Merge ../linux-2.6-watchdog-mm
[linux-drm-fsl-dcu.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/smp_lock.h>
29 #include <linux/pci.h>
30 #include <linux/mc146818rtc.h>
31 #include <linux/acpi.h>
32 #include <linux/sysdev.h>
33 #include <linux/msi.h>
34 #include <linux/htirq.h>
35 #ifdef CONFIG_ACPI
36 #include <acpi/acpi_bus.h>
37 #endif
38
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 static int assign_irq_vector(int irq, cpumask_t mask, cpumask_t *result);
51
52 #define __apicdebuginit  __init
53
54 int sis_apic_bug; /* not actually supported, dummy for compile */
55
56 static int no_timer_check;
57
58 static int disable_timer_pin_1 __initdata;
59
60 int timer_over_8254 __initdata = 1;
61
62 /* Where if anywhere is the i8259 connect in external int mode */
63 static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
64
65 static DEFINE_SPINLOCK(ioapic_lock);
66 DEFINE_SPINLOCK(vector_lock);
67
68 /*
69  * # of IRQ routing registers
70  */
71 int nr_ioapic_registers[MAX_IO_APICS];
72
73 /*
74  * Rough estimation of how many shared IRQs there are, can
75  * be changed anytime.
76  */
77 #define MAX_PLUS_SHARED_IRQS NR_IRQ_VECTORS
78 #define PIN_MAP_SIZE (MAX_PLUS_SHARED_IRQS + NR_IRQS)
79
80 /*
81  * This is performance-critical, we want to do it O(1)
82  *
83  * the indexing order of this array favors 1:1 mappings
84  * between pins and IRQs.
85  */
86
87 static struct irq_pin_list {
88         short apic, pin, next;
89 } irq_2_pin[PIN_MAP_SIZE];
90
91 struct io_apic {
92         unsigned int index;
93         unsigned int unused[3];
94         unsigned int data;
95 };
96
97 static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
98 {
99         return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
100                 + (mp_ioapics[idx].mpc_apicaddr & ~PAGE_MASK);
101 }
102
103 static inline unsigned int io_apic_read(unsigned int apic, unsigned int reg)
104 {
105         struct io_apic __iomem *io_apic = io_apic_base(apic);
106         writel(reg, &io_apic->index);
107         return readl(&io_apic->data);
108 }
109
110 static inline void io_apic_write(unsigned int apic, unsigned int reg, unsigned int value)
111 {
112         struct io_apic __iomem *io_apic = io_apic_base(apic);
113         writel(reg, &io_apic->index);
114         writel(value, &io_apic->data);
115 }
116
117 /*
118  * Re-write a value: to be used for read-modify-write
119  * cycles where the read already set up the index register.
120  */
121 static inline void io_apic_modify(unsigned int apic, unsigned int value)
122 {
123         struct io_apic __iomem *io_apic = io_apic_base(apic);
124         writel(value, &io_apic->data);
125 }
126
127 /*
128  * Synchronize the IO-APIC and the CPU by doing
129  * a dummy read from the IO-APIC
130  */
131 static inline void io_apic_sync(unsigned int apic)
132 {
133         struct io_apic __iomem *io_apic = io_apic_base(apic);
134         readl(&io_apic->data);
135 }
136
137 #define __DO_ACTION(R, ACTION, FINAL)                                   \
138                                                                         \
139 {                                                                       \
140         int pin;                                                        \
141         struct irq_pin_list *entry = irq_2_pin + irq;                   \
142                                                                         \
143         BUG_ON(irq >= NR_IRQS);                                         \
144         for (;;) {                                                      \
145                 unsigned int reg;                                       \
146                 pin = entry->pin;                                       \
147                 if (pin == -1)                                          \
148                         break;                                          \
149                 reg = io_apic_read(entry->apic, 0x10 + R + pin*2);      \
150                 reg ACTION;                                             \
151                 io_apic_modify(entry->apic, reg);                       \
152                 if (!entry->next)                                       \
153                         break;                                          \
154                 entry = irq_2_pin + entry->next;                        \
155         }                                                               \
156         FINAL;                                                          \
157 }
158
159 union entry_union {
160         struct { u32 w1, w2; };
161         struct IO_APIC_route_entry entry;
162 };
163
164 static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
165 {
166         union entry_union eu;
167         unsigned long flags;
168         spin_lock_irqsave(&ioapic_lock, flags);
169         eu.w1 = io_apic_read(apic, 0x10 + 2 * pin);
170         eu.w2 = io_apic_read(apic, 0x11 + 2 * pin);
171         spin_unlock_irqrestore(&ioapic_lock, flags);
172         return eu.entry;
173 }
174
175 /*
176  * When we write a new IO APIC routing entry, we need to write the high
177  * word first! If the mask bit in the low word is clear, we will enable
178  * the interrupt, and we need to make sure the entry is fully populated
179  * before that happens.
180  */
181 static void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
182 {
183         unsigned long flags;
184         union entry_union eu;
185         eu.entry = e;
186         spin_lock_irqsave(&ioapic_lock, flags);
187         io_apic_write(apic, 0x11 + 2*pin, eu.w2);
188         io_apic_write(apic, 0x10 + 2*pin, eu.w1);
189         spin_unlock_irqrestore(&ioapic_lock, flags);
190 }
191
192 /*
193  * When we mask an IO APIC routing entry, we need to write the low
194  * word first, in order to set the mask bit before we change the
195  * high bits!
196  */
197 static void ioapic_mask_entry(int apic, int pin)
198 {
199         unsigned long flags;
200         union entry_union eu = { .entry.mask = 1 };
201
202         spin_lock_irqsave(&ioapic_lock, flags);
203         io_apic_write(apic, 0x10 + 2*pin, eu.w1);
204         io_apic_write(apic, 0x11 + 2*pin, eu.w2);
205         spin_unlock_irqrestore(&ioapic_lock, flags);
206 }
207
208 #ifdef CONFIG_SMP
209 static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, u8 vector)
210 {
211         int apic, pin;
212         struct irq_pin_list *entry = irq_2_pin + irq;
213
214         BUG_ON(irq >= NR_IRQS);
215         for (;;) {
216                 unsigned int reg;
217                 apic = entry->apic;
218                 pin = entry->pin;
219                 if (pin == -1)
220                         break;
221                 io_apic_write(apic, 0x11 + pin*2, dest);
222                 reg = io_apic_read(apic, 0x10 + pin*2);
223                 reg &= ~0x000000ff;
224                 reg |= vector;
225                 io_apic_modify(apic, reg);
226                 if (!entry->next)
227                         break;
228                 entry = irq_2_pin + entry->next;
229         }
230 }
231
232 static void set_ioapic_affinity_irq(unsigned int irq, cpumask_t mask)
233 {
234         unsigned long flags;
235         unsigned int dest;
236         cpumask_t tmp;
237         int vector;
238
239         cpus_and(tmp, mask, cpu_online_map);
240         if (cpus_empty(tmp))
241                 tmp = TARGET_CPUS;
242
243         cpus_and(mask, tmp, CPU_MASK_ALL);
244
245         vector = assign_irq_vector(irq, mask, &tmp);
246         if (vector < 0)
247                 return;
248
249         dest = cpu_mask_to_apicid(tmp);
250
251         /*
252          * Only the high 8 bits are valid.
253          */
254         dest = SET_APIC_LOGICAL_ID(dest);
255
256         spin_lock_irqsave(&ioapic_lock, flags);
257         __target_IO_APIC_irq(irq, dest, vector);
258         set_native_irq_info(irq, mask);
259         spin_unlock_irqrestore(&ioapic_lock, flags);
260 }
261 #endif
262
263 /*
264  * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
265  * shared ISA-space IRQs, so we have to support them. We are super
266  * fast in the common case, and fast for shared ISA-space IRQs.
267  */
268 static void add_pin_to_irq(unsigned int irq, int apic, int pin)
269 {
270         static int first_free_entry = NR_IRQS;
271         struct irq_pin_list *entry = irq_2_pin + irq;
272
273         BUG_ON(irq >= NR_IRQS);
274         while (entry->next)
275                 entry = irq_2_pin + entry->next;
276
277         if (entry->pin != -1) {
278                 entry->next = first_free_entry;
279                 entry = irq_2_pin + entry->next;
280                 if (++first_free_entry >= PIN_MAP_SIZE)
281                         panic("io_apic.c: ran out of irq_2_pin entries!");
282         }
283         entry->apic = apic;
284         entry->pin = pin;
285 }
286
287
288 #define DO_ACTION(name,R,ACTION, FINAL)                                 \
289                                                                         \
290         static void name##_IO_APIC_irq (unsigned int irq)               \
291         __DO_ACTION(R, ACTION, FINAL)
292
293 DO_ACTION( __mask,             0, |= 0x00010000, io_apic_sync(entry->apic) )
294                                                 /* mask = 1 */
295 DO_ACTION( __unmask,           0, &= 0xfffeffff, )
296                                                 /* mask = 0 */
297
298 static void mask_IO_APIC_irq (unsigned int irq)
299 {
300         unsigned long flags;
301
302         spin_lock_irqsave(&ioapic_lock, flags);
303         __mask_IO_APIC_irq(irq);
304         spin_unlock_irqrestore(&ioapic_lock, flags);
305 }
306
307 static void unmask_IO_APIC_irq (unsigned int irq)
308 {
309         unsigned long flags;
310
311         spin_lock_irqsave(&ioapic_lock, flags);
312         __unmask_IO_APIC_irq(irq);
313         spin_unlock_irqrestore(&ioapic_lock, flags);
314 }
315
316 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
317 {
318         struct IO_APIC_route_entry entry;
319
320         /* Check delivery_mode to be sure we're not clearing an SMI pin */
321         entry = ioapic_read_entry(apic, pin);
322         if (entry.delivery_mode == dest_SMI)
323                 return;
324         /*
325          * Disable it in the IO-APIC irq-routing table:
326          */
327         ioapic_mask_entry(apic, pin);
328 }
329
330 static void clear_IO_APIC (void)
331 {
332         int apic, pin;
333
334         for (apic = 0; apic < nr_ioapics; apic++)
335                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
336                         clear_IO_APIC_pin(apic, pin);
337 }
338
339 int skip_ioapic_setup;
340 int ioapic_force;
341
342 /* dummy parsing: see setup.c */
343
344 static int __init disable_ioapic_setup(char *str)
345 {
346         skip_ioapic_setup = 1;
347         return 0;
348 }
349 early_param("noapic", disable_ioapic_setup);
350
351 /* Actually the next is obsolete, but keep it for paranoid reasons -AK */
352 static int __init disable_timer_pin_setup(char *arg)
353 {
354         disable_timer_pin_1 = 1;
355         return 1;
356 }
357 __setup("disable_timer_pin_1", disable_timer_pin_setup);
358
359 static int __init setup_disable_8254_timer(char *s)
360 {
361         timer_over_8254 = -1;
362         return 1;
363 }
364 static int __init setup_enable_8254_timer(char *s)
365 {
366         timer_over_8254 = 2;
367         return 1;
368 }
369
370 __setup("disable_8254_timer", setup_disable_8254_timer);
371 __setup("enable_8254_timer", setup_enable_8254_timer);
372
373
374 /*
375  * Find the IRQ entry number of a certain pin.
376  */
377 static int find_irq_entry(int apic, int pin, int type)
378 {
379         int i;
380
381         for (i = 0; i < mp_irq_entries; i++)
382                 if (mp_irqs[i].mpc_irqtype == type &&
383                     (mp_irqs[i].mpc_dstapic == mp_ioapics[apic].mpc_apicid ||
384                      mp_irqs[i].mpc_dstapic == MP_APIC_ALL) &&
385                     mp_irqs[i].mpc_dstirq == pin)
386                         return i;
387
388         return -1;
389 }
390
391 /*
392  * Find the pin to which IRQ[irq] (ISA) is connected
393  */
394 static int __init find_isa_irq_pin(int irq, int type)
395 {
396         int i;
397
398         for (i = 0; i < mp_irq_entries; i++) {
399                 int lbus = mp_irqs[i].mpc_srcbus;
400
401                 if (test_bit(lbus, mp_bus_not_pci) &&
402                     (mp_irqs[i].mpc_irqtype == type) &&
403                     (mp_irqs[i].mpc_srcbusirq == irq))
404
405                         return mp_irqs[i].mpc_dstirq;
406         }
407         return -1;
408 }
409
410 static int __init find_isa_irq_apic(int irq, int type)
411 {
412         int i;
413
414         for (i = 0; i < mp_irq_entries; i++) {
415                 int lbus = mp_irqs[i].mpc_srcbus;
416
417                 if (test_bit(lbus, mp_bus_not_pci) &&
418                     (mp_irqs[i].mpc_irqtype == type) &&
419                     (mp_irqs[i].mpc_srcbusirq == irq))
420                         break;
421         }
422         if (i < mp_irq_entries) {
423                 int apic;
424                 for(apic = 0; apic < nr_ioapics; apic++) {
425                         if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic)
426                                 return apic;
427                 }
428         }
429
430         return -1;
431 }
432
433 /*
434  * Find a specific PCI IRQ entry.
435  * Not an __init, possibly needed by modules
436  */
437 static int pin_2_irq(int idx, int apic, int pin);
438
439 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
440 {
441         int apic, i, best_guess = -1;
442
443         apic_printk(APIC_DEBUG, "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
444                 bus, slot, pin);
445         if (mp_bus_id_to_pci_bus[bus] == -1) {
446                 apic_printk(APIC_VERBOSE, "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
447                 return -1;
448         }
449         for (i = 0; i < mp_irq_entries; i++) {
450                 int lbus = mp_irqs[i].mpc_srcbus;
451
452                 for (apic = 0; apic < nr_ioapics; apic++)
453                         if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic ||
454                             mp_irqs[i].mpc_dstapic == MP_APIC_ALL)
455                                 break;
456
457                 if (!test_bit(lbus, mp_bus_not_pci) &&
458                     !mp_irqs[i].mpc_irqtype &&
459                     (bus == lbus) &&
460                     (slot == ((mp_irqs[i].mpc_srcbusirq >> 2) & 0x1f))) {
461                         int irq = pin_2_irq(i,apic,mp_irqs[i].mpc_dstirq);
462
463                         if (!(apic || IO_APIC_IRQ(irq)))
464                                 continue;
465
466                         if (pin == (mp_irqs[i].mpc_srcbusirq & 3))
467                                 return irq;
468                         /*
469                          * Use the first all-but-pin matching entry as a
470                          * best-guess fuzzy result for broken mptables.
471                          */
472                         if (best_guess < 0)
473                                 best_guess = irq;
474                 }
475         }
476         BUG_ON(best_guess >= NR_IRQS);
477         return best_guess;
478 }
479
480 /* ISA interrupts are always polarity zero edge triggered,
481  * when listed as conforming in the MP table. */
482
483 #define default_ISA_trigger(idx)        (0)
484 #define default_ISA_polarity(idx)       (0)
485
486 /* PCI interrupts are always polarity one level triggered,
487  * when listed as conforming in the MP table. */
488
489 #define default_PCI_trigger(idx)        (1)
490 #define default_PCI_polarity(idx)       (1)
491
492 static int __init MPBIOS_polarity(int idx)
493 {
494         int bus = mp_irqs[idx].mpc_srcbus;
495         int polarity;
496
497         /*
498          * Determine IRQ line polarity (high active or low active):
499          */
500         switch (mp_irqs[idx].mpc_irqflag & 3)
501         {
502                 case 0: /* conforms, ie. bus-type dependent polarity */
503                         if (test_bit(bus, mp_bus_not_pci))
504                                 polarity = default_ISA_polarity(idx);
505                         else
506                                 polarity = default_PCI_polarity(idx);
507                         break;
508                 case 1: /* high active */
509                 {
510                         polarity = 0;
511                         break;
512                 }
513                 case 2: /* reserved */
514                 {
515                         printk(KERN_WARNING "broken BIOS!!\n");
516                         polarity = 1;
517                         break;
518                 }
519                 case 3: /* low active */
520                 {
521                         polarity = 1;
522                         break;
523                 }
524                 default: /* invalid */
525                 {
526                         printk(KERN_WARNING "broken BIOS!!\n");
527                         polarity = 1;
528                         break;
529                 }
530         }
531         return polarity;
532 }
533
534 static int MPBIOS_trigger(int idx)
535 {
536         int bus = mp_irqs[idx].mpc_srcbus;
537         int trigger;
538
539         /*
540          * Determine IRQ trigger mode (edge or level sensitive):
541          */
542         switch ((mp_irqs[idx].mpc_irqflag>>2) & 3)
543         {
544                 case 0: /* conforms, ie. bus-type dependent */
545                         if (test_bit(bus, mp_bus_not_pci))
546                                 trigger = default_ISA_trigger(idx);
547                         else
548                                 trigger = default_PCI_trigger(idx);
549                         break;
550                 case 1: /* edge */
551                 {
552                         trigger = 0;
553                         break;
554                 }
555                 case 2: /* reserved */
556                 {
557                         printk(KERN_WARNING "broken BIOS!!\n");
558                         trigger = 1;
559                         break;
560                 }
561                 case 3: /* level */
562                 {
563                         trigger = 1;
564                         break;
565                 }
566                 default: /* invalid */
567                 {
568                         printk(KERN_WARNING "broken BIOS!!\n");
569                         trigger = 0;
570                         break;
571                 }
572         }
573         return trigger;
574 }
575
576 static inline int irq_polarity(int idx)
577 {
578         return MPBIOS_polarity(idx);
579 }
580
581 static inline int irq_trigger(int idx)
582 {
583         return MPBIOS_trigger(idx);
584 }
585
586 static int pin_2_irq(int idx, int apic, int pin)
587 {
588         int irq, i;
589         int bus = mp_irqs[idx].mpc_srcbus;
590
591         /*
592          * Debugging check, we are in big trouble if this message pops up!
593          */
594         if (mp_irqs[idx].mpc_dstirq != pin)
595                 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
596
597         if (test_bit(bus, mp_bus_not_pci)) {
598                 irq = mp_irqs[idx].mpc_srcbusirq;
599         } else {
600                 /*
601                  * PCI IRQs are mapped in order
602                  */
603                 i = irq = 0;
604                 while (i < apic)
605                         irq += nr_ioapic_registers[i++];
606                 irq += pin;
607         }
608         BUG_ON(irq >= NR_IRQS);
609         return irq;
610 }
611
612 static inline int IO_APIC_irq_trigger(int irq)
613 {
614         int apic, idx, pin;
615
616         for (apic = 0; apic < nr_ioapics; apic++) {
617                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
618                         idx = find_irq_entry(apic,pin,mp_INT);
619                         if ((idx != -1) && (irq == pin_2_irq(idx,apic,pin)))
620                                 return irq_trigger(idx);
621                 }
622         }
623         /*
624          * nonexistent IRQs are edge default
625          */
626         return 0;
627 }
628
629 /* irq_vectors is indexed by the sum of all RTEs in all I/O APICs. */
630 static u8 irq_vector[NR_IRQ_VECTORS] __read_mostly = {
631         [0] = FIRST_EXTERNAL_VECTOR + 0,
632         [1] = FIRST_EXTERNAL_VECTOR + 1,
633         [2] = FIRST_EXTERNAL_VECTOR + 2,
634         [3] = FIRST_EXTERNAL_VECTOR + 3,
635         [4] = FIRST_EXTERNAL_VECTOR + 4,
636         [5] = FIRST_EXTERNAL_VECTOR + 5,
637         [6] = FIRST_EXTERNAL_VECTOR + 6,
638         [7] = FIRST_EXTERNAL_VECTOR + 7,
639         [8] = FIRST_EXTERNAL_VECTOR + 8,
640         [9] = FIRST_EXTERNAL_VECTOR + 9,
641         [10] = FIRST_EXTERNAL_VECTOR + 10,
642         [11] = FIRST_EXTERNAL_VECTOR + 11,
643         [12] = FIRST_EXTERNAL_VECTOR + 12,
644         [13] = FIRST_EXTERNAL_VECTOR + 13,
645         [14] = FIRST_EXTERNAL_VECTOR + 14,
646         [15] = FIRST_EXTERNAL_VECTOR + 15,
647 };
648
649 static cpumask_t irq_domain[NR_IRQ_VECTORS] __read_mostly = {
650         [0] = CPU_MASK_ALL,
651         [1] = CPU_MASK_ALL,
652         [2] = CPU_MASK_ALL,
653         [3] = CPU_MASK_ALL,
654         [4] = CPU_MASK_ALL,
655         [5] = CPU_MASK_ALL,
656         [6] = CPU_MASK_ALL,
657         [7] = CPU_MASK_ALL,
658         [8] = CPU_MASK_ALL,
659         [9] = CPU_MASK_ALL,
660         [10] = CPU_MASK_ALL,
661         [11] = CPU_MASK_ALL,
662         [12] = CPU_MASK_ALL,
663         [13] = CPU_MASK_ALL,
664         [14] = CPU_MASK_ALL,
665         [15] = CPU_MASK_ALL,
666 };
667
668 static int __assign_irq_vector(int irq, cpumask_t mask, cpumask_t *result)
669 {
670         /*
671          * NOTE! The local APIC isn't very good at handling
672          * multiple interrupts at the same interrupt level.
673          * As the interrupt level is determined by taking the
674          * vector number and shifting that right by 4, we
675          * want to spread these out a bit so that they don't
676          * all fall in the same interrupt level.
677          *
678          * Also, we've got to be careful not to trash gate
679          * 0x80, because int 0x80 is hm, kind of importantish. ;)
680          */
681         static int current_vector = FIRST_DEVICE_VECTOR, current_offset = 0;
682         int old_vector = -1;
683         int cpu;
684
685         BUG_ON((unsigned)irq >= NR_IRQ_VECTORS);
686
687         /* Only try and allocate irqs on cpus that are present */
688         cpus_and(mask, mask, cpu_online_map);
689
690         if (irq_vector[irq] > 0)
691                 old_vector = irq_vector[irq];
692         if (old_vector > 0) {
693                 cpus_and(*result, irq_domain[irq], mask);
694                 if (!cpus_empty(*result))
695                         return old_vector;
696         }
697
698         for_each_cpu_mask(cpu, mask) {
699                 cpumask_t domain, new_mask;
700                 int new_cpu;
701                 int vector, offset;
702
703                 domain = vector_allocation_domain(cpu);
704                 cpus_and(new_mask, domain, cpu_online_map);
705
706                 vector = current_vector;
707                 offset = current_offset;
708 next:
709                 vector += 8;
710                 if (vector >= FIRST_SYSTEM_VECTOR) {
711                         /* If we run out of vectors on large boxen, must share them. */
712                         offset = (offset + 1) % 8;
713                         vector = FIRST_DEVICE_VECTOR + offset;
714                 }
715                 if (unlikely(current_vector == vector))
716                         continue;
717                 if (vector == IA32_SYSCALL_VECTOR)
718                         goto next;
719                 for_each_cpu_mask(new_cpu, new_mask)
720                         if (per_cpu(vector_irq, new_cpu)[vector] != -1)
721                                 goto next;
722                 /* Found one! */
723                 current_vector = vector;
724                 current_offset = offset;
725                 if (old_vector >= 0) {
726                         cpumask_t old_mask;
727                         int old_cpu;
728                         cpus_and(old_mask, irq_domain[irq], cpu_online_map);
729                         for_each_cpu_mask(old_cpu, old_mask)
730                                 per_cpu(vector_irq, old_cpu)[old_vector] = -1;
731                 }
732                 for_each_cpu_mask(new_cpu, new_mask)
733                         per_cpu(vector_irq, new_cpu)[vector] = irq;
734                 irq_vector[irq] = vector;
735                 irq_domain[irq] = domain;
736                 cpus_and(*result, domain, mask);
737                 return vector;
738         }
739         return -ENOSPC;
740 }
741
742 static int assign_irq_vector(int irq, cpumask_t mask, cpumask_t *result)
743 {
744         int vector;
745         unsigned long flags;
746
747         spin_lock_irqsave(&vector_lock, flags);
748         vector = __assign_irq_vector(irq, mask, result);
749         spin_unlock_irqrestore(&vector_lock, flags);
750         return vector;
751 }
752
753 void __setup_vector_irq(int cpu)
754 {
755         /* Initialize vector_irq on a new cpu */
756         /* This function must be called with vector_lock held */
757         int irq, vector;
758
759         /* Mark the inuse vectors */
760         for (irq = 0; irq < NR_IRQ_VECTORS; ++irq) {
761                 if (!cpu_isset(cpu, irq_domain[irq]))
762                         continue;
763                 vector = irq_vector[irq];
764                 per_cpu(vector_irq, cpu)[vector] = irq;
765         }
766         /* Mark the free vectors */
767         for (vector = 0; vector < NR_VECTORS; ++vector) {
768                 irq = per_cpu(vector_irq, cpu)[vector];
769                 if (irq < 0)
770                         continue;
771                 if (!cpu_isset(cpu, irq_domain[irq]))
772                         per_cpu(vector_irq, cpu)[vector] = -1;
773         }
774 }
775
776
777 extern void (*interrupt[NR_IRQS])(void);
778
779 static struct irq_chip ioapic_chip;
780
781 #define IOAPIC_AUTO     -1
782 #define IOAPIC_EDGE     0
783 #define IOAPIC_LEVEL    1
784
785 static void ioapic_register_intr(int irq, int vector, unsigned long trigger)
786 {
787         if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
788                         trigger == IOAPIC_LEVEL)
789                 set_irq_chip_and_handler_name(irq, &ioapic_chip,
790                                               handle_fasteoi_irq, "fasteoi");
791         else {
792                 irq_desc[irq].status |= IRQ_DELAYED_DISABLE;
793                 set_irq_chip_and_handler_name(irq, &ioapic_chip,
794                                               handle_edge_irq, "edge");
795         }
796 }
797
798 static void __init setup_IO_APIC_irqs(void)
799 {
800         struct IO_APIC_route_entry entry;
801         int apic, pin, idx, irq, first_notcon = 1, vector;
802         unsigned long flags;
803
804         apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
805
806         for (apic = 0; apic < nr_ioapics; apic++) {
807         for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
808
809                 /*
810                  * add it to the IO-APIC irq-routing table:
811                  */
812                 memset(&entry,0,sizeof(entry));
813
814                 entry.delivery_mode = INT_DELIVERY_MODE;
815                 entry.dest_mode = INT_DEST_MODE;
816                 entry.mask = 0;                         /* enable IRQ */
817                 entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
818
819                 idx = find_irq_entry(apic,pin,mp_INT);
820                 if (idx == -1) {
821                         if (first_notcon) {
822                                 apic_printk(APIC_VERBOSE, KERN_DEBUG " IO-APIC (apicid-pin) %d-%d", mp_ioapics[apic].mpc_apicid, pin);
823                                 first_notcon = 0;
824                         } else
825                                 apic_printk(APIC_VERBOSE, ", %d-%d", mp_ioapics[apic].mpc_apicid, pin);
826                         continue;
827                 }
828
829                 entry.trigger = irq_trigger(idx);
830                 entry.polarity = irq_polarity(idx);
831
832                 if (irq_trigger(idx)) {
833                         entry.trigger = 1;
834                         entry.mask = 1;
835                         entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
836                 }
837
838                 irq = pin_2_irq(idx, apic, pin);
839                 add_pin_to_irq(irq, apic, pin);
840
841                 if (!apic && !IO_APIC_IRQ(irq))
842                         continue;
843
844                 if (IO_APIC_IRQ(irq)) {
845                         cpumask_t mask;
846                         vector = assign_irq_vector(irq, TARGET_CPUS, &mask);
847                         if (vector < 0)
848                                 continue;
849
850                         entry.dest.logical.logical_dest = cpu_mask_to_apicid(mask);
851                         entry.vector = vector;
852
853                         ioapic_register_intr(irq, vector, IOAPIC_AUTO);
854                         if (!apic && (irq < 16))
855                                 disable_8259A_irq(irq);
856                 }
857                 ioapic_write_entry(apic, pin, entry);
858
859                 spin_lock_irqsave(&ioapic_lock, flags);
860                 set_native_irq_info(irq, TARGET_CPUS);
861                 spin_unlock_irqrestore(&ioapic_lock, flags);
862         }
863         }
864
865         if (!first_notcon)
866                 apic_printk(APIC_VERBOSE," not connected.\n");
867 }
868
869 /*
870  * Set up the 8259A-master output pin as broadcast to all
871  * CPUs.
872  */
873 static void __init setup_ExtINT_IRQ0_pin(unsigned int apic, unsigned int pin, int vector)
874 {
875         struct IO_APIC_route_entry entry;
876         unsigned long flags;
877
878         memset(&entry,0,sizeof(entry));
879
880         disable_8259A_irq(0);
881
882         /* mask LVT0 */
883         apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
884
885         /*
886          * We use logical delivery to get the timer IRQ
887          * to the first CPU.
888          */
889         entry.dest_mode = INT_DEST_MODE;
890         entry.mask = 0;                                 /* unmask IRQ now */
891         entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
892         entry.delivery_mode = INT_DELIVERY_MODE;
893         entry.polarity = 0;
894         entry.trigger = 0;
895         entry.vector = vector;
896
897         /*
898          * The timer IRQ doesn't have to know that behind the
899          * scene we have a 8259A-master in AEOI mode ...
900          */
901         set_irq_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq, "edge");
902
903         /*
904          * Add it to the IO-APIC irq-routing table:
905          */
906         spin_lock_irqsave(&ioapic_lock, flags);
907         io_apic_write(apic, 0x11+2*pin, *(((int *)&entry)+1));
908         io_apic_write(apic, 0x10+2*pin, *(((int *)&entry)+0));
909         spin_unlock_irqrestore(&ioapic_lock, flags);
910
911         enable_8259A_irq(0);
912 }
913
914 void __init UNEXPECTED_IO_APIC(void)
915 {
916 }
917
918 void __apicdebuginit print_IO_APIC(void)
919 {
920         int apic, i;
921         union IO_APIC_reg_00 reg_00;
922         union IO_APIC_reg_01 reg_01;
923         union IO_APIC_reg_02 reg_02;
924         unsigned long flags;
925
926         if (apic_verbosity == APIC_QUIET)
927                 return;
928
929         printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
930         for (i = 0; i < nr_ioapics; i++)
931                 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
932                        mp_ioapics[i].mpc_apicid, nr_ioapic_registers[i]);
933
934         /*
935          * We are a bit conservative about what we expect.  We have to
936          * know about every hardware change ASAP.
937          */
938         printk(KERN_INFO "testing the IO APIC.......................\n");
939
940         for (apic = 0; apic < nr_ioapics; apic++) {
941
942         spin_lock_irqsave(&ioapic_lock, flags);
943         reg_00.raw = io_apic_read(apic, 0);
944         reg_01.raw = io_apic_read(apic, 1);
945         if (reg_01.bits.version >= 0x10)
946                 reg_02.raw = io_apic_read(apic, 2);
947         spin_unlock_irqrestore(&ioapic_lock, flags);
948
949         printk("\n");
950         printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].mpc_apicid);
951         printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
952         printk(KERN_DEBUG ".......    : physical APIC id: %02X\n", reg_00.bits.ID);
953         if (reg_00.bits.__reserved_1 || reg_00.bits.__reserved_2)
954                 UNEXPECTED_IO_APIC();
955
956         printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)&reg_01);
957         printk(KERN_DEBUG ".......     : max redirection entries: %04X\n", reg_01.bits.entries);
958         if (    (reg_01.bits.entries != 0x0f) && /* older (Neptune) boards */
959                 (reg_01.bits.entries != 0x17) && /* typical ISA+PCI boards */
960                 (reg_01.bits.entries != 0x1b) && /* Compaq Proliant boards */
961                 (reg_01.bits.entries != 0x1f) && /* dual Xeon boards */
962                 (reg_01.bits.entries != 0x22) && /* bigger Xeon boards */
963                 (reg_01.bits.entries != 0x2E) &&
964                 (reg_01.bits.entries != 0x3F) &&
965                 (reg_01.bits.entries != 0x03) 
966         )
967                 UNEXPECTED_IO_APIC();
968
969         printk(KERN_DEBUG ".......     : PRQ implemented: %X\n", reg_01.bits.PRQ);
970         printk(KERN_DEBUG ".......     : IO APIC version: %04X\n", reg_01.bits.version);
971         if (    (reg_01.bits.version != 0x01) && /* 82489DX IO-APICs */
972                 (reg_01.bits.version != 0x02) && /* 82801BA IO-APICs (ICH2) */
973                 (reg_01.bits.version != 0x10) && /* oldest IO-APICs */
974                 (reg_01.bits.version != 0x11) && /* Pentium/Pro IO-APICs */
975                 (reg_01.bits.version != 0x13) && /* Xeon IO-APICs */
976                 (reg_01.bits.version != 0x20)    /* Intel P64H (82806 AA) */
977         )
978                 UNEXPECTED_IO_APIC();
979         if (reg_01.bits.__reserved_1 || reg_01.bits.__reserved_2)
980                 UNEXPECTED_IO_APIC();
981
982         if (reg_01.bits.version >= 0x10) {
983                 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
984                 printk(KERN_DEBUG ".......     : arbitration: %02X\n", reg_02.bits.arbitration);
985                 if (reg_02.bits.__reserved_1 || reg_02.bits.__reserved_2)
986                         UNEXPECTED_IO_APIC();
987         }
988
989         printk(KERN_DEBUG ".... IRQ redirection table:\n");
990
991         printk(KERN_DEBUG " NR Log Phy Mask Trig IRR Pol"
992                           " Stat Dest Deli Vect:   \n");
993
994         for (i = 0; i <= reg_01.bits.entries; i++) {
995                 struct IO_APIC_route_entry entry;
996
997                 entry = ioapic_read_entry(apic, i);
998
999                 printk(KERN_DEBUG " %02x %03X %02X  ",
1000                         i,
1001                         entry.dest.logical.logical_dest,
1002                         entry.dest.physical.physical_dest
1003                 );
1004
1005                 printk("%1d    %1d    %1d   %1d   %1d    %1d    %1d    %02X\n",
1006                         entry.mask,
1007                         entry.trigger,
1008                         entry.irr,
1009                         entry.polarity,
1010                         entry.delivery_status,
1011                         entry.dest_mode,
1012                         entry.delivery_mode,
1013                         entry.vector
1014                 );
1015         }
1016         }
1017         printk(KERN_DEBUG "IRQ to pin mappings:\n");
1018         for (i = 0; i < NR_IRQS; i++) {
1019                 struct irq_pin_list *entry = irq_2_pin + i;
1020                 if (entry->pin < 0)
1021                         continue;
1022                 printk(KERN_DEBUG "IRQ%d ", i);
1023                 for (;;) {
1024                         printk("-> %d:%d", entry->apic, entry->pin);
1025                         if (!entry->next)
1026                                 break;
1027                         entry = irq_2_pin + entry->next;
1028                 }
1029                 printk("\n");
1030         }
1031
1032         printk(KERN_INFO ".................................... done.\n");
1033
1034         return;
1035 }
1036
1037 #if 0
1038
1039 static __apicdebuginit void print_APIC_bitfield (int base)
1040 {
1041         unsigned int v;
1042         int i, j;
1043
1044         if (apic_verbosity == APIC_QUIET)
1045                 return;
1046
1047         printk(KERN_DEBUG "0123456789abcdef0123456789abcdef\n" KERN_DEBUG);
1048         for (i = 0; i < 8; i++) {
1049                 v = apic_read(base + i*0x10);
1050                 for (j = 0; j < 32; j++) {
1051                         if (v & (1<<j))
1052                                 printk("1");
1053                         else
1054                                 printk("0");
1055                 }
1056                 printk("\n");
1057         }
1058 }
1059
1060 void __apicdebuginit print_local_APIC(void * dummy)
1061 {
1062         unsigned int v, ver, maxlvt;
1063
1064         if (apic_verbosity == APIC_QUIET)
1065                 return;
1066
1067         printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1068                 smp_processor_id(), hard_smp_processor_id());
1069         v = apic_read(APIC_ID);
1070         printk(KERN_INFO "... APIC ID:      %08x (%01x)\n", v, GET_APIC_ID(v));
1071         v = apic_read(APIC_LVR);
1072         printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1073         ver = GET_APIC_VERSION(v);
1074         maxlvt = get_maxlvt();
1075
1076         v = apic_read(APIC_TASKPRI);
1077         printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1078
1079         v = apic_read(APIC_ARBPRI);
1080         printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
1081                 v & APIC_ARBPRI_MASK);
1082         v = apic_read(APIC_PROCPRI);
1083         printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1084
1085         v = apic_read(APIC_EOI);
1086         printk(KERN_DEBUG "... APIC EOI: %08x\n", v);
1087         v = apic_read(APIC_RRR);
1088         printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1089         v = apic_read(APIC_LDR);
1090         printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1091         v = apic_read(APIC_DFR);
1092         printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1093         v = apic_read(APIC_SPIV);
1094         printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1095
1096         printk(KERN_DEBUG "... APIC ISR field:\n");
1097         print_APIC_bitfield(APIC_ISR);
1098         printk(KERN_DEBUG "... APIC TMR field:\n");
1099         print_APIC_bitfield(APIC_TMR);
1100         printk(KERN_DEBUG "... APIC IRR field:\n");
1101         print_APIC_bitfield(APIC_IRR);
1102
1103         v = apic_read(APIC_ESR);
1104         printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1105
1106         v = apic_read(APIC_ICR);
1107         printk(KERN_DEBUG "... APIC ICR: %08x\n", v);
1108         v = apic_read(APIC_ICR2);
1109         printk(KERN_DEBUG "... APIC ICR2: %08x\n", v);
1110
1111         v = apic_read(APIC_LVTT);
1112         printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1113
1114         if (maxlvt > 3) {                       /* PC is LVT#4. */
1115                 v = apic_read(APIC_LVTPC);
1116                 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1117         }
1118         v = apic_read(APIC_LVT0);
1119         printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1120         v = apic_read(APIC_LVT1);
1121         printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1122
1123         if (maxlvt > 2) {                       /* ERR is LVT#3. */
1124                 v = apic_read(APIC_LVTERR);
1125                 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1126         }
1127
1128         v = apic_read(APIC_TMICT);
1129         printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1130         v = apic_read(APIC_TMCCT);
1131         printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1132         v = apic_read(APIC_TDCR);
1133         printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1134         printk("\n");
1135 }
1136
1137 void print_all_local_APICs (void)
1138 {
1139         on_each_cpu(print_local_APIC, NULL, 1, 1);
1140 }
1141
1142 void __apicdebuginit print_PIC(void)
1143 {
1144         unsigned int v;
1145         unsigned long flags;
1146
1147         if (apic_verbosity == APIC_QUIET)
1148                 return;
1149
1150         printk(KERN_DEBUG "\nprinting PIC contents\n");
1151
1152         spin_lock_irqsave(&i8259A_lock, flags);
1153
1154         v = inb(0xa1) << 8 | inb(0x21);
1155         printk(KERN_DEBUG "... PIC  IMR: %04x\n", v);
1156
1157         v = inb(0xa0) << 8 | inb(0x20);
1158         printk(KERN_DEBUG "... PIC  IRR: %04x\n", v);
1159
1160         outb(0x0b,0xa0);
1161         outb(0x0b,0x20);
1162         v = inb(0xa0) << 8 | inb(0x20);
1163         outb(0x0a,0xa0);
1164         outb(0x0a,0x20);
1165
1166         spin_unlock_irqrestore(&i8259A_lock, flags);
1167
1168         printk(KERN_DEBUG "... PIC  ISR: %04x\n", v);
1169
1170         v = inb(0x4d1) << 8 | inb(0x4d0);
1171         printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1172 }
1173
1174 #endif  /*  0  */
1175
1176 static void __init enable_IO_APIC(void)
1177 {
1178         union IO_APIC_reg_01 reg_01;
1179         int i8259_apic, i8259_pin;
1180         int i, apic;
1181         unsigned long flags;
1182
1183         for (i = 0; i < PIN_MAP_SIZE; i++) {
1184                 irq_2_pin[i].pin = -1;
1185                 irq_2_pin[i].next = 0;
1186         }
1187
1188         /*
1189          * The number of IO-APIC IRQ registers (== #pins):
1190          */
1191         for (apic = 0; apic < nr_ioapics; apic++) {
1192                 spin_lock_irqsave(&ioapic_lock, flags);
1193                 reg_01.raw = io_apic_read(apic, 1);
1194                 spin_unlock_irqrestore(&ioapic_lock, flags);
1195                 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
1196         }
1197         for(apic = 0; apic < nr_ioapics; apic++) {
1198                 int pin;
1199                 /* See if any of the pins is in ExtINT mode */
1200                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1201                         struct IO_APIC_route_entry entry;
1202                         entry = ioapic_read_entry(apic, pin);
1203
1204                         /* If the interrupt line is enabled and in ExtInt mode
1205                          * I have found the pin where the i8259 is connected.
1206                          */
1207                         if ((entry.mask == 0) && (entry.delivery_mode == dest_ExtINT)) {
1208                                 ioapic_i8259.apic = apic;
1209                                 ioapic_i8259.pin  = pin;
1210                                 goto found_i8259;
1211                         }
1212                 }
1213         }
1214  found_i8259:
1215         /* Look to see what if the MP table has reported the ExtINT */
1216         i8259_pin  = find_isa_irq_pin(0, mp_ExtINT);
1217         i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
1218         /* Trust the MP table if nothing is setup in the hardware */
1219         if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
1220                 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
1221                 ioapic_i8259.pin  = i8259_pin;
1222                 ioapic_i8259.apic = i8259_apic;
1223         }
1224         /* Complain if the MP table and the hardware disagree */
1225         if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
1226                 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
1227         {
1228                 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
1229         }
1230
1231         /*
1232          * Do not trust the IO-APIC being empty at bootup
1233          */
1234         clear_IO_APIC();
1235 }
1236
1237 /*
1238  * Not an __init, needed by the reboot code
1239  */
1240 void disable_IO_APIC(void)
1241 {
1242         /*
1243          * Clear the IO-APIC before rebooting:
1244          */
1245         clear_IO_APIC();
1246
1247         /*
1248          * If the i8259 is routed through an IOAPIC
1249          * Put that IOAPIC in virtual wire mode
1250          * so legacy interrupts can be delivered.
1251          */
1252         if (ioapic_i8259.pin != -1) {
1253                 struct IO_APIC_route_entry entry;
1254
1255                 memset(&entry, 0, sizeof(entry));
1256                 entry.mask            = 0; /* Enabled */
1257                 entry.trigger         = 0; /* Edge */
1258                 entry.irr             = 0;
1259                 entry.polarity        = 0; /* High */
1260                 entry.delivery_status = 0;
1261                 entry.dest_mode       = 0; /* Physical */
1262                 entry.delivery_mode   = dest_ExtINT; /* ExtInt */
1263                 entry.vector          = 0;
1264                 entry.dest.physical.physical_dest =
1265                                         GET_APIC_ID(apic_read(APIC_ID));
1266
1267                 /*
1268                  * Add it to the IO-APIC irq-routing table:
1269                  */
1270                 ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
1271         }
1272
1273         disconnect_bsp_APIC(ioapic_i8259.pin != -1);
1274 }
1275
1276 /*
1277  * There is a nasty bug in some older SMP boards, their mptable lies
1278  * about the timer IRQ. We do the following to work around the situation:
1279  *
1280  *      - timer IRQ defaults to IO-APIC IRQ
1281  *      - if this function detects that timer IRQs are defunct, then we fall
1282  *        back to ISA timer IRQs
1283  */
1284 static int __init timer_irq_works(void)
1285 {
1286         unsigned long t1 = jiffies;
1287
1288         local_irq_enable();
1289         /* Let ten ticks pass... */
1290         mdelay((10 * 1000) / HZ);
1291
1292         /*
1293          * Expect a few ticks at least, to be sure some possible
1294          * glue logic does not lock up after one or two first
1295          * ticks in a non-ExtINT mode.  Also the local APIC
1296          * might have cached one ExtINT interrupt.  Finally, at
1297          * least one tick may be lost due to delays.
1298          */
1299
1300         /* jiffies wrap? */
1301         if (jiffies - t1 > 4)
1302                 return 1;
1303         return 0;
1304 }
1305
1306 /*
1307  * In the SMP+IOAPIC case it might happen that there are an unspecified
1308  * number of pending IRQ events unhandled. These cases are very rare,
1309  * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
1310  * better to do it this way as thus we do not have to be aware of
1311  * 'pending' interrupts in the IRQ path, except at this point.
1312  */
1313 /*
1314  * Edge triggered needs to resend any interrupt
1315  * that was delayed but this is now handled in the device
1316  * independent code.
1317  */
1318
1319 /*
1320  * Starting up a edge-triggered IO-APIC interrupt is
1321  * nasty - we need to make sure that we get the edge.
1322  * If it is already asserted for some reason, we need
1323  * return 1 to indicate that is was pending.
1324  *
1325  * This is not complete - we should be able to fake
1326  * an edge even if it isn't on the 8259A...
1327  */
1328
1329 static unsigned int startup_ioapic_irq(unsigned int irq)
1330 {
1331         int was_pending = 0;
1332         unsigned long flags;
1333
1334         spin_lock_irqsave(&ioapic_lock, flags);
1335         if (irq < 16) {
1336                 disable_8259A_irq(irq);
1337                 if (i8259A_irq_pending(irq))
1338                         was_pending = 1;
1339         }
1340         __unmask_IO_APIC_irq(irq);
1341         spin_unlock_irqrestore(&ioapic_lock, flags);
1342
1343         return was_pending;
1344 }
1345
1346 static int ioapic_retrigger_irq(unsigned int irq)
1347 {
1348         cpumask_t mask;
1349         unsigned vector;
1350         unsigned long flags;
1351
1352         spin_lock_irqsave(&vector_lock, flags);
1353         vector = irq_vector[irq];
1354         cpus_clear(mask);
1355         cpu_set(first_cpu(irq_domain[irq]), mask);
1356
1357         send_IPI_mask(mask, vector);
1358         spin_unlock_irqrestore(&vector_lock, flags);
1359
1360         return 1;
1361 }
1362
1363 /*
1364  * Level and edge triggered IO-APIC interrupts need different handling,
1365  * so we use two separate IRQ descriptors. Edge triggered IRQs can be
1366  * handled with the level-triggered descriptor, but that one has slightly
1367  * more overhead. Level-triggered interrupts cannot be handled with the
1368  * edge-triggered handler, without risking IRQ storms and other ugly
1369  * races.
1370  */
1371
1372 static void ack_apic_edge(unsigned int irq)
1373 {
1374         move_native_irq(irq);
1375         ack_APIC_irq();
1376 }
1377
1378 static void ack_apic_level(unsigned int irq)
1379 {
1380         int do_unmask_irq = 0;
1381
1382 #if defined(CONFIG_GENERIC_PENDING_IRQ) || defined(CONFIG_IRQBALANCE)
1383         /* If we are moving the irq we need to mask it */
1384         if (unlikely(irq_desc[irq].status & IRQ_MOVE_PENDING)) {
1385                 do_unmask_irq = 1;
1386                 mask_IO_APIC_irq(irq);
1387         }
1388 #endif
1389
1390         /*
1391          * We must acknowledge the irq before we move it or the acknowledge will
1392          * not propogate properly.
1393          */
1394         ack_APIC_irq();
1395
1396         /* Now we can move and renable the irq */
1397         move_masked_irq(irq);
1398         if (unlikely(do_unmask_irq))
1399                 unmask_IO_APIC_irq(irq);
1400 }
1401
1402 static struct irq_chip ioapic_chip __read_mostly = {
1403         .name           = "IO-APIC",
1404         .startup        = startup_ioapic_irq,
1405         .mask           = mask_IO_APIC_irq,
1406         .unmask         = unmask_IO_APIC_irq,
1407         .ack            = ack_apic_edge,
1408         .eoi            = ack_apic_level,
1409 #ifdef CONFIG_SMP
1410         .set_affinity   = set_ioapic_affinity_irq,
1411 #endif
1412         .retrigger      = ioapic_retrigger_irq,
1413 };
1414
1415 static inline void init_IO_APIC_traps(void)
1416 {
1417         int irq;
1418
1419         /*
1420          * NOTE! The local APIC isn't very good at handling
1421          * multiple interrupts at the same interrupt level.
1422          * As the interrupt level is determined by taking the
1423          * vector number and shifting that right by 4, we
1424          * want to spread these out a bit so that they don't
1425          * all fall in the same interrupt level.
1426          *
1427          * Also, we've got to be careful not to trash gate
1428          * 0x80, because int 0x80 is hm, kind of importantish. ;)
1429          */
1430         for (irq = 0; irq < NR_IRQS ; irq++) {
1431                 int tmp = irq;
1432                 if (IO_APIC_IRQ(tmp) && !irq_vector[tmp]) {
1433                         /*
1434                          * Hmm.. We don't have an entry for this,
1435                          * so default to an old-fashioned 8259
1436                          * interrupt if we can..
1437                          */
1438                         if (irq < 16)
1439                                 make_8259A_irq(irq);
1440                         else
1441                                 /* Strange. Oh, well.. */
1442                                 irq_desc[irq].chip = &no_irq_chip;
1443                 }
1444         }
1445 }
1446
1447 static void enable_lapic_irq (unsigned int irq)
1448 {
1449         unsigned long v;
1450
1451         v = apic_read(APIC_LVT0);
1452         apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
1453 }
1454
1455 static void disable_lapic_irq (unsigned int irq)
1456 {
1457         unsigned long v;
1458
1459         v = apic_read(APIC_LVT0);
1460         apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1461 }
1462
1463 static void ack_lapic_irq (unsigned int irq)
1464 {
1465         ack_APIC_irq();
1466 }
1467
1468 static void end_lapic_irq (unsigned int i) { /* nothing */ }
1469
1470 static struct hw_interrupt_type lapic_irq_type __read_mostly = {
1471         .typename = "local-APIC-edge",
1472         .startup = NULL, /* startup_irq() not used for IRQ0 */
1473         .shutdown = NULL, /* shutdown_irq() not used for IRQ0 */
1474         .enable = enable_lapic_irq,
1475         .disable = disable_lapic_irq,
1476         .ack = ack_lapic_irq,
1477         .end = end_lapic_irq,
1478 };
1479
1480 static void setup_nmi (void)
1481 {
1482         /*
1483          * Dirty trick to enable the NMI watchdog ...
1484          * We put the 8259A master into AEOI mode and
1485          * unmask on all local APICs LVT0 as NMI.
1486          *
1487          * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
1488          * is from Maciej W. Rozycki - so we do not have to EOI from
1489          * the NMI handler or the timer interrupt.
1490          */ 
1491         printk(KERN_INFO "activating NMI Watchdog ...");
1492
1493         enable_NMI_through_LVT0(NULL);
1494
1495         printk(" done.\n");
1496 }
1497
1498 /*
1499  * This looks a bit hackish but it's about the only one way of sending
1500  * a few INTA cycles to 8259As and any associated glue logic.  ICR does
1501  * not support the ExtINT mode, unfortunately.  We need to send these
1502  * cycles as some i82489DX-based boards have glue logic that keeps the
1503  * 8259A interrupt line asserted until INTA.  --macro
1504  */
1505 static inline void unlock_ExtINT_logic(void)
1506 {
1507         int apic, pin, i;
1508         struct IO_APIC_route_entry entry0, entry1;
1509         unsigned char save_control, save_freq_select;
1510         unsigned long flags;
1511
1512         pin  = find_isa_irq_pin(8, mp_INT);
1513         apic = find_isa_irq_apic(8, mp_INT);
1514         if (pin == -1)
1515                 return;
1516
1517         spin_lock_irqsave(&ioapic_lock, flags);
1518         *(((int *)&entry0) + 1) = io_apic_read(apic, 0x11 + 2 * pin);
1519         *(((int *)&entry0) + 0) = io_apic_read(apic, 0x10 + 2 * pin);
1520         spin_unlock_irqrestore(&ioapic_lock, flags);
1521         clear_IO_APIC_pin(apic, pin);
1522
1523         memset(&entry1, 0, sizeof(entry1));
1524
1525         entry1.dest_mode = 0;                   /* physical delivery */
1526         entry1.mask = 0;                        /* unmask IRQ now */
1527         entry1.dest.physical.physical_dest = hard_smp_processor_id();
1528         entry1.delivery_mode = dest_ExtINT;
1529         entry1.polarity = entry0.polarity;
1530         entry1.trigger = 0;
1531         entry1.vector = 0;
1532
1533         spin_lock_irqsave(&ioapic_lock, flags);
1534         io_apic_write(apic, 0x11 + 2 * pin, *(((int *)&entry1) + 1));
1535         io_apic_write(apic, 0x10 + 2 * pin, *(((int *)&entry1) + 0));
1536         spin_unlock_irqrestore(&ioapic_lock, flags);
1537
1538         save_control = CMOS_READ(RTC_CONTROL);
1539         save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
1540         CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
1541                    RTC_FREQ_SELECT);
1542         CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
1543
1544         i = 100;
1545         while (i-- > 0) {
1546                 mdelay(10);
1547                 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
1548                         i -= 10;
1549         }
1550
1551         CMOS_WRITE(save_control, RTC_CONTROL);
1552         CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
1553         clear_IO_APIC_pin(apic, pin);
1554
1555         spin_lock_irqsave(&ioapic_lock, flags);
1556         io_apic_write(apic, 0x11 + 2 * pin, *(((int *)&entry0) + 1));
1557         io_apic_write(apic, 0x10 + 2 * pin, *(((int *)&entry0) + 0));
1558         spin_unlock_irqrestore(&ioapic_lock, flags);
1559 }
1560
1561 /*
1562  * This code may look a bit paranoid, but it's supposed to cooperate with
1563  * a wide range of boards and BIOS bugs.  Fortunately only the timer IRQ
1564  * is so screwy.  Thanks to Brian Perkins for testing/hacking this beast
1565  * fanatically on his truly buggy board.
1566  *
1567  * FIXME: really need to revamp this for modern platforms only.
1568  */
1569 static inline void check_timer(void)
1570 {
1571         int apic1, pin1, apic2, pin2;
1572         int vector;
1573         cpumask_t mask;
1574
1575         /*
1576          * get/set the timer IRQ vector:
1577          */
1578         disable_8259A_irq(0);
1579         vector = assign_irq_vector(0, TARGET_CPUS, &mask);
1580
1581         /*
1582          * Subtle, code in do_timer_interrupt() expects an AEOI
1583          * mode for the 8259A whenever interrupts are routed
1584          * through I/O APICs.  Also IRQ0 has to be enabled in
1585          * the 8259A which implies the virtual wire has to be
1586          * disabled in the local APIC.
1587          */
1588         apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
1589         init_8259A(1);
1590         if (timer_over_8254 > 0)
1591                 enable_8259A_irq(0);
1592
1593         pin1  = find_isa_irq_pin(0, mp_INT);
1594         apic1 = find_isa_irq_apic(0, mp_INT);
1595         pin2  = ioapic_i8259.pin;
1596         apic2 = ioapic_i8259.apic;
1597
1598         apic_printk(APIC_VERBOSE,KERN_INFO "..TIMER: vector=0x%02X apic1=%d pin1=%d apic2=%d pin2=%d\n",
1599                 vector, apic1, pin1, apic2, pin2);
1600
1601         if (pin1 != -1) {
1602                 /*
1603                  * Ok, does IRQ0 through the IOAPIC work?
1604                  */
1605                 unmask_IO_APIC_irq(0);
1606                 if (!no_timer_check && timer_irq_works()) {
1607                         nmi_watchdog_default();
1608                         if (nmi_watchdog == NMI_IO_APIC) {
1609                                 disable_8259A_irq(0);
1610                                 setup_nmi();
1611                                 enable_8259A_irq(0);
1612                         }
1613                         if (disable_timer_pin_1 > 0)
1614                                 clear_IO_APIC_pin(0, pin1);
1615                         return;
1616                 }
1617                 clear_IO_APIC_pin(apic1, pin1);
1618                 apic_printk(APIC_QUIET,KERN_ERR "..MP-BIOS bug: 8254 timer not "
1619                                 "connected to IO-APIC\n");
1620         }
1621
1622         apic_printk(APIC_VERBOSE,KERN_INFO "...trying to set up timer (IRQ0) "
1623                                 "through the 8259A ... ");
1624         if (pin2 != -1) {
1625                 apic_printk(APIC_VERBOSE,"\n..... (found apic %d pin %d) ...",
1626                         apic2, pin2);
1627                 /*
1628                  * legacy devices should be connected to IO APIC #0
1629                  */
1630                 setup_ExtINT_IRQ0_pin(apic2, pin2, vector);
1631                 if (timer_irq_works()) {
1632                         apic_printk(APIC_VERBOSE," works.\n");
1633                         nmi_watchdog_default();
1634                         if (nmi_watchdog == NMI_IO_APIC) {
1635                                 setup_nmi();
1636                         }
1637                         return;
1638                 }
1639                 /*
1640                  * Cleanup, just in case ...
1641                  */
1642                 clear_IO_APIC_pin(apic2, pin2);
1643         }
1644         apic_printk(APIC_VERBOSE," failed.\n");
1645
1646         if (nmi_watchdog == NMI_IO_APIC) {
1647                 printk(KERN_WARNING "timer doesn't work through the IO-APIC - disabling NMI Watchdog!\n");
1648                 nmi_watchdog = 0;
1649         }
1650
1651         apic_printk(APIC_VERBOSE, KERN_INFO "...trying to set up timer as Virtual Wire IRQ...");
1652
1653         disable_8259A_irq(0);
1654         irq_desc[0].chip = &lapic_irq_type;
1655         apic_write(APIC_LVT0, APIC_DM_FIXED | vector);  /* Fixed mode */
1656         enable_8259A_irq(0);
1657
1658         if (timer_irq_works()) {
1659                 apic_printk(APIC_VERBOSE," works.\n");
1660                 return;
1661         }
1662         apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | vector);
1663         apic_printk(APIC_VERBOSE," failed.\n");
1664
1665         apic_printk(APIC_VERBOSE, KERN_INFO "...trying to set up timer as ExtINT IRQ...");
1666
1667         init_8259A(0);
1668         make_8259A_irq(0);
1669         apic_write(APIC_LVT0, APIC_DM_EXTINT);
1670
1671         unlock_ExtINT_logic();
1672
1673         if (timer_irq_works()) {
1674                 apic_printk(APIC_VERBOSE," works.\n");
1675                 return;
1676         }
1677         apic_printk(APIC_VERBOSE," failed :(.\n");
1678         panic("IO-APIC + timer doesn't work! Try using the 'noapic' kernel parameter\n");
1679 }
1680
1681 static int __init notimercheck(char *s)
1682 {
1683         no_timer_check = 1;
1684         return 1;
1685 }
1686 __setup("no_timer_check", notimercheck);
1687
1688 /*
1689  *
1690  * IRQ's that are handled by the PIC in the MPS IOAPIC case.
1691  * - IRQ2 is the cascade IRQ, and cannot be a io-apic IRQ.
1692  *   Linux doesn't really care, as it's not actually used
1693  *   for any interrupt handling anyway.
1694  */
1695 #define PIC_IRQS        (1<<2)
1696
1697 void __init setup_IO_APIC(void)
1698 {
1699         enable_IO_APIC();
1700
1701         if (acpi_ioapic)
1702                 io_apic_irqs = ~0;      /* all IRQs go through IOAPIC */
1703         else
1704                 io_apic_irqs = ~PIC_IRQS;
1705
1706         apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
1707
1708         sync_Arb_IDs();
1709         setup_IO_APIC_irqs();
1710         init_IO_APIC_traps();
1711         check_timer();
1712         if (!acpi_ioapic)
1713                 print_IO_APIC();
1714 }
1715
1716 struct sysfs_ioapic_data {
1717         struct sys_device dev;
1718         struct IO_APIC_route_entry entry[0];
1719 };
1720 static struct sysfs_ioapic_data * mp_ioapic_data[MAX_IO_APICS];
1721
1722 static int ioapic_suspend(struct sys_device *dev, pm_message_t state)
1723 {
1724         struct IO_APIC_route_entry *entry;
1725         struct sysfs_ioapic_data *data;
1726         int i;
1727
1728         data = container_of(dev, struct sysfs_ioapic_data, dev);
1729         entry = data->entry;
1730         for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ )
1731                 *entry = ioapic_read_entry(dev->id, i);
1732
1733         return 0;
1734 }
1735
1736 static int ioapic_resume(struct sys_device *dev)
1737 {
1738         struct IO_APIC_route_entry *entry;
1739         struct sysfs_ioapic_data *data;
1740         unsigned long flags;
1741         union IO_APIC_reg_00 reg_00;
1742         int i;
1743
1744         data = container_of(dev, struct sysfs_ioapic_data, dev);
1745         entry = data->entry;
1746
1747         spin_lock_irqsave(&ioapic_lock, flags);
1748         reg_00.raw = io_apic_read(dev->id, 0);
1749         if (reg_00.bits.ID != mp_ioapics[dev->id].mpc_apicid) {
1750                 reg_00.bits.ID = mp_ioapics[dev->id].mpc_apicid;
1751                 io_apic_write(dev->id, 0, reg_00.raw);
1752         }
1753         spin_unlock_irqrestore(&ioapic_lock, flags);
1754         for (i = 0; i < nr_ioapic_registers[dev->id]; i++)
1755                 ioapic_write_entry(dev->id, i, entry[i]);
1756
1757         return 0;
1758 }
1759
1760 static struct sysdev_class ioapic_sysdev_class = {
1761         set_kset_name("ioapic"),
1762         .suspend = ioapic_suspend,
1763         .resume = ioapic_resume,
1764 };
1765
1766 static int __init ioapic_init_sysfs(void)
1767 {
1768         struct sys_device * dev;
1769         int i, size, error = 0;
1770
1771         error = sysdev_class_register(&ioapic_sysdev_class);
1772         if (error)
1773                 return error;
1774
1775         for (i = 0; i < nr_ioapics; i++ ) {
1776                 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
1777                         * sizeof(struct IO_APIC_route_entry);
1778                 mp_ioapic_data[i] = kmalloc(size, GFP_KERNEL);
1779                 if (!mp_ioapic_data[i]) {
1780                         printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
1781                         continue;
1782                 }
1783                 memset(mp_ioapic_data[i], 0, size);
1784                 dev = &mp_ioapic_data[i]->dev;
1785                 dev->id = i;
1786                 dev->cls = &ioapic_sysdev_class;
1787                 error = sysdev_register(dev);
1788                 if (error) {
1789                         kfree(mp_ioapic_data[i]);
1790                         mp_ioapic_data[i] = NULL;
1791                         printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
1792                         continue;
1793                 }
1794         }
1795
1796         return 0;
1797 }
1798
1799 device_initcall(ioapic_init_sysfs);
1800
1801 /*
1802  * Dynamic irq allocate and deallocation
1803  */
1804 int create_irq(void)
1805 {
1806         /* Allocate an unused irq */
1807         int irq;
1808         int new;
1809         int vector = 0;
1810         unsigned long flags;
1811         cpumask_t mask;
1812
1813         irq = -ENOSPC;
1814         spin_lock_irqsave(&vector_lock, flags);
1815         for (new = (NR_IRQS - 1); new >= 0; new--) {
1816                 if (platform_legacy_irq(new))
1817                         continue;
1818                 if (irq_vector[new] != 0)
1819                         continue;
1820                 vector = __assign_irq_vector(new, TARGET_CPUS, &mask);
1821                 if (likely(vector > 0))
1822                         irq = new;
1823                 break;
1824         }
1825         spin_unlock_irqrestore(&vector_lock, flags);
1826
1827         if (irq >= 0) {
1828                 dynamic_irq_init(irq);
1829         }
1830         return irq;
1831 }
1832
1833 void destroy_irq(unsigned int irq)
1834 {
1835         unsigned long flags;
1836
1837         dynamic_irq_cleanup(irq);
1838
1839         spin_lock_irqsave(&vector_lock, flags);
1840         irq_vector[irq] = 0;
1841         spin_unlock_irqrestore(&vector_lock, flags);
1842 }
1843
1844 /*
1845  * MSI mesage composition
1846  */
1847 #ifdef CONFIG_PCI_MSI
1848 static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq, struct msi_msg *msg)
1849 {
1850         int vector;
1851         unsigned dest;
1852         cpumask_t tmp;
1853
1854         vector = assign_irq_vector(irq, TARGET_CPUS, &tmp);
1855         if (vector >= 0) {
1856                 dest = cpu_mask_to_apicid(tmp);
1857
1858                 msg->address_hi = MSI_ADDR_BASE_HI;
1859                 msg->address_lo =
1860                         MSI_ADDR_BASE_LO |
1861                         ((INT_DEST_MODE == 0) ?
1862                                 MSI_ADDR_DEST_MODE_PHYSICAL:
1863                                 MSI_ADDR_DEST_MODE_LOGICAL) |
1864                         ((INT_DELIVERY_MODE != dest_LowestPrio) ?
1865                                 MSI_ADDR_REDIRECTION_CPU:
1866                                 MSI_ADDR_REDIRECTION_LOWPRI) |
1867                         MSI_ADDR_DEST_ID(dest);
1868
1869                 msg->data =
1870                         MSI_DATA_TRIGGER_EDGE |
1871                         MSI_DATA_LEVEL_ASSERT |
1872                         ((INT_DELIVERY_MODE != dest_LowestPrio) ?
1873                                 MSI_DATA_DELIVERY_FIXED:
1874                                 MSI_DATA_DELIVERY_LOWPRI) |
1875                         MSI_DATA_VECTOR(vector);
1876         }
1877         return vector;
1878 }
1879
1880 #ifdef CONFIG_SMP
1881 static void set_msi_irq_affinity(unsigned int irq, cpumask_t mask)
1882 {
1883         struct msi_msg msg;
1884         unsigned int dest;
1885         cpumask_t tmp;
1886         int vector;
1887
1888         cpus_and(tmp, mask, cpu_online_map);
1889         if (cpus_empty(tmp))
1890                 tmp = TARGET_CPUS;
1891
1892         cpus_and(mask, tmp, CPU_MASK_ALL);
1893
1894         vector = assign_irq_vector(irq, mask, &tmp);
1895         if (vector < 0)
1896                 return;
1897
1898         dest = cpu_mask_to_apicid(tmp);
1899
1900         read_msi_msg(irq, &msg);
1901
1902         msg.data &= ~MSI_DATA_VECTOR_MASK;
1903         msg.data |= MSI_DATA_VECTOR(vector);
1904         msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
1905         msg.address_lo |= MSI_ADDR_DEST_ID(dest);
1906
1907         write_msi_msg(irq, &msg);
1908         set_native_irq_info(irq, mask);
1909 }
1910 #endif /* CONFIG_SMP */
1911
1912 /*
1913  * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
1914  * which implement the MSI or MSI-X Capability Structure.
1915  */
1916 static struct irq_chip msi_chip = {
1917         .name           = "PCI-MSI",
1918         .unmask         = unmask_msi_irq,
1919         .mask           = mask_msi_irq,
1920         .ack            = ack_apic_edge,
1921 #ifdef CONFIG_SMP
1922         .set_affinity   = set_msi_irq_affinity,
1923 #endif
1924         .retrigger      = ioapic_retrigger_irq,
1925 };
1926
1927 int arch_setup_msi_irq(unsigned int irq, struct pci_dev *dev)
1928 {
1929         struct msi_msg msg;
1930         int ret;
1931         ret = msi_compose_msg(dev, irq, &msg);
1932         if (ret < 0)
1933                 return ret;
1934
1935         write_msi_msg(irq, &msg);
1936
1937         set_irq_chip_and_handler_name(irq, &msi_chip, handle_edge_irq, "edge");
1938
1939         return 0;
1940 }
1941
1942 void arch_teardown_msi_irq(unsigned int irq)
1943 {
1944         return;
1945 }
1946
1947 #endif /* CONFIG_PCI_MSI */
1948
1949 /*
1950  * Hypertransport interrupt support
1951  */
1952 #ifdef CONFIG_HT_IRQ
1953
1954 #ifdef CONFIG_SMP
1955
1956 static void target_ht_irq(unsigned int irq, unsigned int dest, u8 vector)
1957 {
1958         struct ht_irq_msg msg;
1959         fetch_ht_irq_msg(irq, &msg);
1960
1961         msg.address_lo &= ~(HT_IRQ_LOW_VECTOR_MASK | HT_IRQ_LOW_DEST_ID_MASK);
1962         msg.address_hi &= ~(HT_IRQ_HIGH_DEST_ID_MASK);
1963
1964         msg.address_lo |= HT_IRQ_LOW_VECTOR(vector) | HT_IRQ_LOW_DEST_ID(dest);
1965         msg.address_hi |= HT_IRQ_HIGH_DEST_ID(dest);
1966
1967         write_ht_irq_msg(irq, &msg);
1968 }
1969
1970 static void set_ht_irq_affinity(unsigned int irq, cpumask_t mask)
1971 {
1972         unsigned int dest;
1973         cpumask_t tmp;
1974         int vector;
1975
1976         cpus_and(tmp, mask, cpu_online_map);
1977         if (cpus_empty(tmp))
1978                 tmp = TARGET_CPUS;
1979
1980         cpus_and(mask, tmp, CPU_MASK_ALL);
1981
1982         vector = assign_irq_vector(irq, mask, &tmp);
1983         if (vector < 0)
1984                 return;
1985
1986         dest = cpu_mask_to_apicid(tmp);
1987
1988         target_ht_irq(irq, dest, vector);
1989         set_native_irq_info(irq, mask);
1990 }
1991 #endif
1992
1993 static struct irq_chip ht_irq_chip = {
1994         .name           = "PCI-HT",
1995         .mask           = mask_ht_irq,
1996         .unmask         = unmask_ht_irq,
1997         .ack            = ack_apic_edge,
1998 #ifdef CONFIG_SMP
1999         .set_affinity   = set_ht_irq_affinity,
2000 #endif
2001         .retrigger      = ioapic_retrigger_irq,
2002 };
2003
2004 int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev)
2005 {
2006         int vector;
2007         cpumask_t tmp;
2008
2009         vector = assign_irq_vector(irq, TARGET_CPUS, &tmp);
2010         if (vector >= 0) {
2011                 struct ht_irq_msg msg;
2012                 unsigned dest;
2013
2014                 dest = cpu_mask_to_apicid(tmp);
2015
2016                 msg.address_hi = HT_IRQ_HIGH_DEST_ID(dest);
2017
2018                 msg.address_lo =
2019                         HT_IRQ_LOW_BASE |
2020                         HT_IRQ_LOW_DEST_ID(dest) |
2021                         HT_IRQ_LOW_VECTOR(vector) |
2022                         ((INT_DEST_MODE == 0) ?
2023                                 HT_IRQ_LOW_DM_PHYSICAL :
2024                                 HT_IRQ_LOW_DM_LOGICAL) |
2025                         HT_IRQ_LOW_RQEOI_EDGE |
2026                         ((INT_DELIVERY_MODE != dest_LowestPrio) ?
2027                                 HT_IRQ_LOW_MT_FIXED :
2028                                 HT_IRQ_LOW_MT_ARBITRATED) |
2029                         HT_IRQ_LOW_IRQ_MASKED;
2030
2031                 write_ht_irq_msg(irq, &msg);
2032
2033                 set_irq_chip_and_handler_name(irq, &ht_irq_chip,
2034                                               handle_edge_irq, "edge");
2035         }
2036         return vector;
2037 }
2038 #endif /* CONFIG_HT_IRQ */
2039
2040 /* --------------------------------------------------------------------------
2041                           ACPI-based IOAPIC Configuration
2042    -------------------------------------------------------------------------- */
2043
2044 #ifdef CONFIG_ACPI
2045
2046 #define IO_APIC_MAX_ID          0xFE
2047
2048 int __init io_apic_get_redir_entries (int ioapic)
2049 {
2050         union IO_APIC_reg_01    reg_01;
2051         unsigned long flags;
2052
2053         spin_lock_irqsave(&ioapic_lock, flags);
2054         reg_01.raw = io_apic_read(ioapic, 1);
2055         spin_unlock_irqrestore(&ioapic_lock, flags);
2056
2057         return reg_01.bits.entries;
2058 }
2059
2060
2061 int io_apic_set_pci_routing (int ioapic, int pin, int irq, int triggering, int polarity)
2062 {
2063         struct IO_APIC_route_entry entry;
2064         unsigned long flags;
2065         int vector;
2066         cpumask_t mask;
2067
2068         if (!IO_APIC_IRQ(irq)) {
2069                 apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
2070                         ioapic);
2071                 return -EINVAL;
2072         }
2073
2074         /*
2075          * IRQs < 16 are already in the irq_2_pin[] map
2076          */
2077         if (irq >= 16)
2078                 add_pin_to_irq(irq, ioapic, pin);
2079
2080
2081         vector = assign_irq_vector(irq, TARGET_CPUS, &mask);
2082         if (vector < 0)
2083                 return vector;
2084
2085         /*
2086          * Generate a PCI IRQ routing entry and program the IOAPIC accordingly.
2087          * Note that we mask (disable) IRQs now -- these get enabled when the
2088          * corresponding device driver registers for this IRQ.
2089          */
2090
2091         memset(&entry,0,sizeof(entry));
2092
2093         entry.delivery_mode = INT_DELIVERY_MODE;
2094         entry.dest_mode = INT_DEST_MODE;
2095         entry.dest.logical.logical_dest = cpu_mask_to_apicid(mask);
2096         entry.trigger = triggering;
2097         entry.polarity = polarity;
2098         entry.mask = 1;                                  /* Disabled (masked) */
2099         entry.vector = vector & 0xff;
2100
2101         apic_printk(APIC_VERBOSE,KERN_DEBUG "IOAPIC[%d]: Set PCI routing entry (%d-%d -> 0x%x -> "
2102                 "IRQ %d Mode:%i Active:%i)\n", ioapic, 
2103                mp_ioapics[ioapic].mpc_apicid, pin, entry.vector, irq,
2104                triggering, polarity);
2105
2106         ioapic_register_intr(irq, entry.vector, triggering);
2107
2108         if (!ioapic && (irq < 16))
2109                 disable_8259A_irq(irq);
2110
2111         ioapic_write_entry(ioapic, pin, entry);
2112
2113         spin_lock_irqsave(&ioapic_lock, flags);
2114         set_native_irq_info(irq, TARGET_CPUS);
2115         spin_unlock_irqrestore(&ioapic_lock, flags);
2116
2117         return 0;
2118 }
2119
2120 #endif /* CONFIG_ACPI */
2121
2122
2123 /*
2124  * This function currently is only a helper for the i386 smp boot process where
2125  * we need to reprogram the ioredtbls to cater for the cpus which have come online
2126  * so mask in all cases should simply be TARGET_CPUS
2127  */
2128 #ifdef CONFIG_SMP
2129 void __init setup_ioapic_dest(void)
2130 {
2131         int pin, ioapic, irq, irq_entry;
2132
2133         if (skip_ioapic_setup == 1)
2134                 return;
2135
2136         for (ioapic = 0; ioapic < nr_ioapics; ioapic++) {
2137                 for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
2138                         irq_entry = find_irq_entry(ioapic, pin, mp_INT);
2139                         if (irq_entry == -1)
2140                                 continue;
2141                         irq = pin_2_irq(irq_entry, ioapic, pin);
2142                         set_ioapic_affinity_irq(irq, TARGET_CPUS);
2143                 }
2144
2145         }
2146 }
2147 #endif