -1)
169                         break;
170                 reg = io_apic_read(entry->apic, 0x10 + pin*2);
171                 /* Is the remote IRR bit set? */
172                 pending |= (reg >> 14) & 1;
173                 if (!entry->next)
174                         break;
175                 entry = irq_2_pin + entry->next;
176         }
177         spin_unlock_irqrestore(&ioapic_lock, flags);
178         return pending;
179 }
180
181 /*
182  * Synchronize the IO-APIC and the CPU by doing
183  * a dummy read from the IO-APIC
184  */
185 static inline void io_apic_sync(unsigned int apic)
186 {
187         struct io_apic __iomem *io_apic = io_apic_base(apic);
188         readl(&io_apic->data);
189 }
190
191 #define __DO_ACTION(R, ACTION, FINAL)                                   \
192                                                                         \
193 {                                                                       \
194         int pin;                                                        \
195         struct irq_pin_list *entry = irq_2_pin + irq;                   \
196                                                                         \
197         BUG_ON(irq >= NR_IRQS);                                         \
198         for (;;) {                                                      \
199                 unsigned int reg;                                       \
200                 pin = entry->pin;                                       \
201                 if (pin == -1)                                          \
202                         break;                                          \
203                 reg = io_apic_read(entry->apic, 0x10 + R + pin*2);      \
204                 reg ACTION;                                             \
205                 io_apic_modify(entry->apic, reg);                       \
206                 FINAL;                                                  \
207                 if (!entry->next)                                       \
208                         break;                                          \
209                 entry = irq_2_pin + entry->next;                        \
210         }                                                               \
211 }
212
213 union entry_union {
214         struct { u32 w1, w2; };
215         struct IO_APIC_route_entry entry;
216 };
217
218 static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
219 {
220         union entry_union eu;
221         unsigned long flags;
222         spin_lock_irqsave(&ioapic_lock, flags);
223         eu.w1 = io_apic_read(apic, 0x10 + 2 * pin);
224         eu.w2 = io_apic_read(apic, 0x11 + 2 * pin);
225         spin_unlock_irqrestore(&ioapic_lock, flags);
226         return eu.entry;
227 }
228
229 /*
230  * When we write a new IO APIC routing entry, we need to write the high
231  * word first! If the mask bit in the low word is clear, we will enable
232  * the interrupt, and we need to make sure the entry is fully populated
233  * before that happens.
234  */
235 static void
236 __ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
237 {
238         union entry_union eu;
239         eu.entry = e;
240         io_apic_write(apic, 0x11 + 2*pin, eu.w2);
241         io_apic_write(apic, 0x10 + 2*pin, eu.w1);
242 }
243
244 static void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
245 {
246         unsigned long flags;
247         spin_lock_irqsave(&ioapic_lock, flags);
248         __ioapic_write_entry(apic, pin, e);
249         spin_unlock_irqrestore(&ioapic_lock, flags);
250 }
251
252 /*
253  * When we mask an IO APIC routing entry, we need to write the low
254  * word first, in order to set the mask bit before we change the
255  * high bits!
256  */
257 static void ioapic_mask_entry(int apic, int pin)
258 {
259         unsigned long flags;
260         union entry_union eu = { .entry.mask = 1 };
261
262         spin_lock_irqsave(&ioapic_lock, flags);
263         io_apic_write(apic, 0x10 + 2*pin, eu.w1);
264         io_apic_write(apic, 0x11 + 2*pin, eu.w2);
265         spin_unlock_irqrestore(&ioapic_lock, flags);
266 }
267
268 #ifdef CONFIG_SMP
269 static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, u8 vector)
270 {
271         int apic, pin;
272         struct irq_pin_list *entry = irq_2_pin + irq;
273
274         BUG_ON(irq >= NR_IRQS);
275         for (;;) {
276                 unsigned int reg;
277                 apic = entry->apic;
278                 pin = entry->pin;
279                 if (pin == -1)
280                         break;
281                 io_apic_write(apic, 0x11 + pin*2, dest);
282                 reg = io_apic_read(apic, 0x10 + pin*2);
283                 reg &= ~0x000000ff;
284                 reg |= vector;
285                 io_apic_modify(apic, reg);
286                 if (!entry->next)
287                         break;
288                 entry = irq_2_pin + entry->next;
289         }
290 }
291
292 static void set_ioapic_affinity_irq(unsigned int irq, cpumask_t mask)
293 {
294         struct irq_cfg *cfg = irq_cfg + irq;
295         unsigned long flags;
296         unsigned int dest;
297         cpumask_t tmp;
298
299         cpus_and(tmp, mask, cpu_online_map);
300         if (cpus_empty(tmp))
301                 return;
302
303         if (assign_irq_vector(irq, mask))
304                 return;
305
306         cpus_and(tmp, cfg->domain, mask);
307         dest = cpu_mask_to_apicid(tmp);
308
309         /*
310          * Only the high 8 bits are valid.
311          */
312         dest = SET_APIC_LOGICAL_ID(dest);
313
314         spin_lock_irqsave(&ioapic_lock, flags);
315         __target_IO_APIC_irq(irq, dest, cfg->vector);
316         irq_desc[irq].affinity = mask;
317         spin_unlock_irqrestore(&ioapic_lock, flags);
318 }
319 #endif
320
321 /*
322  * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
323  * shared ISA-space IRQs, so we have to support them. We are super
324  * fast in the common case, and fast for shared ISA-space IRQs.
325  */
326 static void add_pin_to_irq(unsigned int irq, int apic, int pin)
327 {
328         static int first_free_entry = NR_IRQS;
329         struct irq_pin_list *entry = irq_2_pin + irq;
330
331         BUG_ON(irq >= NR_IRQS);
332         while (entry->next)
333                 entry = irq_2_pin + entry->next;
334
335         if (entry->pin != -1) {
336                 entry->next = first_free_entry;
337                 entry = irq_2_pin + entry->next;
338                 if (++first_free_entry >= PIN_MAP_SIZE)
339                         panic("io_apic.c: ran out of irq_2_pin entries!");
340         }
341         entry->apic = apic;
342         entry->pin = pin;
343 }
344
345
346 #define DO_ACTION(name,R,ACTION, FINAL)                                 \
347                                                                         \
348         static void name##_IO_APIC_irq (unsigned int irq)               \
349         __DO_ACTION(R, ACTION, FINAL)
350
351 DO_ACTION( __mask,             0, |= 0x00010000, io_apic_sync(entry->apic) )
352                                                 /* mask = 1 */
353 DO_ACTION( __unmask,           0, &= 0xfffeffff, )
354                                                 /* mask = 0 */
355
356 static void mask_IO_APIC_irq (unsigned int irq)
357 {
358         unsigned long flags;
359
360         spin_lock_irqsave(&ioapic_lock, flags);
361         __mask_IO_APIC_irq(irq);
362         spin_unlock_irqrestore(&ioapic_lock, flags);
363 }
364
365 static void unmask_IO_APIC_irq (unsigned int irq)
366 {
367         unsigned long flags;
368
369         spin_lock_irqsave(&ioapic_lock, flags);
370         __unmask_IO_APIC_irq(irq);
371         spin_unlock_irqrestore(&ioapic_lock, flags);
372 }
373
374 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
375 {
376         struct IO_APIC_route_entry entry;
377
378         /* Check delivery_mode to be sure we're not clearing an SMI pin */
379         entry = ioapic_read_entry(apic, pin);
380         if (entry.delivery_mode == dest_SMI)
381                 return;
382         /*
383          * Disable it in the IO-APIC irq-routing table:
384          */
385         ioapic_mask_entry(apic, pin);
386 }
387
388 static void clear_IO_APIC (void)
389 {
390         int apic, pin;
391
392         for (apic = 0; apic < nr_ioapics; apic++)
393                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
394                         clear_IO_APIC_pin(apic, pin);
395 }
396
397 int skip_ioapic_setup;
398 int ioapic_force;
399
400 /* dummy parsing: see setup.c */
401
402 static int __init disable_ioapic_setup(char *str)
403 {
404         skip_ioapic_setup = 1;
405         return 0;
406 }
407 early_param("noapic", disable_ioapic_setup);
408
409 /* Actually the next is obsolete, but keep it for paranoid reasons -AK */
410 static int __init disable_timer_pin_setup(char *arg)
411 {
412         disable_timer_pin_1 = 1;
413         return 1;
414 }
415 __setup("disable_timer_pin_1", disable_timer_pin_setup);
416
417 static int __init setup_disable_8254_timer(char *s)
418 {
419         timer_over_8254 = -1;
420         return 1;
421 }
422 static int __init setup_enable_8254_timer(char *s)
423 {
424         timer_over_8254 = 2;
425         return 1;
426 }
427
428 __setup("disable_8254_timer", setup_disable_8254_timer);
429 __setup("enable_8254_timer", setup_enable_8254_timer);
430
431
432 /*
433  * Find the IRQ entry number of a certain pin.
434  */
435 static int find_irq_entry(int apic, int pin, int type)
436 {
437         int i;
438
439         for (i = 0; i < mp_irq_entries; i++)
440                 if (mp_irqs[i].mpc_irqtype == type &&
441                     (mp_irqs[i].mpc_dstapic == mp_ioapics[apic].mpc_apicid ||
442                      mp_irqs[i].mpc_dstapic == MP_APIC_ALL) &&
443                     mp_irqs[i].mpc_dstirq == pin)
444                         return i;
445
446         return -1;
447 }
448
449 /*
450  * Find the pin to which IRQ[irq] (ISA) is connected
451  */
452 static int __init find_isa_irq_pin(int irq, int type)
453 {
454         int i;
455
456         for (i = 0; i < mp_irq_entries; i++) {
457                 int lbus = mp_irqs[i].mpc_srcbus;
458
459                 if (test_bit(lbus, mp_bus_not_pci) &&
460                     (mp_irqs[i].mpc_irqtype == type) &&
461                     (mp_irqs[i].mpc_srcbusirq == irq))
462
463                         return mp_irqs[i].mpc_dstirq;
464         }
465         return -1;
466 }
467
468 static int __init find_isa_irq_apic(int irq, int type)
469 {
470         int i;
471
472         for (i = 0; i < mp_irq_entries; i++) {
473                 int lbus = mp_irqs[i].mpc_srcbus;
474
475                 if (test_bit(lbus, mp_bus_not_pci) &&
476                     (mp_irqs[i].mpc_irqtype == type) &&
477                     (mp_irqs[i].mpc_srcbusirq == irq))
478                         break;
479         }
480         if (i < mp_irq_entries) {
481                 int apic;
482                 for(apic = 0; apic < nr_ioapics; apic++) {
483                         if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic)
484                                 return apic;
485                 }
486         }
487
488         return -1;
489 }
490
491 /*
492  * Find a specific PCI IRQ entry.
493  * Not an __init, possibly needed by modules
494  */
495 static int pin_2_irq(int idx, int apic, int pin);
496
497 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
498 {
499         int apic, i, best_guess = -1;
500
501         apic_printk(APIC_DEBUG, "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
502                 bus, slot, pin);
503         if (mp_bus_id_to_pci_bus[bus] == -1) {
504                 apic_printk(APIC_VERBOSE, "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
505                 return -1;
506         }
507         for (i = 0; i < mp_irq_entries; i++) {
508                 int lbus = mp_irqs[i].mpc_srcbus;
509
510                 for (apic = 0; apic < nr_ioapics; apic++)
511                         if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic ||
512                             mp_irqs[i].mpc_dstapic == MP_APIC_ALL)
513                                 break;
514
515                 if (!test_bit(lbus, mp_bus_not_pci) &&
516                     !mp_irqs[i].mpc_irqtype &&
517                     (bus == lbus) &&
518                     (slot == ((mp_irqs[i].mpc_srcbusirq >> 2) & 0x1f))) {
519                         int irq = pin_2_irq(i,apic,mp_irqs[i].mpc_dstirq);
520
521                         if (!(apic || IO_APIC_IRQ(irq)))
522                                 continue;
523
524                         if (pin == (mp_irqs[i].mpc_srcbusirq & 3))
525                                 return irq;
526                         /*
527                          * Use the first all-but-pin matching entry as a
528                          * best-guess fuzzy result for broken mptables.
529                          */
530                         if (best_guess < 0)
531                                 best_guess = irq;
532                 }
533         }
534         BUG_ON(best_guess >= NR_IRQS);
535         return best_guess;
536 }
537
538 /* ISA interrupts are always polarity zero edge triggered,
539  * when listed as conforming in the MP table. */
540
541 #define default_ISA_trigger(idx)        (0)
542 #define default_ISA_polarity(idx)       (0)
543
544 /* PCI interrupts are always polarity one level triggered,
545  * when listed as conforming in the MP table. */
546
547 #define default_PCI_trigger(idx)        (1)
548 #define default_PCI_polarity(idx)       (1)
549
550 static int __init MPBIOS_polarity(int idx)
551 {
552         int bus = mp_irqs[idx].mpc_srcbus;
553         int polarity;
554
555         /*
556          * Determine IRQ line polarity (high active or low active):
557          */
558         switch (mp_irqs[idx].mpc_irqflag & 3)
559         {
560                 case 0: /* conforms, ie. bus-type dependent polarity */
561                         if (test_bit(bus, mp_bus_not_pci))
562                                 polarity = default_ISA_polarity(idx);
563                         else
564                                 polarity = default_PCI_polarity(idx);
565                         break;
566                 case 1: /* high active */
567                 {
568                         polarity = 0;
569                         break;
570                 }
571                 case 2: /* reserved */
572                 {
573                         printk(KERN_WARNING "broken BIOS!!\n");
574                         polarity = 1;
575                         break;
576                 }
577                 case 3: /* low active */
578                 {
579                         polarity = 1;
580                         break;
581                 }
582                 default: /* invalid */
583                 {
584                         printk(KERN_WARNING "broken BIOS!!\n");
585                         polarity = 1;
586                         break;
587                 }
588         }
589         return polarity;
590 }
591
592 static int MPBIOS_trigger(int idx)
593 {
594         int bus = mp_irqs[idx].mpc_srcbus;
595         int trigger;
596
597         /*
598          * Determine IRQ trigger mode (edge or level sensitive):
599          */
600         switch ((mp_irqs[idx].mpc_irqflag>>2) & 3)
601         {
602                 case 0: /* conforms, ie. bus-type dependent */
603                         if (test_bit(bus, mp_bus_not_pci))
604                                 trigger = default_ISA_trigger(idx);
605                         else
606                                 trigger = default_PCI_trigger(idx);
607                         break;
608                 case 1: /* edge */
609                 {
610                         trigger = 0;
611                         break;
612                 }
613                 case 2: /* reserved */
614                 {
615                         printk(KERN_WARNING "broken BIOS!!\n");
616                         trigger = 1;
617                         break;
618                 }
619                 case 3: /* level */
620                 {
621                         trigger = 1;
622                         break;
623                 }
624                 default: /* invalid */
625                 {
626                         printk(KERN_WARNING "broken BIOS!!\n");
627                         trigger = 0;
628                         break;
629                 }
630         }
631         return trigger;
632 }
633
634 static inline int irq_polarity(int idx)
635 {
636         return MPBIOS_polarity(idx);
637 }
638
639 static inline int irq_trigger(int idx)
640 {
641         return MPBIOS_trigger(idx);
642 }
643
644 static int pin_2_irq(int idx, int apic, int pin)
645 {
646         int irq, i;
647         int bus = mp_irqs[idx].mpc_srcbus;
648
649         /*
650          * Debugging check, we are in big trouble if this message pops up!
651          */
652         if (mp_irqs[idx].mpc_dstirq != pin)
653                 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
654
655         if (test_bit(bus, mp_bus_not_pci)) {
656                 irq = mp_irqs[idx].mpc_srcbusirq;
657         } else {
658                 /*
659                  * PCI IRQs are mapped in order
660                  */
661                 i = irq = 0;
662                 while (i < apic)
663                         irq += nr_ioapic_registers[i++];
664                 irq += pin;
665         }
666         BUG_ON(irq >= NR_IRQS);
667         return irq;
668 }
669
670 static int __assign_irq_vector(int irq, cpumask_t mask)
671 {
672         /*
673          * NOTE! The local APIC isn't very good at handling
674          * multiple interrupts at the same interrupt level.
675          * As the interrupt level is determined by taking the
676          * vector number and shifting that right by 4, we
677          * want to spread these out a bit so that they don't
678          * all fall in the same interrupt level.
679          *
680          * Also, we've got to be careful not to trash gate
681          * 0x80, because int 0x80 is hm, kind of importantish. ;)
682          */
683         static int current_vector = FIRST_DEVICE_VECTOR, current_offset = 0;
684         unsigned int old_vector;
685         int cpu;
686         struct irq_cfg *cfg;
687
688         BUG_ON((unsigned)irq >= NR_IRQS);
689         cfg = &irq_cfg[irq];
690
691         /* Only try and allocate irqs on cpus that are present */
692         cpus_and(mask, mask, cpu_online_map);
693
694         if ((cfg->move_in_progress) || cfg->move_cleanup_count)
695                 return -EBUSY;
696
697         old_vector = cfg->vector;
698         if (old_vector) {
699                 cpumask_t tmp;
700                 cpus_and(tmp, cfg->domain, mask);
701                 if (!cpus_empty(tmp))
702                         return 0;
703         }
704
705         for_each_cpu_mask(cpu, mask) {
706                 cpumask_t domain, new_mask;
707                 int new_cpu;
708                 int vector, offset;
709
710                 domain = vector_allocation_domain(cpu);
711                 cpus_and(new_mask, domain, cpu_online_map);
712
713                 vector = current_vector;
714                 offset = current_offset;
715 next:
716                 vector += 8;
717                 if (vector >= FIRST_SYSTEM_VECTOR) {
718                         /* If we run out of vectors on large boxen, must share them. */
719                         offset = (offset + 1) % 8;
720                         vector = FIRST_DEVICE_VECTOR + offset;
721                 }
722                 if (unlikely(current_vector == vector))
723                         continue;
724                 if (vector == IA32_SYSCALL_VECTOR)
725                         goto next;
726                 for_each_cpu_mask(new_cpu, new_mask)
727                         if (per_cpu(vector_irq, new_cpu)[vector] != -1)
728                                 goto next;
729                 /* Found one! */
730                 current_vector = vector;
731                 current_offset = offset;
732                 if (old_vector) {
733                         cfg->move_in_progress = 1;
734                         cfg->old_domain = cfg->domain;
735                 }
736                 for_each_cpu_mask(new_cpu, new_mask)
737                         per_cpu(vector_irq, new_cpu)[vector] = irq;
738                 cfg->vector = vector;
739                 cfg->domain = domain;
740                 return 0;
741         }
742         return -ENOSPC;
743 }
744
745 static int assign_irq_vector(int irq, cpumask_t mask)
746 {
747         int err;
748         unsigned long flags;
749
750         spin_lock_irqsave(&vector_lock, flags);
751         err = __assign_irq_vector(irq, mask);
752         spin_unlock_irqrestore(&vector_lock, flags);
753         return err;
754 }
755
756 static void __clear_irq_vector(int irq)
757 {
758         struct irq_cfg *cfg;
759         cpumask_t mask;
760         int cpu, vector;
761
762         BUG_ON((unsigned)irq >= NR_IRQS);
763         cfg = &irq_cfg[irq];
764         BUG_ON(!cfg->vector);
765
766         vector = cfg->vector;
767         cpus_and(mask, cfg->domain, cpu_online_map);
768         for_each_cpu_mask(cpu, mask)
769                 per_cpu(vector_irq, cpu)[vector] = -1;
770
771         cfg->vector = 0;
772         cfg->domain = CPU_MASK_NONE;
773 }
774
775 void __setup_vector_irq(int cpu)
776 {
777         /* Initialize vector_irq on a new cpu */
778         /* This function must be called with vector_lock held */
779         int irq, vector;
780
781         /* Mark the inuse vectors */
782         for (irq = 0; irq < NR_IRQS; ++irq) {
783                 if (!cpu_isset(cpu, irq_cfg[irq].domain))
784                         continue;
785                 vector = irq_cfg[irq].vector;
786                 per_cpu(vector_irq, cpu)[vector] = irq;
787         }
788         /* Mark the free vectors */
789         for (vector = 0; vector < NR_VECTORS; ++vector) {
790                 irq = per_cpu(vector_irq, cpu)[vector];
791                 if (irq < 0)
792                         continue;
793                 if (!cpu_isset(cpu, irq_cfg[irq].domain))
794                         per_cpu(vector_irq, cpu)[vector] = -1;
795         }
796 }
797
798
799 static struct irq_chip ioapic_chip;
800
801 static void ioapic_register_intr(int irq, unsigned long trigger)
802 {
803         if (trigger)
804                 set_irq_chip_and_handler_name(irq, &ioapic_chip,
805                                               handle_fasteoi_irq, "fasteoi");
806         else
807                 set_irq_chip_and_handler_name(irq, &ioapic_chip,
808                                               handle_edge_irq, "edge");
809 }
810
811 static void setup_IO_APIC_irq(int apic, int pin, unsigned int irq,
812                               int trigger, int polarity)
813 {
814         struct irq_cfg *cfg = irq_cfg + irq;
815         struct IO_APIC_route_entry entry;
816         cpumask_t mask;
817
818         if (!IO_APIC_IRQ(irq))
819                 return;
820
821         mask = TARGET_CPUS;
822         if (assign_irq_vector(irq, mask))
823                 return;
824
825         cpus_and(mask, cfg->domain, mask);
826
827         apic_printk(APIC_VERBOSE,KERN_DEBUG
828                     "IOAPIC[%d]: Set routing entry (%d-%d -> 0x%x -> "
829                     "IRQ %d Mode:%i Active:%i)\n",
830                     apic, mp_ioapics[apic].mpc_apicid, pin, cfg->vector,
831                     irq, trigger, polarity);
832
833         /*
834          * add it to the IO-APIC irq-routing table:
835          */
836         memset(&entry,0,sizeof(entry));
837
838         entry.delivery_mode = INT_DELIVERY_MODE;
839         entry.dest_mode = INT_DEST_MODE;
840         entry.dest = cpu_mask_to_apicid(mask);
841         entry.mask = 0;                         /* enable IRQ */
842         entry.trigger = trigger;
843         entry.polarity = polarity;
844         entry.vector = cfg->vector;
845
846         /* Mask level triggered irqs.
847          * Use IRQ_DELAYED_DISABLE for edge triggered irqs.
848          */
849         if (trigger)
850                 entry.mask = 1;
851
852         ioapic_register_intr(irq, trigger);
853         if (irq < 16)
854                 disable_8259A_irq(irq);
855
856         ioapic_write_entry(apic, pin, entry);
857 }
858
859 static void __init setup_IO_APIC_irqs(void)
860 {
861         int apic, pin, idx, irq, first_notcon = 1;
862
863         apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
864
865         for (apic = 0; apic < nr_ioapics; apic++) {
866         for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
867
868                 idx = find_irq_entry(apic,pin,mp_INT);
869                 if (idx == -1) {
870                         if (first_notcon) {
871                                 apic_printk(APIC_VERBOSE, KERN_DEBUG " IO-APIC (apicid-pin) %d-%d", mp_ioapics[apic].mpc_apicid, pin);
872                                 first_notcon = 0;
873                         } else
874                                 apic_printk(APIC_VERBOSE, ", %d-%d", mp_ioapics[apic].mpc_apicid, pin);
875                         continue;
876                 }
877
878                 irq = pin_2_irq(idx, apic, pin);
879                 add_pin_to_irq(irq, apic, pin);
880
881                 setup_IO_APIC_irq(apic, pin, irq,
882                                   irq_trigger(idx), irq_polarity(idx));
883         }
884         }
885
886         if (!first_notcon)
887                 apic_printk(APIC_VERBOSE," not connected.\n");
888 }
889
890 /*
891  * Set up the 8259A-master output pin as broadcast to all
892  * CPUs.
893  */
894 static void __init setup_ExtINT_IRQ0_pin(unsigned int apic, unsigned int pin, int vector)
895 {
896         struct IO_APIC_route_entry entry;
897         unsigned long flags;
898
899         memset(&entry,0,sizeof(entry));
900
901         disable_8259A_irq(0);
902
903         /* mask LVT0 */
904         apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
905
906         /*
907          * We use logical delivery to get the timer IRQ
908          * to the first CPU.
909          */
910         entry.dest_mode = INT_DEST_MODE;
911         entry.mask = 0;                                 /* unmask IRQ now */
912         entry.dest = cpu_mask_to_apicid(TARGET_CPUS);
913         entry.delivery_mode = INT_DELIVERY_MODE;
914         entry.polarity = 0;
915         entry.trigger = 0;
916         entry.vector = vector;
917
918         /*
919          * The timer IRQ doesn't have to know that behind the
920          * scene we have a 8259A-master in AEOI mode ...
921          */
922         set_irq_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq, "edge");
923
924         /*
925          * Add it to the IO-APIC irq-routing table:
926          */
927         spin_lock_irqsave(&ioapic_lock, flags);
928         io_apic_write(apic, 0x11+2*pin, *(((int *)&entry)+1));
929         io_apic_write(apic, 0x10+2*pin, *(((int *)&entry)+0));
930         spin_unlock_irqrestore(&ioapic_lock, flags);
931
932         enable_8259A_irq(0);
933 }
934
935 void __apicdebuginit print_IO_APIC(void)
936 {
937         int apic, i;
938         union IO_APIC_reg_00 reg_00;
939         union IO_APIC_reg_01 reg_01;
940         union IO_APIC_reg_02 reg_02;
941         unsigned long flags;
942
943         if (apic_verbosity == APIC_QUIET)
944                 return;
945
946         printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
947         for (i = 0; i < nr_ioapics; i++)
948                 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
949                        mp_ioapics[i].mpc_apicid, nr_ioapic_registers[i]);
950
951         /*
952          * We are a bit conservative about what we expect.  We have to
953          * know about every hardware change ASAP.
954          */
955         printk(KERN_INFO "testing the IO APIC.......................\n");
956
957         for (apic = 0; apic < nr_ioapics; apic++) {
958
959         spin_lock_irqsave(&ioapic_lock, flags);
960         reg_00.raw = io_apic_read(apic, 0);
961         reg_01.raw = io_apic_read(apic, 1);
962         if (reg_01.bits.version >= 0x10)
963                 reg_02.raw = io_apic_read(apic, 2);
964         spin_unlock_irqrestore(&ioapic_lock, flags);
965
966         printk("\n");
967         printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].mpc_apicid);
968         printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
969         printk(KERN_DEBUG ".......    : physical APIC id: %02X\n", reg_00.bits.ID);
970
971         printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)&reg_01);
972         printk(KERN_DEBUG ".......     : max redirection entries: %04X\n", reg_01.bits.entries);
973
974         printk(KERN_DEBUG ".......     : PRQ implemented: %X\n", reg_01.bits.PRQ);
975         printk(KERN_DEBUG ".......     : IO APIC version: %04X\n", reg_01.bits.version);
976
977         if (reg_01.bits.version >= 0x10) {
978                 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
979                 printk(KERN_DEBUG ".......     : arbitration: %02X\n", reg_02.bits.arbitration);
980         }
981
982         printk(KERN_DEBUG ".... IRQ redirection table:\n");
983
984         printk(KERN_DEBUG " NR Dst Mask Trig IRR Pol"
985                           " Stat Dmod Deli Vect:   \n");
986
987         for (i = 0; i <= reg_01.bits.entries; i++) {
988                 struct IO_APIC_route_entry entry;
989
990                 entry = ioapic_read_entry(apic, i);
991
992                 printk(KERN_DEBUG " %02x %03X ",
993                         i,
994                         entry.dest
995                 );
996
997                 printk("%1d    %1d    %1d   %1d   %1d    %1d    %1d    %02X\n",
998                         entry.mask,
999                         entry.trigger,
1000                         entry.irr,
1001                         entry.polarity,
1002                         entry.delivery_status,
1003                         entry.dest_mode,
1004                         entry.delivery_mode,
1005                         entry.vector
1006                 );
1007         }
1008         }
1009         printk(KERN_DEBUG "IRQ to pin mappings:\n");
1010         for (i = 0; i < NR_IRQS; i++) {
1011                 struct irq_pin_list *entry = irq_2_pin + i;
1012                 if (entry->pin < 0)
1013                         continue;
1014                 printk(KERN_DEBUG "IRQ%d ", i);
1015                 for (;;) {
1016                         printk("-> %d:%d", entry->apic, entry->pin);
1017                         if (!entry->next)
1018                                 break;
1019                         entry = irq_2_pin + entry->next;
1020                 }
1021                 printk("\n");
1022         }
1023
1024         printk(KERN_INFO ".................................... done.\n");
1025
1026         return;
1027 }
1028
1029 #if 0
1030
1031 static __apicdebuginit void print_APIC_bitfield (int base)
1032 {
1033         unsigned int v;
1034         int i, j;
1035
1036         if (apic_verbosity == APIC_QUIET)
1037                 return;
1038
1039         printk(KERN_DEBUG "0123456789abcdef0123456789abcdef\n" KERN_DEBUG);
1040         for (i = 0; i < 8; i++) {
1041                 v = apic_read(base + i*0x10);
1042                 for (j = 0; j < 32; j++) {
1043                         if (v & (1<<j))
1044                                 printk("1");
1045                         else
1046                                 printk("0");
1047                 }
1048                 printk("\n");
1049         }
1050 }
1051
1052 void __apicdebuginit print_local_APIC(void * dummy)
1053 {
1054         unsigned int v, ver, maxlvt;
1055
1056         if (apic_verbosity == APIC_QUIET)
1057                 return;
1058
1059         printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1060                 smp_processor_id(), hard_smp_processor_id());
1061         v = apic_read(APIC_ID);
1062         printk(KERN_INFO "... APIC ID:      %08x (%01x)\n", v, GET_APIC_ID(v));
1063         v = apic_read(APIC_LVR);
1064         printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1065         ver = GET_APIC_VERSION(v);
1066         maxlvt = get_maxlvt();
1067
1068         v = apic_read(APIC_TASKPRI);
1069         printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1070
1071         v = apic_read(APIC_ARBPRI);
1072         printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
1073                 v & APIC_ARBPRI_MASK);
1074         v = apic_read(APIC_PROCPRI);
1075         printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1076
1077         v = apic_read(APIC_EOI);
1078         printk(KERN_DEBUG "... APIC EOI: %08x\n", v);
1079         v = apic_read(APIC_RRR);
1080         printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1081         v = apic_read(APIC_LDR);
1082         printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1083         v = apic_read(APIC_DFR);
1084         printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1085         v = apic_read(APIC_SPIV);
1086         printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1087
1088         printk(KERN_DEBUG "... APIC ISR field:\n");
1089         print_APIC_bitfield(APIC_ISR);
1090         printk(KERN_DEBUG "... APIC TMR field:\n");
1091         print_APIC_bitfield(APIC_TMR);
1092         printk(KERN_DEBUG "... APIC IRR field:\n");
1093         print_APIC_bitfield(APIC_IRR);
1094
1095         v = apic_read(APIC_ESR);
1096         printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1097
1098         v = apic_read(APIC_ICR);
1099         printk(KERN_DEBUG "... APIC ICR: %08x\n", v);
1100         v = apic_read(APIC_ICR2);
1101         printk(KERN_DEBUG "... APIC ICR2: %08x\n", v);
1102
1103         v = apic_read(APIC_LVTT);
1104         printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1105
1106         if (maxlvt > 3) {                       /* PC is LVT#4. */
1107                 v = apic_read(APIC_LVTPC);
1108                 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1109         }
1110         v = apic_read(APIC_LVT0);
1111         printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1112         v = apic_read(APIC_LVT1);
1113         printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1114
1115         if (maxlvt > 2) {                       /* ERR is LVT#3. */
1116                 v = apic_read(APIC_LVTERR);
1117                 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1118         }
1119
1120         v = apic_read(APIC_TMICT);
1121         printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1122         v = apic_read(APIC_TMCCT);
1123         printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1124         v = apic_read(APIC_TDCR);
1125         printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1126         printk("\n");
1127 }
1128
1129 void print_all_local_APICs (void)
1130 {
1131         on_each_cpu(print_local_APIC, NULL, 1, 1);
1132 }
1133
1134 void __apicdebuginit print_PIC(void)
1135 {
1136         unsigned int v;
1137         unsigned long flags;
1138
1139         if (apic_verbosity == APIC_QUIET)
1140                 return;
1141
1142         printk(KERN_DEBUG "\nprinting PIC contents\n");
1143
1144         spin_lock_irqsave(&i8259A_lock, flags);
1145
1146         v = inb(0xa1) << 8 | inb(0x21);
1147         printk(KERN_DEBUG "... PIC  IMR: %04x\n", v);
1148
1149         v = inb(0xa0) << 8 | inb(0x20);
1150         printk(KERN_DEBUG "... PIC  IRR: %04x\n", v);
1151
1152         outb(0x0b,0xa0);
1153         outb(0x0b,0x20);
1154         v = inb(0xa0) << 8 | inb(0x20);
1155         outb(0x0a,0xa0);
1156         outb(0x0a,0x20);
1157
1158         spin_unlock_irqrestore(&i8259A_lock, flags);
1159
1160         printk(KERN_DEBUG "... PIC  ISR: %04x\n", v);
1161
1162         v = inb(0x4d1) << 8 | inb(0x4d0);
1163         printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1164 }
1165
1166 #endif  /*  0  */
1167
1168 static void __init enable_IO_APIC(void)
1169 {
1170         union IO_APIC_reg_01 reg_01;
1171         int i8259_apic, i8259_pin;
1172         int i, apic;
1173         unsigned long flags;
1174
1175         for (i = 0; i < PIN_MAP_SIZE; i++) {
1176                 irq_2_pin[i].pin = -1;
1177                 irq_2_pin[i].next = 0;
1178         }
1179
1180         /*
1181          * The number of IO-APIC IRQ registers (== #pins):
1182          */
1183         for (apic = 0; apic < nr_ioapics; apic++) {
1184                 spin_lock_irqsave(&ioapic_lock, flags);
1185                 reg_01.raw = io_apic_read(apic, 1);
1186                 spin_unlock_irqrestore(&ioapic_lock, flags);
1187                 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
1188         }
1189         for(apic = 0; apic < nr_ioapics; apic++) {
1190                 int pin;
1191                 /* See if any of the pins is in ExtINT mode */
1192                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1193                         struct IO_APIC_route_entry entry;
1194                         entry = ioapic_read_entry(apic, pin);
1195
1196                         /* If the interrupt line is enabled and in ExtInt mode
1197                          * I have found the pin where the i8259 is connected.
1198                          */
1199                         if ((entry.mask == 0) && (entry.delivery_mode == dest_ExtINT)) {
1200                                 ioapic_i8259.apic = apic;
1201                                 ioapic_i8259.pin  = pin;
1202                                 goto found_i8259;
1203                         }
1204                 }
1205         }
1206  found_i8259:
1207         /* Look to see what if the MP table has reported the ExtINT */
1208         i8259_pin  = find_isa_irq_pin(0, mp_ExtINT);
1209         i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
1210         /* Trust the MP table if nothing is setup in the hardware */
1211         if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
1212                 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
1213                 ioapic_i8259.pin  = i8259_pin;
1214                 ioapic_i8259.apic = i8259_apic;
1215         }
1216         /* Complain if the MP table and the hardware disagree */
1217         if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
1218                 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
1219         {
1220                 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
1221         }
1222
1223         /*
1224          * Do not trust the IO-APIC being empty at bootup
1225          */
1226         clear_IO_APIC();
1227 }
1228
1229 /*
1230  * Not an __init, needed by the reboot code
1231  */
1232 void disable_IO_APIC(void)
1233 {
1234         /*
1235          * Clear the IO-APIC before rebooting:
1236          */
1237         clear_IO_APIC();
1238
1239         /*
1240          * If the i8259 is routed through an IOAPIC
1241          * Put that IOAPIC in virtual wire mode
1242          * so legacy interrupts can be delivered.
1243          */
1244         if (ioapic_i8259.pin != -1) {
1245                 struct IO_APIC_route_entry entry;
1246
1247                 memset(&entry, 0, sizeof(entry));
1248                 entry.mask            = 0; /* Enabled */
1249                 entry.trigger         = 0; /* Edge */
1250                 entry.irr             = 0;
1251                 entry.polarity        = 0; /* High */
1252                 entry.delivery_status = 0;
1253                 entry.dest_mode       = 0; /* Physical */
1254                 entry.delivery_mode   = dest_ExtINT; /* ExtInt */
1255                 entry.vector          = 0;
1256                 entry.dest          = GET_APIC_ID(apic_read(APIC_ID));
1257
1258                 /*
1259                  * Add it to the IO-APIC irq-routing table:
1260                  */
1261                 ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
1262         }
1263
1264         disconnect_bsp_APIC(ioapic_i8259.pin != -1);
1265 }
1266
1267 /*
1268  * There is a nasty bug in some older SMP boards, their mptable lies
1269  * about the timer IRQ. We do the following to work around the situation:
1270  *
1271  *      - timer IRQ defaults to IO-APIC IRQ
1272  *      - if this function detects that timer IRQs are defunct, then we fall
1273  *        back to ISA timer IRQs
1274  */
1275 static int __init timer_irq_works(void)
1276 {
1277         unsigned long t1 = jiffies;
1278
1279         local_irq_enable();
1280         /* Let ten ticks pass... */
1281         mdelay((10 * 1000) / HZ);
1282
1283         /*
1284          * Expect a few ticks at least, to be sure some possible
1285          * glue logic does not lock up after one or two first
1286          * ticks in a non-ExtINT mode.  Also the local APIC
1287          * might have cached one ExtINT interrupt.  Finally, at
1288          * least one tick may be lost due to delays.
1289          */
1290
1291         /* jiffies wrap? */
1292         if (jiffies - t1 > 4)
1293                 return 1;
1294         return 0;
1295 }
1296
1297 /*
1298  * In the SMP+IOAPIC case it might happen that there are an unspecified
1299  * number of pending IRQ events unhandled. These cases are very rare,
1300  * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
1301  * better to do it this way as thus we do not have to be aware of
1302  * 'pending' interrupts in the IRQ path, except at this point.
1303  */
1304 /*
1305  * Edge triggered needs to resend any interrupt
1306  * that was delayed but this is now handled in the device
1307  * independent code.
1308  */
1309
1310 /*
1311  * Starting up a edge-triggered IO-APIC interrupt is
1312  * nasty - we need to make sure that we get the edge.
1313  * If it is already asserted for some reason, we need
1314  * return 1 to indicate that is was pending.
1315  *
1316  * This is not complete - we should be able to fake
1317  * an edge even if it isn't on the 8259A...
1318  */
1319
1320 static unsigned int startup_ioapic_irq(unsigned int irq)
1321 {
1322         int was_pending = 0;
1323         unsigned long flags;
1324
1325         spin_lock_irqsave(&ioapic_lock, flags);
1326         if (irq < 16) {
1327                 disable_8259A_irq(irq);
1328                 if (i8259A_irq_pending(irq))
1329                         was_pending = 1;
1330         }
1331         __unmask_IO_APIC_irq(irq);
1332         spin_unlock_irqrestore(&ioapic_lock, flags);
1333
1334         return was_pending;
1335 }
1336
1337 static int ioapic_retrigger_irq(unsigned int irq)
1338 {
1339         struct irq_cfg *cfg = &irq_cfg[irq];
1340         cpumask_t mask;
1341         unsigned long flags;
1342
1343         spin_lock_irqsave(&vector_lock, flags);
1344         cpus_clear(mask);
1345         cpu_set(first_cpu(cfg->domain), mask);
1346
1347         send_IPI_mask(mask, cfg->vector);
1348         spin_unlock_irqrestore(&vector_lock, flags);
1349
1350         return 1;
1351 }
1352
1353 /*
1354  * Level and edge triggered IO-APIC interrupts need different handling,
1355  * so we use two separate IRQ descriptors. Edge triggered IRQs can be
1356  * handled with the level-triggered descriptor, but that one has slightly
1357  * more overhead. Level-triggered interrupts cannot be handled with the
1358  * edge-triggered handler, without risking IRQ storms and other ugly
1359  * races.
1360  */
1361
1362 #ifdef CONFIG_SMP
1363 asmlinkage void smp_irq_move_cleanup_interrupt(void)
1364 {
1365         unsigned vector, me;
1366         ack_APIC_irq();
1367         exit_idle();
1368         irq_enter();
1369
1370         me = smp_processor_id();
1371         for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
1372                 unsigned int irq;
1373                 struct irq_desc *desc;
1374                 struct irq_cfg *cfg;
1375                 irq = __get_cpu_var(vector_irq)[vector];
1376                 if (irq >= NR_IRQS)
1377                         continue;
1378
1379                 desc = irq_desc + irq;
1380                 cfg = irq_cfg + irq;
1381                 spin_lock(&desc->lock);
1382                 if (!cfg->move_cleanup_count)
1383                         goto unlock;
1384
1385                 if ((vector == cfg->vector) && cpu_isset(me, cfg->domain))
1386                         goto unlock;
1387
1388                 __get_cpu_var(vector_irq)[vector] = -1;
1389                 cfg->move_cleanup_count--;
1390 unlock:
1391                 spin_unlock(&desc->lock);
1392         }
1393
1394         irq_exit();
1395 }
1396
1397 static void irq_complete_move(unsigned int irq)
1398 {
1399         struct irq_cfg *cfg = irq_cfg + irq;
1400         unsigned vector, me;
1401
1402         if (likely(!cfg->move_in_progress))
1403                 return;
1404
1405         vector = ~get_irq_regs()->orig_rax;
1406         me = smp_processor_id();
1407         if ((vector == cfg->vector) && cpu_isset(me, cfg->domain)) {
1408                 cpumask_t cleanup_mask;
1409
1410                 cpus_and(cleanup_mask, cfg->old_domain, cpu_online_map);
1411                 cfg->move_cleanup_count = cpus_weight(cleanup_mask);
1412                 send_IPI_mask(cleanup_mask, IRQ_MOVE_CLEANUP_VECTOR);
1413                 cfg->move_in_progress = 0;
1414         }
1415 }
1416 #else
1417 static inline void irq_complete_move(unsigned int irq) {}
1418 #endif
1419
1420 static void ack_apic_edge(unsigned int irq)
1421 {
1422         irq_complete_move(irq);
1423         move_native_irq(irq);
1424         ack_APIC_irq();
1425 }
1426
1427 static void ack_apic_level(unsigned int irq)
1428 {
1429         int do_unmask_irq = 0;
1430
1431         irq_complete_move(irq);
1432 #if defined(CONFIG_GENERIC_PENDING_IRQ) || defined(CONFIG_IRQBALANCE)
1433         /* If we are moving the irq we need to mask it */
1434         if (unlikely(irq_desc[irq].status & IRQ_MOVE_PENDING)) {
1435                 do_unmask_irq = 1;
1436                 mask_IO_APIC_irq(irq);
1437         }
1438 #endif
1439
1440         /*
1441          * We must acknowledge the irq before we move it or the acknowledge will
1442          * not propagate properly.
1443          */
1444         ack_APIC_irq();
1445
1446         /* Now we can move and renable the irq */
1447         if (unlikely(do_unmask_irq)) {
1448                 /* Only migrate the irq if the ack has been received.
1449                  *
1450                  * On rare occasions the broadcast level triggered ack gets
1451                  * delayed going to ioapics, and if we reprogram the
1452                  * vector while Remote IRR is still set the irq will never
1453                  * fire again.
1454                  *
1455                  * To prevent this scenario we read the Remote IRR bit
1456                  * of the ioapic.  This has two effects.
1457                  * - On any sane system the read of the ioapic will
1458                  *   flush writes (and acks) going to the ioapic from
1459                  *   this cpu.
1460                  * - We get to see if the ACK has actually been delivered.
1461                  *
1462                  * Based on failed experiments of reprogramming the
1463                  * ioapic entry from outside of irq context starting
1464                  * with masking the ioapic entry and then polling until
1465                  * Remote IRR was clear before reprogramming the
1466                  * ioapic I don't trust the Remote IRR bit to be
1467                  * completey accurate.
1468                  *
1469                  * However there appears to be no other way to plug
1470                  * this race, so if the Remote IRR bit is not
1471                  * accurate and is causing problems then it is a hardware bug
1472                  * and you can go talk to the chipset vendor about it.
1473                  */
1474                 if (!io_apic_level_ack_pending(irq))
1475                         move_masked_irq(irq);
1476                 unmask_IO_APIC_irq(irq);
1477         }
1478 }
1479
1480 static struct irq_chip ioapic_chip __read_mostly = {
1481         .name           = "IO-APIC",
1482         .startup        = startup_ioapic_irq,
1483         .mask           = mask_IO_APIC_irq,
1484         .unmask         = unmask_IO_APIC_irq,
1485         .ack            = ack_apic_edge,
1486         .eoi            = ack_apic_level,
1487 #ifdef CONFIG_SMP
1488         .set_affinity   = set_ioapic_affinity_irq,
1489 #endif
1490         .retrigger      = ioapic_retrigger_irq,
1491 };
1492
1493 static inline void init_IO_APIC_traps(void)
1494 {
1495         int irq;
1496
1497         /*
1498          * NOTE! The local APIC isn't very good at handling
1499          * multiple interrupts at the same interrupt level.
1500          * As the interrupt level is determined by taking the
1501          * vector number and shifting that right by 4, we
1502          * want to spread these out a bit so that they don't
1503          * all fall in the same interrupt level.
1504          *
1505          * Also, we've got to be careful not to trash gate
1506          * 0x80, because int 0x80 is hm, kind of importantish. ;)
1507          */
1508         for (irq = 0; irq < NR_IRQS ; irq++) {
1509                 int tmp = irq;
1510                 if (IO_APIC_IRQ(tmp) && !irq_cfg[tmp].vector) {
1511                         /*
1512                          * Hmm.. We don't have an entry for this,
1513                          * so default to an old-fashioned 8259
1514                          * interrupt if we can..
1515                          */
1516                         if (irq < 16)
1517                                 make_8259A_irq(irq);
1518                         else
1519                                 /* Strange. Oh, well.. */
1520                                 irq_desc[irq].chip = &no_irq_chip;
1521                 }
1522         }
1523 }
1524
1525 static void enable_lapic_irq (unsigned int irq)
1526 {
1527         unsigned long v;
1528
1529         v = apic_read(APIC_LVT0);
1530         apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
1531 }
1532
1533 static void disable_lapic_irq (unsigned int irq)
1534 {
1535         unsigned long v;
1536
1537         v = apic_read(APIC_LVT0);
1538         apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1539 }
1540
1541 static void ack_lapic_irq (unsigned int irq)
1542 {
1543         ack_APIC_irq();
1544 }
1545
1546 static void end_lapic_irq (unsigned int i) { /* nothing */ }
1547
1548 static struct hw_interrupt_type lapic_irq_type __read_mostly = {
1549         .name = "local-APIC",
1550         .typename = "local-APIC-edge",
1551         .startup = NULL, /* startup_irq() not used for IRQ0 */
1552         .shutdown = NULL, /* shutdown_irq() not used for IRQ0 */
1553         .enable = enable_lapic_irq,
1554         .disable = disable_lapic_irq,
1555         .ack = ack_lapic_irq,
1556         .end = end_lapic_irq,
1557 };
1558
1559 static void setup_nmi (void)
1560 {
1561         /*
1562          * Dirty trick to enable the NMI watchdog ...
1563          * We put the 8259A master into AEOI mode and
1564          * unmask on all local APICs LVT0 as NMI.
1565          *
1566          * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
1567          * is from Maciej W. Rozycki - so we do not have to EOI from
1568          * the NMI handler or the timer interrupt.
1569          */ 
1570         printk(KERN_INFO "activating NMI Watchdog ...");
1571
1572         enable_NMI_through_LVT0(NULL);
1573
1574         printk(" done.\n");
1575 }
1576
1577 /*
1578  * This looks a bit hackish but it's about the only one way of sending
1579  * a few INTA cycles to 8259As and any associated glue logic.  ICR does
1580  * not support the ExtINT mode, unfortunately.  We need to send these
1581  * cycles as some i82489DX-based boards have glue logic that keeps the
1582  * 8259A interrupt line asserted until INTA.  --macro
1583  */
1584 static inline void unlock_ExtINT_logic(void)
1585 {
1586         int apic, pin, i;
1587         struct IO_APIC_route_entry entry0, entry1;
1588         unsigned char save_control, save_freq_select;
1589         unsigned long flags;
1590
1591         pin  = find_isa_irq_pin(8, mp_INT);
1592         apic = find_isa_irq_apic(8, mp_INT);
1593         if (pin == -1)
1594                 return;
1595
1596         spin_lock_irqsave(&ioapic_lock, flags);
1597         *(((int *)&entry0) + 1) = io_apic_read(apic, 0x11 + 2 * pin);
1598         *(((int *)&entry0) + 0) = io_apic_read(apic, 0x10 + 2 * pin);
1599         spin_unlock_irqrestore(&ioapic_lock, flags);
1600         clear_IO_APIC_pin(apic, pin);
1601
1602         memset(&entry1, 0, sizeof(entry1));
1603
1604         entry1.dest_mode = 0;                   /* physical delivery */
1605         entry1.mask = 0;                        /* unmask IRQ now */
1606         entry1.dest = hard_smp_processor_id();
1607         entry1.delivery_mode = dest_ExtINT;
1608         entry1.polarity = entry0.polarity;
1609         entry1.trigger = 0;
1610         entry1.vector = 0;
1611
1612         spin_lock_irqsave(&ioapic_lock, flags);
1613         io_apic_write(apic, 0x11 + 2 * pin, *(((int *)&entry1) + 1));
1614         io_apic_write(apic, 0x10 + 2 * pin, *(((int *)&entry1) + 0));
1615         spin_unlock_irqrestore(&ioapic_lock, flags);
1616
1617         save_control = CMOS_READ(RTC_CONTROL);
1618         save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
1619         CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
1620                    RTC_FREQ_SELECT);
1621         CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
1622
1623         i = 100;
1624         while (i-- > 0) {
1625                 mdelay(10);
1626                 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
1627                         i -= 10;
1628         }
1629
1630         CMOS_WRITE(save_control, RTC_CONTROL);
1631         CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
1632         clear_IO_APIC_pin(apic, pin);
1633
1634         spin_lock_irqsave(&ioapic_lock, flags);
1635         io_apic_write(apic, 0x11 + 2 * pin, *(((int *)&entry0) + 1));
1636         io_apic_write(apic, 0x10 + 2 * pin, *(((int *)&entry0) + 0));
1637         spin_unlock_irqrestore(&ioapic_lock, flags);
1638 }
1639
1640 /*
1641  * This code may look a bit paranoid, but it's supposed to cooperate with
1642  * a wide range of boards and BIOS bugs.  Fortunately only the timer IRQ
1643  * is so screwy.  Thanks to Brian Perkins for testing/hacking this beast
1644  * fanatically on his truly buggy board.
1645  *
1646  * FIXME: really need to revamp this for modern platforms only.
1647  */
1648 static inline void check_timer(void)
1649 {
1650         struct irq_cfg *cfg = irq_cfg + 0;
1651         int apic1, pin1, apic2, pin2;
1652
1653         /*
1654          * get/set the timer IRQ vector:
1655          */
1656         disable_8259A_irq(0);
1657         assign_irq_vector(0, TARGET_CPUS);
1658
1659         /*
1660          * Subtle, code in do_timer_interrupt() expects an AEOI
1661          * mode for the 8259A whenever interrupts are routed
1662          * through I/O APICs.  Also IRQ0 has to be enabled in
1663          * the 8259A which implies the virtual wire has to be
1664          * disabled in the local APIC.
1665          */
1666         apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
1667         init_8259A(1);
1668         if (timer_over_8254 > 0)
1669                 enable_8259A_irq(0);
1670
1671         pin1  = find_isa_irq_pin(0, mp_INT);
1672         apic1 = find_isa_irq_apic(0, mp_INT);
1673         pin2  = ioapic_i8259.pin;
1674         apic2 = ioapic_i8259.apic;
1675
1676         apic_printk(APIC_VERBOSE,KERN_INFO "..TIMER: vector=0x%02X apic1=%d pin1=%d apic2=%d pin2=%d\n",
1677                 cfg->vector, apic1, pin1, apic2, pin2);
1678
1679         if (pin1 != -1) {
1680                 /*
1681                  * Ok, does IRQ0 through the IOAPIC work?
1682                  */
1683                 unmask_IO_APIC_irq(0);
1684                 if (!no_timer_check && timer_irq_works()) {
1685                         nmi_watchdog_default();
1686                         if (nmi_watchdog == NMI_IO_APIC) {
1687                                 disable_8259A_irq(0);
1688                                 setup_nmi();
1689                                 enable_8259A_irq(0);
1690                         }
1691                         if (disable_timer_pin_1 > 0)
1692                                 clear_IO_APIC_pin(0, pin1);
1693                         return;
1694                 }
1695                 clear_IO_APIC_pin(apic1, pin1);
1696                 apic_printk(APIC_QUIET,KERN_ERR "..MP-BIOS bug: 8254 timer not "
1697                                 "connected to IO-APIC\n");
1698         }
1699
1700         apic_printk(APIC_VERBOSE,KERN_INFO "...trying to set up timer (IRQ0) "
1701                                 "through the 8259A ... ");
1702         if (pin2 != -1) {
1703                 apic_printk(APIC_VERBOSE,"\n..... (found apic %d pin %d) ...",
1704                         apic2, pin2);
1705                 /*
1706                  * legacy devices should be connected to IO APIC #0
1707                  */
1708                 setup_ExtINT_IRQ0_pin(apic2, pin2, cfg->vector);
1709                 if (timer_irq_works()) {
1710                         apic_printk(APIC_VERBOSE," works.\n");
1711                         nmi_watchdog_default();
1712                         if (nmi_watchdog == NMI_IO_APIC) {
1713                                 setup_nmi();
1714                         }
1715                         return;
1716                 }
1717                 /*
1718                  * Cleanup, just in case ...
1719                  */
1720                 clear_IO_APIC_pin(apic2, pin2);
1721         }
1722         apic_printk(APIC_VERBOSE," failed.\n");
1723
1724         if (nmi_watchdog == NMI_IO_APIC) {
1725                 printk(KERN_WARNING "timer doesn't work through the IO-APIC - disabling NMI Watchdog!\n");
1726                 nmi_watchdog = 0;
1727         }
1728
1729         apic_printk(APIC_VERBOSE, KERN_INFO "...trying to set up timer as Virtual Wire IRQ...");
1730
1731         disable_8259A_irq(0);
1732         irq_desc[0].chip = &lapic_irq_type;
1733         apic_write(APIC_LVT0, APIC_DM_FIXED | cfg->vector);     /* Fixed mode */
1734         enable_8259A_irq(0);
1735
1736         if (timer_irq_works()) {
1737                 apic_printk(APIC_VERBOSE," works.\n");
1738                 return;
1739         }
1740         apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | cfg->vector);
1741         apic_printk(APIC_VERBOSE," failed.\n");
1742
1743         apic_printk(APIC_VERBOSE, KERN_INFO "...trying to set up timer as ExtINT IRQ...");
1744
1745         init_8259A(0);
1746         make_8259A_irq(0);
1747         apic_write(APIC_LVT0, APIC_DM_EXTINT);
1748
1749         unlock_ExtINT_logic();
1750
1751         if (timer_irq_works()) {
1752                 apic_printk(APIC_VERBOSE," works.\n");
1753                 return;
1754         }
1755         apic_printk(APIC_VERBOSE," failed :(.\n");
1756         panic("IO-APIC + timer doesn't work! Try using the 'noapic' kernel parameter\n");
1757 }
1758
1759 static int __init notimercheck(char *s)
1760 {
1761         no_timer_check = 1;
1762         return 1;
1763 }
1764 __setup("no_timer_check", notimercheck);
1765
1766 /*
1767  *
1768  * IRQ's that are handled by the PIC in the MPS IOAPIC case.
1769  * - IRQ2 is the cascade IRQ, and cannot be a io-apic IRQ.
1770  *   Linux doesn't really care, as it's not actually used
1771  *   for any interrupt handling anyway.
1772  */
1773 #define PIC_IRQS        (1<<2)
1774
1775 void __init setup_IO_APIC(void)
1776 {
1777         enable_IO_APIC();
1778
1779         if (acpi_ioapic)
1780                 io_apic_irqs = ~0;      /* all IRQs go through IOAPIC */
1781         else
1782                 io_apic_irqs = ~PIC_IRQS;
1783
1784         apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
1785
1786         sync_Arb_IDs();
1787         setup_IO_APIC_irqs();
1788         init_IO_APIC_traps();
1789         check_timer();
1790         if (!acpi_ioapic)
1791                 print_IO_APIC();
1792 }
1793
1794 struct sysfs_ioapic_data {
1795         struct sys_device dev;
1796         struct IO_APIC_route_entry entry[0];
1797 };
1798 static struct sysfs_ioapic_data * mp_ioapic_data[MAX_IO_APICS];
1799
1800 static int ioapic_suspend(struct sys_device *dev, pm_message_t state)
1801 {
1802         struct IO_APIC_route_entry *entry;
1803         struct sysfs_ioapic_data *data;
1804         int i;
1805
1806         data = container_of(dev, struct sysfs_ioapic_data, dev);
1807         entry = data->entry;
1808         for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ )
1809                 *entry = ioapic_read_entry(dev->id, i);
1810
1811         return 0;
1812 }
1813
1814 static int ioapic_resume(struct sys_device *dev)
1815 {
1816         struct IO_APIC_route_entry *entry;
1817         struct sysfs_ioapic_data *data;
1818         unsigned long flags;
1819         union IO_APIC_reg_00 reg_00;
1820         int i;
1821
1822         data = container_of(dev, struct sysfs_ioapic_data, dev);
1823         entry = data->entry;
1824
1825         spin_lock_irqsave(&ioapic_lock, flags);
1826         reg_00.raw = io_apic_read(dev->id, 0);
1827         if (reg_00.bits.ID != mp_ioapics[dev->id].mpc_apicid) {
1828                 reg_00.bits.ID = mp_ioapics[dev->id].mpc_apicid;
1829                 io_apic_write(dev->id, 0, reg_00.raw);
1830         }
1831         spin_unlock_irqrestore(&ioapic_lock, flags);
1832         for (i = 0; i < nr_ioapic_registers[dev->id]; i++)
1833                 ioapic_write_entry(dev->id, i, entry[i]);
1834
1835         return 0;
1836 }
1837
1838 static struct sysdev_class ioapic_sysdev_class = {
1839         set_kset_name("ioapic"),
1840         .suspend = ioapic_suspend,
1841         .resume = ioapic_resume,
1842 };
1843
1844 static int __init ioapic_init_sysfs(void)
1845 {
1846         struct sys_device * dev;
1847         int i, size, error = 0;
1848
1849         error = sysdev_class_register(&ioapic_sysdev_class);
1850         if (error)
1851                 return error;
1852
1853         for (i = 0; i < nr_ioapics; i++ ) {
1854                 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
1855                         * sizeof(struct IO_APIC_route_entry);
1856                 mp_ioapic_data[i] = kmalloc(size, GFP_KERNEL);
1857                 if (!mp_ioapic_data[i]) {
1858                         printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
1859                         continue;
1860                 }
1861                 memset(mp_ioapic_data[i], 0, size);
1862                 dev = &mp_ioapic_data[i]->dev;
1863                 dev->id = i;
1864                 dev->cls = &ioapic_sysdev_class;
1865                 error = sysdev_register(dev);
1866                 if (error) {
1867                         kfree(mp_ioapic_data[i]);
1868                         mp_ioapic_data[i] = NULL;
1869                         printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
1870                         continue;
1871                 }
1872         }
1873
1874         return 0;
1875 }
1876
1877 device_initcall(ioapic_init_sysfs);
1878
1879 /*
1880  * Dynamic irq allocate and deallocation
1881  */
1882 int create_irq(void)
1883 {
1884         /* Allocate an unused irq */
1885         int irq;
1886         int new;
1887         unsigned long flags;
1888
1889         irq = -ENOSPC;
1890         spin_lock_irqsave(&vector_lock, flags);
1891         for (new = (NR_IRQS - 1); new >= 0; new--) {
1892                 if (platform_legacy_irq(new))
1893                         continue;
1894                 if (irq_cfg[new].vector != 0)
1895                         continue;
1896                 if (__assign_irq_vector(new, TARGET_CPUS) == 0)
1897                         irq = new;
1898                 break;
1899         }
1900         spin_unlock_irqrestore(&vector_lock, flags);
1901
1902         if (irq >= 0) {
1903                 dynamic_irq_init(irq);
1904         }
1905         return irq;
1906 }
1907
1908 void destroy_irq(unsigned int irq)
1909 {
1910         unsigned long flags;
1911
1912         dynamic_irq_cleanup(irq);
1913
1914         spin_lock_irqsave(&vector_lock, flags);
1915         __clear_irq_vector(irq);
1916         spin_unlock_irqrestore(&vector_lock, flags);
1917 }
1918
1919 /*
1920  * MSI mesage composition
1921  */
1922 #ifdef CONFIG_PCI_MSI
1923 static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq, struct msi_msg *msg)
1924 {
1925         struct irq_cfg *cfg = irq_cfg + irq;
1926         int err;
1927         unsigned dest;
1928         cpumask_t tmp;
1929
1930         tmp = TARGET_CPUS;
1931         err = assign_irq_vector(irq, tmp);
1932         if (!err) {
1933                 cpus_and(tmp, cfg->domain, tmp);
1934                 dest = cpu_mask_to_apicid(tmp);
1935
1936                 msg->address_hi = MSI_ADDR_BASE_HI;
1937                 msg->address_lo =
1938                         MSI_ADDR_BASE_LO |
1939                         ((INT_DEST_MODE == 0) ?
1940                                 MSI_ADDR_DEST_MODE_PHYSICAL:
1941                                 MSI_ADDR_DEST_MODE_LOGICAL) |
1942                         ((INT_DELIVERY_MODE != dest_LowestPrio) ?
1943                                 MSI_ADDR_REDIRECTION_CPU:
1944                                 MSI_ADDR_REDIRECTION_LOWPRI) |
1945                         MSI_ADDR_DEST_ID(dest);
1946
1947                 msg->data =
1948                         MSI_DATA_TRIGGER_EDGE |
1949                         MSI_DATA_LEVEL_ASSERT |
1950                         ((INT_DELIVERY_MODE != dest_LowestPrio) ?
1951                                 MSI_DATA_DELIVERY_FIXED:
1952                                 MSI_DATA_DELIVERY_LOWPRI) |
1953                         MSI_DATA_VECTOR(cfg->vector);
1954         }
1955         return err;
1956 }
1957
1958 #ifdef CONFIG_SMP
1959 static void set_msi_irq_affinity(unsigned int irq, cpumask_t mask)
1960 {
1961         struct irq_cfg *cfg = irq_cfg + irq;
1962         struct msi_msg msg;
1963         unsigned int dest;
1964         cpumask_t tmp;
1965
1966         cpus_and(tmp, mask, cpu_online_map);
1967         if (cpus_empty(tmp))
1968                 return;
1969
1970         if (assign_irq_vector(irq, mask))
1971                 return;
1972
1973         cpus_and(tmp, cfg->domain, mask);
1974         dest = cpu_mask_to_apicid(tmp);
1975
1976         read_msi_msg(irq, &msg);
1977
1978         msg.data &= ~MSI_DATA_VECTOR_MASK;
1979         msg.data |= MSI_DATA_VECTOR(cfg->vector);
1980         msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
1981         msg.address_lo |= MSI_ADDR_DEST_ID(dest);
1982
1983         write_msi_msg(irq, &msg);
1984         irq_desc[irq].affinity = mask;
1985 }
1986 #endif /* CONFIG_SMP */
1987
1988 /*
1989  * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
1990  * which implement the MSI or MSI-X Capability Structure.
1991  */
1992 static struct irq_chip msi_chip = {
1993         .name           = "PCI-MSI",
1994         .unmask         = unmask_msi_irq,
1995         .mask           = mask_msi_irq,
1996         .ack            = ack_apic_edge,
1997 #ifdef CONFIG_SMP
1998         .set_affinity   = set_msi_irq_affinity,
1999 #endif
2000         .retrigger      = ioapic_retrigger_irq,
2001 };
2002
2003 int arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc)
2004 {
2005         struct msi_msg msg;
2006         int irq, ret;
2007         irq = create_irq();
2008         if (irq < 0)
2009                 return irq;
2010
2011         ret = msi_compose_msg(dev, irq, &msg);
2012         if (ret < 0) {
2013                 destroy_irq(irq);
2014                 return ret;
2015         }
2016
2017         set_irq_msi(irq, desc);
2018         write_msi_msg(irq, &msg);
2019
2020         set_irq_chip_and_handler_name(irq, &msi_chip, handle_edge_irq, "edge");
2021
2022         return 0;
2023 }
2024
2025 void arch_teardown_msi_irq(unsigned int irq)
2026 {
2027         destroy_irq(irq);
2028 }
2029
2030 #endif /* CONFIG_PCI_MSI */
2031
2032 /*
2033  * Hypertransport interrupt support
2034  */
2035 #ifdef CONFIG_HT_IRQ
2036
2037 #ifdef CONFIG_SMP
2038
2039 static void target_ht_irq(unsigned int irq, unsigned int dest, u8 vector)
2040 {
2041         struct ht_irq_msg msg;
2042         fetch_ht_irq_msg(irq, &msg);
2043
2044         msg.address_lo &= ~(HT_IRQ_LOW_VECTOR_MASK | HT_IRQ_LOW_DEST_ID_MASK);
2045         msg.address_hi &= ~(HT_IRQ_HIGH_DEST_ID_MASK);
2046
2047         msg.address_lo |= HT_IRQ_LOW_VECTOR(vector) | HT_IRQ_LOW_DEST_ID(dest);
2048         msg.address_hi |= HT_IRQ_HIGH_DEST_ID(dest);
2049
2050         write_ht_irq_msg(irq, &msg);
2051 }
2052
2053 static void set_ht_irq_affinity(unsigned int irq, cpumask_t mask)
2054 {
2055         struct irq_cfg *cfg = irq_cfg + irq;
2056         unsigned int dest;
2057         cpumask_t tmp;
2058
2059         cpus_and(tmp, mask, cpu_online_map);
2060         if (cpus_empty(tmp))
2061                 return;
2062
2063         if (assign_irq_vector(irq, mask))
2064                 return;
2065
2066         cpus_and(tmp, cfg->domain, mask);
2067         dest = cpu_mask_to_apicid(tmp);
2068
2069         target_ht_irq(irq, dest, cfg->vector);
2070         irq_desc[irq].affinity = mask;
2071 }
2072 #endif
2073
2074 static struct irq_chip ht_irq_chip = {
2075         .name           = "PCI-HT",
2076         .mask           = mask_ht_irq,
2077         .unmask         = unmask_ht_irq,
2078         .ack            = ack_apic_edge,
2079 #ifdef CONFIG_SMP
2080         .set_affinity   = set_ht_irq_affinity,
2081 #endif
2082         .retrigger      = ioapic_retrigger_irq,
2083 };
2084
2085 int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev)
2086 {
2087         struct irq_cfg *cfg = irq_cfg + irq;
2088         int err;
2089         cpumask_t tmp;
2090
2091         tmp = TARGET_CPUS;
2092         err = assign_irq_vector(irq, tmp);
2093         if (!err) {
2094                 struct ht_irq_msg msg;
2095                 unsigned dest;
2096
2097                 cpus_and(tmp, cfg->domain, tmp);
2098                 dest = cpu_mask_to_apicid(tmp);
2099
2100                 msg.address_hi = HT_IRQ_HIGH_DEST_ID(dest);
2101
2102                 msg.address_lo =
2103                         HT_IRQ_LOW_BASE |
2104                         HT_IRQ_LOW_DEST_ID(dest) |
2105                         HT_IRQ_LOW_VECTOR(cfg->vector) |
2106                         ((INT_DEST_MODE == 0) ?
2107                                 HT_IRQ_LOW_DM_PHYSICAL :
2108                                 HT_IRQ_LOW_DM_LOGICAL) |
2109                         HT_IRQ_LOW_RQEOI_EDGE |
2110                         ((INT_DELIVERY_MODE != dest_LowestPrio) ?
2111                                 HT_IRQ_LOW_MT_FIXED :
2112                                 HT_IRQ_LOW_MT_ARBITRATED) |
2113                         HT_IRQ_LOW_IRQ_MASKED;
2114
2115                 write_ht_irq_msg(irq, &msg);
2116
2117                 set_irq_chip_and_handler_name(irq, &ht_irq_chip,
2118                                               handle_edge_irq, "edge");
2119         }
2120         return err;
2121 }
2122 #endif /* CONFIG_HT_IRQ */
2123
2124 /* --------------------------------------------------------------------------
2125                           ACPI-based IOAPIC Configuration
2126    -------------------------------------------------------------------------- */
2127
2128 #ifdef CONFIG_ACPI
2129
2130 #define IO_APIC_MAX_ID          0xFE
2131
2132 int __init io_apic_get_redir_entries (int ioapic)
2133 {
2134         union IO_APIC_reg_01    reg_01;
2135         unsigned long flags;
2136
2137         spin_lock_irqsave(&ioapic_lock, flags);
2138         reg_01.raw = io_apic_read(ioapic, 1);
2139         spin_unlock_irqrestore(&ioapic_lock, flags);
2140
2141         return reg_01.bits.entries;
2142 }
2143
2144
2145 int io_apic_set_pci_routing (int ioapic, int pin, int irq, int triggering, int polarity)
2146 {
2147         if (!IO_APIC_IRQ(irq)) {
2148                 apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
2149                         ioapic);
2150                 return -EINVAL;
2151         }
2152
2153         /*
2154          * IRQs < 16 are already in the irq_2_pin[] map
2155          */
2156         if (irq >= 16)
2157                 add_pin_to_irq(irq, ioapic, pin);
2158
2159         setup_IO_APIC_irq(ioapic, pin, irq, triggering, polarity);
2160
2161         return 0;
2162 }
2163
2164 #endif /* CONFIG_ACPI */
2165
2166
2167 /*
2168  * This function currently is only a helper for the i386 smp boot process where
2169  * we need to reprogram the ioredtbls to cater for the cpus which have come online
2170  * so mask in all cases should simply be TARGET_CPUS
2171  */
2172 #ifdef CONFIG_SMP
2173 void __init setup_ioapic_dest(void)
2174 {
2175         int pin, ioapic, irq, irq_entry;
2176
2177         if (skip_ioapic_setup == 1)
2178                 return;
2179
2180         for (ioapic = 0; ioapic < nr_ioapics; ioapic++) {
2181                 for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
2182                         irq_entry = find_irq_entry(ioapic, pin, mp_INT);
2183                         if (irq_entry == -1)
2184                                 continue;
2185                         irq = pin_2_irq(irq_entry, ioapic, pin);
2186
2187                         /* setup_IO_APIC_irqs could fail to get vector for some device
2188                          * when you have too many devices, because at that time only boot
2189                          * cpu is online.
2190                          */
2191                         if (!irq_cfg[irq].vector)
2192                                 setup_IO_APIC_irq(ioapic, pin, irq,
2193                                                   irq_trigger(irq_entry),
2194                                                   irq_polarity(irq_entry));
2195                         else
2196                                 set_ioapic_affinity_irq(irq, TARGET_CPUS);
2197                 }
2198
2199         }
2200 }
2201 #endif