Automatic merge with /usr/src/ntfs-2.6.git
[pandora-kernel.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/irq.h>
25 #include <linux/interrupt.h>
26 #include <linux/init.h>
27 #include <linux/delay.h>
28 #include <linux/sched.h>
29 #include <linux/config.h>
30 #include <linux/smp_lock.h>
31 #include <linux/mc146818rtc.h>
32 #include <linux/acpi.h>
33 #include <linux/sysdev.h>
34
35 #include <asm/io.h>
36 #include <asm/smp.h>
37 #include <asm/desc.h>
38 #include <asm/proto.h>
39 #include <asm/mach_apic.h>
40 #include <asm/acpi.h>
41
42 #define __apicdebuginit  __init
43
44 int sis_apic_bug; /* not actually supported, dummy for compile */
45
46 static int no_timer_check;
47
48 static DEFINE_SPINLOCK(ioapic_lock);
49
50 /*
51  * # of IRQ routing registers
52  */
53 int nr_ioapic_registers[MAX_IO_APICS];
54
55 /*
56  * Rough estimation of how many shared IRQs there are, can
57  * be changed anytime.
58  */
59 #define MAX_PLUS_SHARED_IRQS NR_IRQS
60 #define PIN_MAP_SIZE (MAX_PLUS_SHARED_IRQS + NR_IRQS)
61
62 /*
63  * This is performance-critical, we want to do it O(1)
64  *
65  * the indexing order of this array favors 1:1 mappings
66  * between pins and IRQs.
67  */
68
69 static struct irq_pin_list {
70         short apic, pin, next;
71 } irq_2_pin[PIN_MAP_SIZE];
72
73 int vector_irq[NR_VECTORS] = { [0 ... NR_VECTORS - 1] = -1};
74 #ifdef CONFIG_PCI_MSI
75 #define vector_to_irq(vector)   \
76         (platform_legacy_irq(vector) ? vector : vector_irq[vector])
77 #else
78 #define vector_to_irq(vector)   (vector)
79 #endif
80
81 /*
82  * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
83  * shared ISA-space IRQs, so we have to support them. We are super
84  * fast in the common case, and fast for shared ISA-space IRQs.
85  */
86 static void add_pin_to_irq(unsigned int irq, int apic, int pin)
87 {
88         static int first_free_entry = NR_IRQS;
89         struct irq_pin_list *entry = irq_2_pin + irq;
90
91         while (entry->next)
92                 entry = irq_2_pin + entry->next;
93
94         if (entry->pin != -1) {
95                 entry->next = first_free_entry;
96                 entry = irq_2_pin + entry->next;
97                 if (++first_free_entry >= PIN_MAP_SIZE)
98                         panic("io_apic.c: whoops");
99         }
100         entry->apic = apic;
101         entry->pin = pin;
102 }
103
104 #define __DO_ACTION(R, ACTION, FINAL)                                   \
105                                                                         \
106 {                                                                       \
107         int pin;                                                        \
108         struct irq_pin_list *entry = irq_2_pin + irq;                   \
109                                                                         \
110         for (;;) {                                                      \
111                 unsigned int reg;                                       \
112                 pin = entry->pin;                                       \
113                 if (pin == -1)                                          \
114                         break;                                          \
115                 reg = io_apic_read(entry->apic, 0x10 + R + pin*2);      \
116                 reg ACTION;                                             \
117                 io_apic_modify(entry->apic, reg);                       \
118                 if (!entry->next)                                       \
119                         break;                                          \
120                 entry = irq_2_pin + entry->next;                        \
121         }                                                               \
122         FINAL;                                                          \
123 }
124
125 #define DO_ACTION(name,R,ACTION, FINAL)                                 \
126                                                                         \
127         static void name##_IO_APIC_irq (unsigned int irq)               \
128         __DO_ACTION(R, ACTION, FINAL)
129
130 DO_ACTION( __mask,             0, |= 0x00010000, io_apic_sync(entry->apic) )
131                                                 /* mask = 1 */
132 DO_ACTION( __unmask,           0, &= 0xfffeffff, )
133                                                 /* mask = 0 */
134
135 static void mask_IO_APIC_irq (unsigned int irq)
136 {
137         unsigned long flags;
138
139         spin_lock_irqsave(&ioapic_lock, flags);
140         __mask_IO_APIC_irq(irq);
141         spin_unlock_irqrestore(&ioapic_lock, flags);
142 }
143
144 static void unmask_IO_APIC_irq (unsigned int irq)
145 {
146         unsigned long flags;
147
148         spin_lock_irqsave(&ioapic_lock, flags);
149         __unmask_IO_APIC_irq(irq);
150         spin_unlock_irqrestore(&ioapic_lock, flags);
151 }
152
153 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
154 {
155         struct IO_APIC_route_entry entry;
156         unsigned long flags;
157
158         /* Check delivery_mode to be sure we're not clearing an SMI pin */
159         spin_lock_irqsave(&ioapic_lock, flags);
160         *(((int*)&entry) + 0) = io_apic_read(apic, 0x10 + 2 * pin);
161         *(((int*)&entry) + 1) = io_apic_read(apic, 0x11 + 2 * pin);
162         spin_unlock_irqrestore(&ioapic_lock, flags);
163         if (entry.delivery_mode == dest_SMI)
164                 return;
165         /*
166          * Disable it in the IO-APIC irq-routing table:
167          */
168         memset(&entry, 0, sizeof(entry));
169         entry.mask = 1;
170         spin_lock_irqsave(&ioapic_lock, flags);
171         io_apic_write(apic, 0x10 + 2 * pin, *(((int *)&entry) + 0));
172         io_apic_write(apic, 0x11 + 2 * pin, *(((int *)&entry) + 1));
173         spin_unlock_irqrestore(&ioapic_lock, flags);
174 }
175
176 static void clear_IO_APIC (void)
177 {
178         int apic, pin;
179
180         for (apic = 0; apic < nr_ioapics; apic++)
181                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
182                         clear_IO_APIC_pin(apic, pin);
183 }
184
185 /*
186  * support for broken MP BIOSs, enables hand-redirection of PIRQ0-7 to
187  * specific CPU-side IRQs.
188  */
189
190 #define MAX_PIRQS 8
191 static int pirq_entries [MAX_PIRQS];
192 static int pirqs_enabled;
193 int skip_ioapic_setup;
194 int ioapic_force;
195
196 /* dummy parsing: see setup.c */
197
198 static int __init disable_ioapic_setup(char *str)
199 {
200         skip_ioapic_setup = 1;
201         return 1;
202 }
203
204 static int __init enable_ioapic_setup(char *str)
205 {
206         ioapic_force = 1;
207         skip_ioapic_setup = 0;
208         return 1;
209 }
210
211 __setup("noapic", disable_ioapic_setup);
212 __setup("apic", enable_ioapic_setup);
213
214 #include <asm/pci-direct.h>
215 #include <linux/pci_ids.h>
216 #include <linux/pci.h>
217
218 /* Temporary Hack. Nvidia and VIA boards currently only work with IO-APIC
219    off. Check for an Nvidia or VIA PCI bridge and turn it off.
220    Use pci direct infrastructure because this runs before the PCI subsystem. 
221
222    Can be overwritten with "apic"
223
224    And another hack to disable the IOMMU on VIA chipsets.
225
226    Kludge-O-Rama. */
227 void __init check_ioapic(void) 
228
229         int num,slot,func; 
230         if (ioapic_force) 
231                 return; 
232
233         /* Poor man's PCI discovery */
234         for (num = 0; num < 32; num++) { 
235                 for (slot = 0; slot < 32; slot++) { 
236                         for (func = 0; func < 8; func++) { 
237                                 u32 class;
238                                 u32 vendor;
239                                 u8 type;
240                                 class = read_pci_config(num,slot,func,
241                                                         PCI_CLASS_REVISION);
242                                 if (class == 0xffffffff)
243                                         break; 
244
245                                 if ((class >> 16) != PCI_CLASS_BRIDGE_PCI)
246                                         continue; 
247
248                                 vendor = read_pci_config(num, slot, func, 
249                                                          PCI_VENDOR_ID);
250                                 vendor &= 0xffff;
251                                 switch (vendor) { 
252                                 case PCI_VENDOR_ID_VIA:
253 #ifdef CONFIG_GART_IOMMU
254                                         if ((end_pfn >= (0xffffffff>>PAGE_SHIFT) ||
255                                              force_iommu) &&
256                                             !iommu_aperture_allowed) {
257                                                 printk(KERN_INFO
258     "Looks like a VIA chipset. Disabling IOMMU. Overwrite with \"iommu=allowed\"\n");
259                                                 iommu_aperture_disabled = 1;
260                                         }
261 #endif
262                                         return;
263                                 case PCI_VENDOR_ID_NVIDIA:
264 #ifdef CONFIG_ACPI
265                                         /* All timer overrides on Nvidia
266                                            seem to be wrong. Skip them. */
267                                         acpi_skip_timer_override = 1;
268                                         printk(KERN_INFO 
269              "Nvidia board detected. Ignoring ACPI timer override.\n");
270 #endif
271                                         /* RED-PEN skip them on mptables too? */
272                                         return;
273                                 } 
274
275                                 /* No multi-function device? */
276                                 type = read_pci_config_byte(num,slot,func,
277                                                             PCI_HEADER_TYPE);
278                                 if (!(type & 0x80))
279                                         break;
280                         } 
281                 }
282         }
283
284
285 static int __init ioapic_pirq_setup(char *str)
286 {
287         int i, max;
288         int ints[MAX_PIRQS+1];
289
290         get_options(str, ARRAY_SIZE(ints), ints);
291
292         for (i = 0; i < MAX_PIRQS; i++)
293                 pirq_entries[i] = -1;
294
295         pirqs_enabled = 1;
296         apic_printk(APIC_VERBOSE, "PIRQ redirection, working around broken MP-BIOS.\n");
297         max = MAX_PIRQS;
298         if (ints[0] < MAX_PIRQS)
299                 max = ints[0];
300
301         for (i = 0; i < max; i++) {
302                 apic_printk(APIC_VERBOSE, "... PIRQ%d -> IRQ %d\n", i, ints[i+1]);
303                 /*
304                  * PIRQs are mapped upside down, usually.
305                  */
306                 pirq_entries[MAX_PIRQS-i-1] = ints[i+1];
307         }
308         return 1;
309 }
310
311 __setup("pirq=", ioapic_pirq_setup);
312
313 /*
314  * Find the IRQ entry number of a certain pin.
315  */
316 static int find_irq_entry(int apic, int pin, int type)
317 {
318         int i;
319
320         for (i = 0; i < mp_irq_entries; i++)
321                 if (mp_irqs[i].mpc_irqtype == type &&
322                     (mp_irqs[i].mpc_dstapic == mp_ioapics[apic].mpc_apicid ||
323                      mp_irqs[i].mpc_dstapic == MP_APIC_ALL) &&
324                     mp_irqs[i].mpc_dstirq == pin)
325                         return i;
326
327         return -1;
328 }
329
330 /*
331  * Find the pin to which IRQ[irq] (ISA) is connected
332  */
333 static int __init find_isa_irq_pin(int irq, int type)
334 {
335         int i;
336
337         for (i = 0; i < mp_irq_entries; i++) {
338                 int lbus = mp_irqs[i].mpc_srcbus;
339
340                 if ((mp_bus_id_to_type[lbus] == MP_BUS_ISA ||
341                      mp_bus_id_to_type[lbus] == MP_BUS_EISA ||
342                      mp_bus_id_to_type[lbus] == MP_BUS_MCA) &&
343                     (mp_irqs[i].mpc_irqtype == type) &&
344                     (mp_irqs[i].mpc_srcbusirq == irq))
345
346                         return mp_irqs[i].mpc_dstirq;
347         }
348         return -1;
349 }
350
351 /*
352  * Find a specific PCI IRQ entry.
353  * Not an __init, possibly needed by modules
354  */
355 static int pin_2_irq(int idx, int apic, int pin);
356
357 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
358 {
359         int apic, i, best_guess = -1;
360
361         apic_printk(APIC_DEBUG, "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
362                 bus, slot, pin);
363         if (mp_bus_id_to_pci_bus[bus] == -1) {
364                 apic_printk(APIC_VERBOSE, "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
365                 return -1;
366         }
367         for (i = 0; i < mp_irq_entries; i++) {
368                 int lbus = mp_irqs[i].mpc_srcbus;
369
370                 for (apic = 0; apic < nr_ioapics; apic++)
371                         if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic ||
372                             mp_irqs[i].mpc_dstapic == MP_APIC_ALL)
373                                 break;
374
375                 if ((mp_bus_id_to_type[lbus] == MP_BUS_PCI) &&
376                     !mp_irqs[i].mpc_irqtype &&
377                     (bus == lbus) &&
378                     (slot == ((mp_irqs[i].mpc_srcbusirq >> 2) & 0x1f))) {
379                         int irq = pin_2_irq(i,apic,mp_irqs[i].mpc_dstirq);
380
381                         if (!(apic || IO_APIC_IRQ(irq)))
382                                 continue;
383
384                         if (pin == (mp_irqs[i].mpc_srcbusirq & 3))
385                                 return irq;
386                         /*
387                          * Use the first all-but-pin matching entry as a
388                          * best-guess fuzzy result for broken mptables.
389                          */
390                         if (best_guess < 0)
391                                 best_guess = irq;
392                 }
393         }
394         return best_guess;
395 }
396
397 /*
398  * EISA Edge/Level control register, ELCR
399  */
400 static int EISA_ELCR(unsigned int irq)
401 {
402         if (irq < 16) {
403                 unsigned int port = 0x4d0 + (irq >> 3);
404                 return (inb(port) >> (irq & 7)) & 1;
405         }
406         apic_printk(APIC_VERBOSE, "Broken MPtable reports ISA irq %d\n", irq);
407         return 0;
408 }
409
410 /* EISA interrupts are always polarity zero and can be edge or level
411  * trigger depending on the ELCR value.  If an interrupt is listed as
412  * EISA conforming in the MP table, that means its trigger type must
413  * be read in from the ELCR */
414
415 #define default_EISA_trigger(idx)       (EISA_ELCR(mp_irqs[idx].mpc_srcbusirq))
416 #define default_EISA_polarity(idx)      (0)
417
418 /* ISA interrupts are always polarity zero edge triggered,
419  * when listed as conforming in the MP table. */
420
421 #define default_ISA_trigger(idx)        (0)
422 #define default_ISA_polarity(idx)       (0)
423
424 /* PCI interrupts are always polarity one level triggered,
425  * when listed as conforming in the MP table. */
426
427 #define default_PCI_trigger(idx)        (1)
428 #define default_PCI_polarity(idx)       (1)
429
430 /* MCA interrupts are always polarity zero level triggered,
431  * when listed as conforming in the MP table. */
432
433 #define default_MCA_trigger(idx)        (1)
434 #define default_MCA_polarity(idx)       (0)
435
436 static int __init MPBIOS_polarity(int idx)
437 {
438         int bus = mp_irqs[idx].mpc_srcbus;
439         int polarity;
440
441         /*
442          * Determine IRQ line polarity (high active or low active):
443          */
444         switch (mp_irqs[idx].mpc_irqflag & 3)
445         {
446                 case 0: /* conforms, ie. bus-type dependent polarity */
447                 {
448                         switch (mp_bus_id_to_type[bus])
449                         {
450                                 case MP_BUS_ISA: /* ISA pin */
451                                 {
452                                         polarity = default_ISA_polarity(idx);
453                                         break;
454                                 }
455                                 case MP_BUS_EISA: /* EISA pin */
456                                 {
457                                         polarity = default_EISA_polarity(idx);
458                                         break;
459                                 }
460                                 case MP_BUS_PCI: /* PCI pin */
461                                 {
462                                         polarity = default_PCI_polarity(idx);
463                                         break;
464                                 }
465                                 case MP_BUS_MCA: /* MCA pin */
466                                 {
467                                         polarity = default_MCA_polarity(idx);
468                                         break;
469                                 }
470                                 default:
471                                 {
472                                         printk(KERN_WARNING "broken BIOS!!\n");
473                                         polarity = 1;
474                                         break;
475                                 }
476                         }
477                         break;
478                 }
479                 case 1: /* high active */
480                 {
481                         polarity = 0;
482                         break;
483                 }
484                 case 2: /* reserved */
485                 {
486                         printk(KERN_WARNING "broken BIOS!!\n");
487                         polarity = 1;
488                         break;
489                 }
490                 case 3: /* low active */
491                 {
492                         polarity = 1;
493                         break;
494                 }
495                 default: /* invalid */
496                 {
497                         printk(KERN_WARNING "broken BIOS!!\n");
498                         polarity = 1;
499                         break;
500                 }
501         }
502         return polarity;
503 }
504
505 static int MPBIOS_trigger(int idx)
506 {
507         int bus = mp_irqs[idx].mpc_srcbus;
508         int trigger;
509
510         /*
511          * Determine IRQ trigger mode (edge or level sensitive):
512          */
513         switch ((mp_irqs[idx].mpc_irqflag>>2) & 3)
514         {
515                 case 0: /* conforms, ie. bus-type dependent */
516                 {
517                         switch (mp_bus_id_to_type[bus])
518                         {
519                                 case MP_BUS_ISA: /* ISA pin */
520                                 {
521                                         trigger = default_ISA_trigger(idx);
522                                         break;
523                                 }
524                                 case MP_BUS_EISA: /* EISA pin */
525                                 {
526                                         trigger = default_EISA_trigger(idx);
527                                         break;
528                                 }
529                                 case MP_BUS_PCI: /* PCI pin */
530                                 {
531                                         trigger = default_PCI_trigger(idx);
532                                         break;
533                                 }
534                                 case MP_BUS_MCA: /* MCA pin */
535                                 {
536                                         trigger = default_MCA_trigger(idx);
537                                         break;
538                                 }
539                                 default:
540                                 {
541                                         printk(KERN_WARNING "broken BIOS!!\n");
542                                         trigger = 1;
543                                         break;
544                                 }
545                         }
546                         break;
547                 }
548                 case 1: /* edge */
549                 {
550                         trigger = 0;
551                         break;
552                 }
553                 case 2: /* reserved */
554                 {
555                         printk(KERN_WARNING "broken BIOS!!\n");
556                         trigger = 1;
557                         break;
558                 }
559                 case 3: /* level */
560                 {
561                         trigger = 1;
562                         break;
563                 }
564                 default: /* invalid */
565                 {
566                         printk(KERN_WARNING "broken BIOS!!\n");
567                         trigger = 0;
568                         break;
569                 }
570         }
571         return trigger;
572 }
573
574 static inline int irq_polarity(int idx)
575 {
576         return MPBIOS_polarity(idx);
577 }
578
579 static inline int irq_trigger(int idx)
580 {
581         return MPBIOS_trigger(idx);
582 }
583
584 static int pin_2_irq(int idx, int apic, int pin)
585 {
586         int irq, i;
587         int bus = mp_irqs[idx].mpc_srcbus;
588
589         /*
590          * Debugging check, we are in big trouble if this message pops up!
591          */
592         if (mp_irqs[idx].mpc_dstirq != pin)
593                 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
594
595         switch (mp_bus_id_to_type[bus])
596         {
597                 case MP_BUS_ISA: /* ISA pin */
598                 case MP_BUS_EISA:
599                 case MP_BUS_MCA:
600                 {
601                         irq = mp_irqs[idx].mpc_srcbusirq;
602                         break;
603                 }
604                 case MP_BUS_PCI: /* PCI pin */
605                 {
606                         /*
607                          * PCI IRQs are mapped in order
608                          */
609                         i = irq = 0;
610                         while (i < apic)
611                                 irq += nr_ioapic_registers[i++];
612                         irq += pin;
613                         break;
614                 }
615                 default:
616                 {
617                         printk(KERN_ERR "unknown bus type %d.\n",bus); 
618                         irq = 0;
619                         break;
620                 }
621         }
622
623         /*
624          * PCI IRQ command line redirection. Yes, limits are hardcoded.
625          */
626         if ((pin >= 16) && (pin <= 23)) {
627                 if (pirq_entries[pin-16] != -1) {
628                         if (!pirq_entries[pin-16]) {
629                                 apic_printk(APIC_VERBOSE, "disabling PIRQ%d\n", pin-16);
630                         } else {
631                                 irq = pirq_entries[pin-16];
632                                 apic_printk(APIC_VERBOSE, "using PIRQ%d -> IRQ %d\n",
633                                                 pin-16, irq);
634                         }
635                 }
636         }
637         return irq;
638 }
639
640 static inline int IO_APIC_irq_trigger(int irq)
641 {
642         int apic, idx, pin;
643
644         for (apic = 0; apic < nr_ioapics; apic++) {
645                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
646                         idx = find_irq_entry(apic,pin,mp_INT);
647                         if ((idx != -1) && (irq == pin_2_irq(idx,apic,pin)))
648                                 return irq_trigger(idx);
649                 }
650         }
651         /*
652          * nonexistent IRQs are edge default
653          */
654         return 0;
655 }
656
657 /* irq_vectors is indexed by the sum of all RTEs in all I/O APICs. */
658 u8 irq_vector[NR_IRQ_VECTORS] = { FIRST_DEVICE_VECTOR , 0 };
659
660 int assign_irq_vector(int irq)
661 {
662         static int current_vector = FIRST_DEVICE_VECTOR, offset = 0;
663
664         BUG_ON(irq >= NR_IRQ_VECTORS);
665         if (IO_APIC_VECTOR(irq) > 0)
666                 return IO_APIC_VECTOR(irq);
667 next:
668         current_vector += 8;
669         if (current_vector == IA32_SYSCALL_VECTOR)
670                 goto next;
671
672         if (current_vector >= FIRST_SYSTEM_VECTOR) {
673                 offset++;
674                 if (!(offset%8))
675                         return -ENOSPC;
676                 current_vector = FIRST_DEVICE_VECTOR + offset;
677         }
678
679         vector_irq[current_vector] = irq;
680         if (irq != AUTO_ASSIGN)
681                 IO_APIC_VECTOR(irq) = current_vector;
682
683         return current_vector;
684 }
685
686 extern void (*interrupt[NR_IRQS])(void);
687 static struct hw_interrupt_type ioapic_level_type;
688 static struct hw_interrupt_type ioapic_edge_type;
689
690 #define IOAPIC_AUTO     -1
691 #define IOAPIC_EDGE     0
692 #define IOAPIC_LEVEL    1
693
694 static inline void ioapic_register_intr(int irq, int vector, unsigned long trigger)
695 {
696         if (use_pci_vector() && !platform_legacy_irq(irq)) {
697                 if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
698                                 trigger == IOAPIC_LEVEL)
699                         irq_desc[vector].handler = &ioapic_level_type;
700                 else
701                         irq_desc[vector].handler = &ioapic_edge_type;
702                 set_intr_gate(vector, interrupt[vector]);
703         } else  {
704                 if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
705                                 trigger == IOAPIC_LEVEL)
706                         irq_desc[irq].handler = &ioapic_level_type;
707                 else
708                         irq_desc[irq].handler = &ioapic_edge_type;
709                 set_intr_gate(vector, interrupt[irq]);
710         }
711 }
712
713 static void __init setup_IO_APIC_irqs(void)
714 {
715         struct IO_APIC_route_entry entry;
716         int apic, pin, idx, irq, first_notcon = 1, vector;
717         unsigned long flags;
718
719         apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
720
721         for (apic = 0; apic < nr_ioapics; apic++) {
722         for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
723
724                 /*
725                  * add it to the IO-APIC irq-routing table:
726                  */
727                 memset(&entry,0,sizeof(entry));
728
729                 entry.delivery_mode = INT_DELIVERY_MODE;
730                 entry.dest_mode = INT_DEST_MODE;
731                 entry.mask = 0;                         /* enable IRQ */
732                 entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
733
734                 idx = find_irq_entry(apic,pin,mp_INT);
735                 if (idx == -1) {
736                         if (first_notcon) {
737                                 apic_printk(APIC_VERBOSE, KERN_DEBUG " IO-APIC (apicid-pin) %d-%d", mp_ioapics[apic].mpc_apicid, pin);
738                                 first_notcon = 0;
739                         } else
740                                 apic_printk(APIC_VERBOSE, ", %d-%d", mp_ioapics[apic].mpc_apicid, pin);
741                         continue;
742                 }
743
744                 entry.trigger = irq_trigger(idx);
745                 entry.polarity = irq_polarity(idx);
746
747                 if (irq_trigger(idx)) {
748                         entry.trigger = 1;
749                         entry.mask = 1;
750                         entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
751                 }
752
753                 irq = pin_2_irq(idx, apic, pin);
754                 add_pin_to_irq(irq, apic, pin);
755
756                 if (!apic && !IO_APIC_IRQ(irq))
757                         continue;
758
759                 if (IO_APIC_IRQ(irq)) {
760                         vector = assign_irq_vector(irq);
761                         entry.vector = vector;
762
763                         ioapic_register_intr(irq, vector, IOAPIC_AUTO);
764                         if (!apic && (irq < 16))
765                                 disable_8259A_irq(irq);
766                 }
767                 spin_lock_irqsave(&ioapic_lock, flags);
768                 io_apic_write(apic, 0x11+2*pin, *(((int *)&entry)+1));
769                 io_apic_write(apic, 0x10+2*pin, *(((int *)&entry)+0));
770                 spin_unlock_irqrestore(&ioapic_lock, flags);
771         }
772         }
773
774         if (!first_notcon)
775                 apic_printk(APIC_VERBOSE," not connected.\n");
776 }
777
778 /*
779  * Set up the 8259A-master output pin as broadcast to all
780  * CPUs.
781  */
782 static void __init setup_ExtINT_IRQ0_pin(unsigned int pin, int vector)
783 {
784         struct IO_APIC_route_entry entry;
785         unsigned long flags;
786
787         memset(&entry,0,sizeof(entry));
788
789         disable_8259A_irq(0);
790
791         /* mask LVT0 */
792         apic_write_around(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
793
794         /*
795          * We use logical delivery to get the timer IRQ
796          * to the first CPU.
797          */
798         entry.dest_mode = INT_DEST_MODE;
799         entry.mask = 0;                                 /* unmask IRQ now */
800         entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
801         entry.delivery_mode = INT_DELIVERY_MODE;
802         entry.polarity = 0;
803         entry.trigger = 0;
804         entry.vector = vector;
805
806         /*
807          * The timer IRQ doesn't have to know that behind the
808          * scene we have a 8259A-master in AEOI mode ...
809          */
810         irq_desc[0].handler = &ioapic_edge_type;
811
812         /*
813          * Add it to the IO-APIC irq-routing table:
814          */
815         spin_lock_irqsave(&ioapic_lock, flags);
816         io_apic_write(0, 0x11+2*pin, *(((int *)&entry)+1));
817         io_apic_write(0, 0x10+2*pin, *(((int *)&entry)+0));
818         spin_unlock_irqrestore(&ioapic_lock, flags);
819
820         enable_8259A_irq(0);
821 }
822
823 void __init UNEXPECTED_IO_APIC(void)
824 {
825 }
826
827 void __apicdebuginit print_IO_APIC(void)
828 {
829         int apic, i;
830         union IO_APIC_reg_00 reg_00;
831         union IO_APIC_reg_01 reg_01;
832         union IO_APIC_reg_02 reg_02;
833         unsigned long flags;
834
835         if (apic_verbosity == APIC_QUIET)
836                 return;
837
838         printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
839         for (i = 0; i < nr_ioapics; i++)
840                 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
841                        mp_ioapics[i].mpc_apicid, nr_ioapic_registers[i]);
842
843         /*
844          * We are a bit conservative about what we expect.  We have to
845          * know about every hardware change ASAP.
846          */
847         printk(KERN_INFO "testing the IO APIC.......................\n");
848
849         for (apic = 0; apic < nr_ioapics; apic++) {
850
851         spin_lock_irqsave(&ioapic_lock, flags);
852         reg_00.raw = io_apic_read(apic, 0);
853         reg_01.raw = io_apic_read(apic, 1);
854         if (reg_01.bits.version >= 0x10)
855                 reg_02.raw = io_apic_read(apic, 2);
856         spin_unlock_irqrestore(&ioapic_lock, flags);
857
858         printk("\n");
859         printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].mpc_apicid);
860         printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
861         printk(KERN_DEBUG ".......    : physical APIC id: %02X\n", reg_00.bits.ID);
862         if (reg_00.bits.__reserved_1 || reg_00.bits.__reserved_2)
863                 UNEXPECTED_IO_APIC();
864
865         printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)&reg_01);
866         printk(KERN_DEBUG ".......     : max redirection entries: %04X\n", reg_01.bits.entries);
867         if (    (reg_01.bits.entries != 0x0f) && /* older (Neptune) boards */
868                 (reg_01.bits.entries != 0x17) && /* typical ISA+PCI boards */
869                 (reg_01.bits.entries != 0x1b) && /* Compaq Proliant boards */
870                 (reg_01.bits.entries != 0x1f) && /* dual Xeon boards */
871                 (reg_01.bits.entries != 0x22) && /* bigger Xeon boards */
872                 (reg_01.bits.entries != 0x2E) &&
873                 (reg_01.bits.entries != 0x3F) &&
874                 (reg_01.bits.entries != 0x03) 
875         )
876                 UNEXPECTED_IO_APIC();
877
878         printk(KERN_DEBUG ".......     : PRQ implemented: %X\n", reg_01.bits.PRQ);
879         printk(KERN_DEBUG ".......     : IO APIC version: %04X\n", reg_01.bits.version);
880         if (    (reg_01.bits.version != 0x01) && /* 82489DX IO-APICs */
881                 (reg_01.bits.version != 0x02) && /* 82801BA IO-APICs (ICH2) */
882                 (reg_01.bits.version != 0x10) && /* oldest IO-APICs */
883                 (reg_01.bits.version != 0x11) && /* Pentium/Pro IO-APICs */
884                 (reg_01.bits.version != 0x13) && /* Xeon IO-APICs */
885                 (reg_01.bits.version != 0x20)    /* Intel P64H (82806 AA) */
886         )
887                 UNEXPECTED_IO_APIC();
888         if (reg_01.bits.__reserved_1 || reg_01.bits.__reserved_2)
889                 UNEXPECTED_IO_APIC();
890
891         if (reg_01.bits.version >= 0x10) {
892                 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
893                 printk(KERN_DEBUG ".......     : arbitration: %02X\n", reg_02.bits.arbitration);
894                 if (reg_02.bits.__reserved_1 || reg_02.bits.__reserved_2)
895                         UNEXPECTED_IO_APIC();
896         }
897
898         printk(KERN_DEBUG ".... IRQ redirection table:\n");
899
900         printk(KERN_DEBUG " NR Log Phy Mask Trig IRR Pol"
901                           " Stat Dest Deli Vect:   \n");
902
903         for (i = 0; i <= reg_01.bits.entries; i++) {
904                 struct IO_APIC_route_entry entry;
905
906                 spin_lock_irqsave(&ioapic_lock, flags);
907                 *(((int *)&entry)+0) = io_apic_read(apic, 0x10+i*2);
908                 *(((int *)&entry)+1) = io_apic_read(apic, 0x11+i*2);
909                 spin_unlock_irqrestore(&ioapic_lock, flags);
910
911                 printk(KERN_DEBUG " %02x %03X %02X  ",
912                         i,
913                         entry.dest.logical.logical_dest,
914                         entry.dest.physical.physical_dest
915                 );
916
917                 printk("%1d    %1d    %1d   %1d   %1d    %1d    %1d    %02X\n",
918                         entry.mask,
919                         entry.trigger,
920                         entry.irr,
921                         entry.polarity,
922                         entry.delivery_status,
923                         entry.dest_mode,
924                         entry.delivery_mode,
925                         entry.vector
926                 );
927         }
928         }
929         if (use_pci_vector())
930                 printk(KERN_INFO "Using vector-based indexing\n");
931         printk(KERN_DEBUG "IRQ to pin mappings:\n");
932         for (i = 0; i < NR_IRQS; i++) {
933                 struct irq_pin_list *entry = irq_2_pin + i;
934                 if (entry->pin < 0)
935                         continue;
936                 if (use_pci_vector() && !platform_legacy_irq(i))
937                         printk(KERN_DEBUG "IRQ%d ", IO_APIC_VECTOR(i));
938                 else
939                         printk(KERN_DEBUG "IRQ%d ", i);
940                 for (;;) {
941                         printk("-> %d:%d", entry->apic, entry->pin);
942                         if (!entry->next)
943                                 break;
944                         entry = irq_2_pin + entry->next;
945                 }
946                 printk("\n");
947         }
948
949         printk(KERN_INFO ".................................... done.\n");
950
951         return;
952 }
953
954 #if 0
955
956 static __apicdebuginit void print_APIC_bitfield (int base)
957 {
958         unsigned int v;
959         int i, j;
960
961         if (apic_verbosity == APIC_QUIET)
962                 return;
963
964         printk(KERN_DEBUG "0123456789abcdef0123456789abcdef\n" KERN_DEBUG);
965         for (i = 0; i < 8; i++) {
966                 v = apic_read(base + i*0x10);
967                 for (j = 0; j < 32; j++) {
968                         if (v & (1<<j))
969                                 printk("1");
970                         else
971                                 printk("0");
972                 }
973                 printk("\n");
974         }
975 }
976
977 void __apicdebuginit print_local_APIC(void * dummy)
978 {
979         unsigned int v, ver, maxlvt;
980
981         if (apic_verbosity == APIC_QUIET)
982                 return;
983
984         printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
985                 smp_processor_id(), hard_smp_processor_id());
986         v = apic_read(APIC_ID);
987         printk(KERN_INFO "... APIC ID:      %08x (%01x)\n", v, GET_APIC_ID(v));
988         v = apic_read(APIC_LVR);
989         printk(KERN_INFO "... APIC VERSION: %08x\n", v);
990         ver = GET_APIC_VERSION(v);
991         maxlvt = get_maxlvt();
992
993         v = apic_read(APIC_TASKPRI);
994         printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
995
996         if (APIC_INTEGRATED(ver)) {                     /* !82489DX */
997                 v = apic_read(APIC_ARBPRI);
998                 printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
999                         v & APIC_ARBPRI_MASK);
1000                 v = apic_read(APIC_PROCPRI);
1001                 printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1002         }
1003
1004         v = apic_read(APIC_EOI);
1005         printk(KERN_DEBUG "... APIC EOI: %08x\n", v);
1006         v = apic_read(APIC_RRR);
1007         printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1008         v = apic_read(APIC_LDR);
1009         printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1010         v = apic_read(APIC_DFR);
1011         printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1012         v = apic_read(APIC_SPIV);
1013         printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1014
1015         printk(KERN_DEBUG "... APIC ISR field:\n");
1016         print_APIC_bitfield(APIC_ISR);
1017         printk(KERN_DEBUG "... APIC TMR field:\n");
1018         print_APIC_bitfield(APIC_TMR);
1019         printk(KERN_DEBUG "... APIC IRR field:\n");
1020         print_APIC_bitfield(APIC_IRR);
1021
1022         if (APIC_INTEGRATED(ver)) {             /* !82489DX */
1023                 if (maxlvt > 3)         /* Due to the Pentium erratum 3AP. */
1024                         apic_write(APIC_ESR, 0);
1025                 v = apic_read(APIC_ESR);
1026                 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1027         }
1028
1029         v = apic_read(APIC_ICR);
1030         printk(KERN_DEBUG "... APIC ICR: %08x\n", v);
1031         v = apic_read(APIC_ICR2);
1032         printk(KERN_DEBUG "... APIC ICR2: %08x\n", v);
1033
1034         v = apic_read(APIC_LVTT);
1035         printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1036
1037         if (maxlvt > 3) {                       /* PC is LVT#4. */
1038                 v = apic_read(APIC_LVTPC);
1039                 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1040         }
1041         v = apic_read(APIC_LVT0);
1042         printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1043         v = apic_read(APIC_LVT1);
1044         printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1045
1046         if (maxlvt > 2) {                       /* ERR is LVT#3. */
1047                 v = apic_read(APIC_LVTERR);
1048                 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1049         }
1050
1051         v = apic_read(APIC_TMICT);
1052         printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1053         v = apic_read(APIC_TMCCT);
1054         printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1055         v = apic_read(APIC_TDCR);
1056         printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1057         printk("\n");
1058 }
1059
1060 void print_all_local_APICs (void)
1061 {
1062         on_each_cpu(print_local_APIC, NULL, 1, 1);
1063 }
1064
1065 void __apicdebuginit print_PIC(void)
1066 {
1067         extern spinlock_t i8259A_lock;
1068         unsigned int v;
1069         unsigned long flags;
1070
1071         if (apic_verbosity == APIC_QUIET)
1072                 return;
1073
1074         printk(KERN_DEBUG "\nprinting PIC contents\n");
1075
1076         spin_lock_irqsave(&i8259A_lock, flags);
1077
1078         v = inb(0xa1) << 8 | inb(0x21);
1079         printk(KERN_DEBUG "... PIC  IMR: %04x\n", v);
1080
1081         v = inb(0xa0) << 8 | inb(0x20);
1082         printk(KERN_DEBUG "... PIC  IRR: %04x\n", v);
1083
1084         outb(0x0b,0xa0);
1085         outb(0x0b,0x20);
1086         v = inb(0xa0) << 8 | inb(0x20);
1087         outb(0x0a,0xa0);
1088         outb(0x0a,0x20);
1089
1090         spin_unlock_irqrestore(&i8259A_lock, flags);
1091
1092         printk(KERN_DEBUG "... PIC  ISR: %04x\n", v);
1093
1094         v = inb(0x4d1) << 8 | inb(0x4d0);
1095         printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1096 }
1097
1098 #endif  /*  0  */
1099
1100 static void __init enable_IO_APIC(void)
1101 {
1102         union IO_APIC_reg_01 reg_01;
1103         int i;
1104         unsigned long flags;
1105
1106         for (i = 0; i < PIN_MAP_SIZE; i++) {
1107                 irq_2_pin[i].pin = -1;
1108                 irq_2_pin[i].next = 0;
1109         }
1110         if (!pirqs_enabled)
1111                 for (i = 0; i < MAX_PIRQS; i++)
1112                         pirq_entries[i] = -1;
1113
1114         /*
1115          * The number of IO-APIC IRQ registers (== #pins):
1116          */
1117         for (i = 0; i < nr_ioapics; i++) {
1118                 spin_lock_irqsave(&ioapic_lock, flags);
1119                 reg_01.raw = io_apic_read(i, 1);
1120                 spin_unlock_irqrestore(&ioapic_lock, flags);
1121                 nr_ioapic_registers[i] = reg_01.bits.entries+1;
1122         }
1123
1124         /*
1125          * Do not trust the IO-APIC being empty at bootup
1126          */
1127         clear_IO_APIC();
1128 }
1129
1130 /*
1131  * Not an __init, needed by the reboot code
1132  */
1133 void disable_IO_APIC(void)
1134 {
1135         /*
1136          * Clear the IO-APIC before rebooting:
1137          */
1138         clear_IO_APIC();
1139
1140         disconnect_bsp_APIC();
1141 }
1142
1143 /*
1144  * function to set the IO-APIC physical IDs based on the
1145  * values stored in the MPC table.
1146  *
1147  * by Matt Domsch <Matt_Domsch@dell.com>  Tue Dec 21 12:25:05 CST 1999
1148  */
1149
1150 static void __init setup_ioapic_ids_from_mpc (void)
1151 {
1152         union IO_APIC_reg_00 reg_00;
1153         int apic;
1154         int i;
1155         unsigned char old_id;
1156         unsigned long flags;
1157
1158         /*
1159          * Set the IOAPIC ID to the value stored in the MPC table.
1160          */
1161         for (apic = 0; apic < nr_ioapics; apic++) {
1162
1163                 /* Read the register 0 value */
1164                 spin_lock_irqsave(&ioapic_lock, flags);
1165                 reg_00.raw = io_apic_read(apic, 0);
1166                 spin_unlock_irqrestore(&ioapic_lock, flags);
1167                 
1168                 old_id = mp_ioapics[apic].mpc_apicid;
1169
1170
1171                 printk(KERN_INFO "Using IO-APIC %d\n", mp_ioapics[apic].mpc_apicid);
1172
1173
1174                 /*
1175                  * We need to adjust the IRQ routing table
1176                  * if the ID changed.
1177                  */
1178                 if (old_id != mp_ioapics[apic].mpc_apicid)
1179                         for (i = 0; i < mp_irq_entries; i++)
1180                                 if (mp_irqs[i].mpc_dstapic == old_id)
1181                                         mp_irqs[i].mpc_dstapic
1182                                                 = mp_ioapics[apic].mpc_apicid;
1183
1184                 /*
1185                  * Read the right value from the MPC table and
1186                  * write it into the ID register.
1187                  */
1188                 apic_printk(APIC_VERBOSE,KERN_INFO "...changing IO-APIC physical APIC ID to %d ...",
1189                                 mp_ioapics[apic].mpc_apicid);
1190
1191                 reg_00.bits.ID = mp_ioapics[apic].mpc_apicid;
1192                 spin_lock_irqsave(&ioapic_lock, flags);
1193                 io_apic_write(apic, 0, reg_00.raw);
1194                 spin_unlock_irqrestore(&ioapic_lock, flags);
1195
1196                 /*
1197                  * Sanity check
1198                  */
1199                 spin_lock_irqsave(&ioapic_lock, flags);
1200                 reg_00.raw = io_apic_read(apic, 0);
1201                 spin_unlock_irqrestore(&ioapic_lock, flags);
1202                 if (reg_00.bits.ID != mp_ioapics[apic].mpc_apicid)
1203                         printk("could not set ID!\n");
1204                 else
1205                         apic_printk(APIC_VERBOSE," ok.\n");
1206         }
1207 }
1208
1209 /*
1210  * There is a nasty bug in some older SMP boards, their mptable lies
1211  * about the timer IRQ. We do the following to work around the situation:
1212  *
1213  *      - timer IRQ defaults to IO-APIC IRQ
1214  *      - if this function detects that timer IRQs are defunct, then we fall
1215  *        back to ISA timer IRQs
1216  */
1217 static int __init timer_irq_works(void)
1218 {
1219         unsigned long t1 = jiffies;
1220
1221         local_irq_enable();
1222         /* Let ten ticks pass... */
1223         mdelay((10 * 1000) / HZ);
1224
1225         /*
1226          * Expect a few ticks at least, to be sure some possible
1227          * glue logic does not lock up after one or two first
1228          * ticks in a non-ExtINT mode.  Also the local APIC
1229          * might have cached one ExtINT interrupt.  Finally, at
1230          * least one tick may be lost due to delays.
1231          */
1232
1233         /* jiffies wrap? */
1234         if (jiffies - t1 > 4)
1235                 return 1;
1236         return 0;
1237 }
1238
1239 /*
1240  * In the SMP+IOAPIC case it might happen that there are an unspecified
1241  * number of pending IRQ events unhandled. These cases are very rare,
1242  * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
1243  * better to do it this way as thus we do not have to be aware of
1244  * 'pending' interrupts in the IRQ path, except at this point.
1245  */
1246 /*
1247  * Edge triggered needs to resend any interrupt
1248  * that was delayed but this is now handled in the device
1249  * independent code.
1250  */
1251
1252 /*
1253  * Starting up a edge-triggered IO-APIC interrupt is
1254  * nasty - we need to make sure that we get the edge.
1255  * If it is already asserted for some reason, we need
1256  * return 1 to indicate that is was pending.
1257  *
1258  * This is not complete - we should be able to fake
1259  * an edge even if it isn't on the 8259A...
1260  */
1261
1262 static unsigned int startup_edge_ioapic_irq(unsigned int irq)
1263 {
1264         int was_pending = 0;
1265         unsigned long flags;
1266
1267         spin_lock_irqsave(&ioapic_lock, flags);
1268         if (irq < 16) {
1269                 disable_8259A_irq(irq);
1270                 if (i8259A_irq_pending(irq))
1271                         was_pending = 1;
1272         }
1273         __unmask_IO_APIC_irq(irq);
1274         spin_unlock_irqrestore(&ioapic_lock, flags);
1275
1276         return was_pending;
1277 }
1278
1279 /*
1280  * Once we have recorded IRQ_PENDING already, we can mask the
1281  * interrupt for real. This prevents IRQ storms from unhandled
1282  * devices.
1283  */
1284 static void ack_edge_ioapic_irq(unsigned int irq)
1285 {
1286         if ((irq_desc[irq].status & (IRQ_PENDING | IRQ_DISABLED))
1287                                         == (IRQ_PENDING | IRQ_DISABLED))
1288                 mask_IO_APIC_irq(irq);
1289         ack_APIC_irq();
1290 }
1291
1292 /*
1293  * Level triggered interrupts can just be masked,
1294  * and shutting down and starting up the interrupt
1295  * is the same as enabling and disabling them -- except
1296  * with a startup need to return a "was pending" value.
1297  *
1298  * Level triggered interrupts are special because we
1299  * do not touch any IO-APIC register while handling
1300  * them. We ack the APIC in the end-IRQ handler, not
1301  * in the start-IRQ-handler. Protection against reentrance
1302  * from the same interrupt is still provided, both by the
1303  * generic IRQ layer and by the fact that an unacked local
1304  * APIC does not accept IRQs.
1305  */
1306 static unsigned int startup_level_ioapic_irq (unsigned int irq)
1307 {
1308         unmask_IO_APIC_irq(irq);
1309
1310         return 0; /* don't check for pending */
1311 }
1312
1313 static void end_level_ioapic_irq (unsigned int irq)
1314 {
1315         ack_APIC_irq();
1316 }
1317
1318 static void set_ioapic_affinity_irq(unsigned int irq, cpumask_t mask)
1319 {
1320         unsigned long flags;
1321         unsigned int dest;
1322
1323         dest = cpu_mask_to_apicid(mask);
1324
1325         /*
1326          * Only the high 8 bits are valid.
1327          */
1328         dest = SET_APIC_LOGICAL_ID(dest);
1329
1330         spin_lock_irqsave(&ioapic_lock, flags);
1331         __DO_ACTION(1, = dest, )
1332         spin_unlock_irqrestore(&ioapic_lock, flags);
1333 }
1334
1335 #ifdef CONFIG_PCI_MSI
1336 static unsigned int startup_edge_ioapic_vector(unsigned int vector)
1337 {
1338         int irq = vector_to_irq(vector);
1339
1340         return startup_edge_ioapic_irq(irq);
1341 }
1342
1343 static void ack_edge_ioapic_vector(unsigned int vector)
1344 {
1345         int irq = vector_to_irq(vector);
1346
1347         ack_edge_ioapic_irq(irq);
1348 }
1349
1350 static unsigned int startup_level_ioapic_vector (unsigned int vector)
1351 {
1352         int irq = vector_to_irq(vector);
1353
1354         return startup_level_ioapic_irq (irq);
1355 }
1356
1357 static void end_level_ioapic_vector (unsigned int vector)
1358 {
1359         int irq = vector_to_irq(vector);
1360
1361         end_level_ioapic_irq(irq);
1362 }
1363
1364 static void mask_IO_APIC_vector (unsigned int vector)
1365 {
1366         int irq = vector_to_irq(vector);
1367
1368         mask_IO_APIC_irq(irq);
1369 }
1370
1371 static void unmask_IO_APIC_vector (unsigned int vector)
1372 {
1373         int irq = vector_to_irq(vector);
1374
1375         unmask_IO_APIC_irq(irq);
1376 }
1377
1378 static void set_ioapic_affinity_vector (unsigned int vector,
1379                                         cpumask_t cpu_mask)
1380 {
1381         int irq = vector_to_irq(vector);
1382
1383         set_ioapic_affinity_irq(irq, cpu_mask);
1384 }
1385 #endif
1386
1387 /*
1388  * Level and edge triggered IO-APIC interrupts need different handling,
1389  * so we use two separate IRQ descriptors. Edge triggered IRQs can be
1390  * handled with the level-triggered descriptor, but that one has slightly
1391  * more overhead. Level-triggered interrupts cannot be handled with the
1392  * edge-triggered handler, without risking IRQ storms and other ugly
1393  * races.
1394  */
1395
1396 static struct hw_interrupt_type ioapic_edge_type = {
1397         .typename = "IO-APIC-edge",
1398         .startup        = startup_edge_ioapic,
1399         .shutdown       = shutdown_edge_ioapic,
1400         .enable         = enable_edge_ioapic,
1401         .disable        = disable_edge_ioapic,
1402         .ack            = ack_edge_ioapic,
1403         .end            = end_edge_ioapic,
1404         .set_affinity = set_ioapic_affinity,
1405 };
1406
1407 static struct hw_interrupt_type ioapic_level_type = {
1408         .typename = "IO-APIC-level",
1409         .startup        = startup_level_ioapic,
1410         .shutdown       = shutdown_level_ioapic,
1411         .enable         = enable_level_ioapic,
1412         .disable        = disable_level_ioapic,
1413         .ack            = mask_and_ack_level_ioapic,
1414         .end            = end_level_ioapic,
1415         .set_affinity = set_ioapic_affinity,
1416 };
1417
1418 static inline void init_IO_APIC_traps(void)
1419 {
1420         int irq;
1421
1422         /*
1423          * NOTE! The local APIC isn't very good at handling
1424          * multiple interrupts at the same interrupt level.
1425          * As the interrupt level is determined by taking the
1426          * vector number and shifting that right by 4, we
1427          * want to spread these out a bit so that they don't
1428          * all fall in the same interrupt level.
1429          *
1430          * Also, we've got to be careful not to trash gate
1431          * 0x80, because int 0x80 is hm, kind of importantish. ;)
1432          */
1433         for (irq = 0; irq < NR_IRQS ; irq++) {
1434                 int tmp = irq;
1435                 if (use_pci_vector()) {
1436                         if (!platform_legacy_irq(tmp))
1437                                 if ((tmp = vector_to_irq(tmp)) == -1)
1438                                         continue;
1439                 }
1440                 if (IO_APIC_IRQ(tmp) && !IO_APIC_VECTOR(tmp)) {
1441                         /*
1442                          * Hmm.. We don't have an entry for this,
1443                          * so default to an old-fashioned 8259
1444                          * interrupt if we can..
1445                          */
1446                         if (irq < 16)
1447                                 make_8259A_irq(irq);
1448                         else
1449                                 /* Strange. Oh, well.. */
1450                                 irq_desc[irq].handler = &no_irq_type;
1451                 }
1452         }
1453 }
1454
1455 static void enable_lapic_irq (unsigned int irq)
1456 {
1457         unsigned long v;
1458
1459         v = apic_read(APIC_LVT0);
1460         apic_write_around(APIC_LVT0, v & ~APIC_LVT_MASKED);
1461 }
1462
1463 static void disable_lapic_irq (unsigned int irq)
1464 {
1465         unsigned long v;
1466
1467         v = apic_read(APIC_LVT0);
1468         apic_write_around(APIC_LVT0, v | APIC_LVT_MASKED);
1469 }
1470
1471 static void ack_lapic_irq (unsigned int irq)
1472 {
1473         ack_APIC_irq();
1474 }
1475
1476 static void end_lapic_irq (unsigned int i) { /* nothing */ }
1477
1478 static struct hw_interrupt_type lapic_irq_type = {
1479         .typename = "local-APIC-edge",
1480         .startup = NULL, /* startup_irq() not used for IRQ0 */
1481         .shutdown = NULL, /* shutdown_irq() not used for IRQ0 */
1482         .enable = enable_lapic_irq,
1483         .disable = disable_lapic_irq,
1484         .ack = ack_lapic_irq,
1485         .end = end_lapic_irq,
1486 };
1487
1488 static void setup_nmi (void)
1489 {
1490         /*
1491          * Dirty trick to enable the NMI watchdog ...
1492          * We put the 8259A master into AEOI mode and
1493          * unmask on all local APICs LVT0 as NMI.
1494          *
1495          * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
1496          * is from Maciej W. Rozycki - so we do not have to EOI from
1497          * the NMI handler or the timer interrupt.
1498          */ 
1499         printk(KERN_INFO "activating NMI Watchdog ...");
1500
1501         enable_NMI_through_LVT0(NULL);
1502
1503         printk(" done.\n");
1504 }
1505
1506 /*
1507  * This looks a bit hackish but it's about the only one way of sending
1508  * a few INTA cycles to 8259As and any associated glue logic.  ICR does
1509  * not support the ExtINT mode, unfortunately.  We need to send these
1510  * cycles as some i82489DX-based boards have glue logic that keeps the
1511  * 8259A interrupt line asserted until INTA.  --macro
1512  */
1513 static inline void unlock_ExtINT_logic(void)
1514 {
1515         int pin, i;
1516         struct IO_APIC_route_entry entry0, entry1;
1517         unsigned char save_control, save_freq_select;
1518         unsigned long flags;
1519
1520         pin = find_isa_irq_pin(8, mp_INT);
1521         if (pin == -1)
1522                 return;
1523
1524         spin_lock_irqsave(&ioapic_lock, flags);
1525         *(((int *)&entry0) + 1) = io_apic_read(0, 0x11 + 2 * pin);
1526         *(((int *)&entry0) + 0) = io_apic_read(0, 0x10 + 2 * pin);
1527         spin_unlock_irqrestore(&ioapic_lock, flags);
1528         clear_IO_APIC_pin(0, pin);
1529
1530         memset(&entry1, 0, sizeof(entry1));
1531
1532         entry1.dest_mode = 0;                   /* physical delivery */
1533         entry1.mask = 0;                        /* unmask IRQ now */
1534         entry1.dest.physical.physical_dest = hard_smp_processor_id();
1535         entry1.delivery_mode = dest_ExtINT;
1536         entry1.polarity = entry0.polarity;
1537         entry1.trigger = 0;
1538         entry1.vector = 0;
1539
1540         spin_lock_irqsave(&ioapic_lock, flags);
1541         io_apic_write(0, 0x11 + 2 * pin, *(((int *)&entry1) + 1));
1542         io_apic_write(0, 0x10 + 2 * pin, *(((int *)&entry1) + 0));
1543         spin_unlock_irqrestore(&ioapic_lock, flags);
1544
1545         save_control = CMOS_READ(RTC_CONTROL);
1546         save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
1547         CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
1548                    RTC_FREQ_SELECT);
1549         CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
1550
1551         i = 100;
1552         while (i-- > 0) {
1553                 mdelay(10);
1554                 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
1555                         i -= 10;
1556         }
1557
1558         CMOS_WRITE(save_control, RTC_CONTROL);
1559         CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
1560         clear_IO_APIC_pin(0, pin);
1561
1562         spin_lock_irqsave(&ioapic_lock, flags);
1563         io_apic_write(0, 0x11 + 2 * pin, *(((int *)&entry0) + 1));
1564         io_apic_write(0, 0x10 + 2 * pin, *(((int *)&entry0) + 0));
1565         spin_unlock_irqrestore(&ioapic_lock, flags);
1566 }
1567
1568 /*
1569  * This code may look a bit paranoid, but it's supposed to cooperate with
1570  * a wide range of boards and BIOS bugs.  Fortunately only the timer IRQ
1571  * is so screwy.  Thanks to Brian Perkins for testing/hacking this beast
1572  * fanatically on his truly buggy board.
1573  */
1574 static inline void check_timer(void)
1575 {
1576         int pin1, pin2;
1577         int vector;
1578
1579         /*
1580          * get/set the timer IRQ vector:
1581          */
1582         disable_8259A_irq(0);
1583         vector = assign_irq_vector(0);
1584         set_intr_gate(vector, interrupt[0]);
1585
1586         /*
1587          * Subtle, code in do_timer_interrupt() expects an AEOI
1588          * mode for the 8259A whenever interrupts are routed
1589          * through I/O APICs.  Also IRQ0 has to be enabled in
1590          * the 8259A which implies the virtual wire has to be
1591          * disabled in the local APIC.
1592          */
1593         apic_write_around(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
1594         init_8259A(1);
1595         enable_8259A_irq(0);
1596
1597         pin1 = find_isa_irq_pin(0, mp_INT);
1598         pin2 = find_isa_irq_pin(0, mp_ExtINT);
1599
1600         apic_printk(APIC_VERBOSE,KERN_INFO "..TIMER: vector=0x%02X pin1=%d pin2=%d\n", vector, pin1, pin2);
1601
1602         if (pin1 != -1) {
1603                 /*
1604                  * Ok, does IRQ0 through the IOAPIC work?
1605                  */
1606                 unmask_IO_APIC_irq(0);
1607                 if (!no_timer_check && timer_irq_works()) {
1608                         nmi_watchdog_default();
1609                         if (nmi_watchdog == NMI_IO_APIC) {
1610                                 disable_8259A_irq(0);
1611                                 setup_nmi();
1612                                 enable_8259A_irq(0);
1613                         }
1614                         return;
1615                 }
1616                 clear_IO_APIC_pin(0, pin1);
1617                 apic_printk(APIC_QUIET,KERN_ERR "..MP-BIOS bug: 8254 timer not connected to IO-APIC\n");
1618         }
1619
1620         apic_printk(APIC_VERBOSE,KERN_INFO "...trying to set up timer (IRQ0) through the 8259A ... ");
1621         if (pin2 != -1) {
1622                 apic_printk(APIC_VERBOSE,"\n..... (found pin %d) ...", pin2);
1623                 /*
1624                  * legacy devices should be connected to IO APIC #0
1625                  */
1626                 setup_ExtINT_IRQ0_pin(pin2, vector);
1627                 if (timer_irq_works()) {
1628                         printk("works.\n");
1629                         nmi_watchdog_default();
1630                         if (nmi_watchdog == NMI_IO_APIC) {
1631                                 setup_nmi();
1632                         }
1633                         return;
1634                 }
1635                 /*
1636                  * Cleanup, just in case ...
1637                  */
1638                 clear_IO_APIC_pin(0, pin2);
1639         }
1640         printk(" failed.\n");
1641
1642         if (nmi_watchdog) {
1643                 printk(KERN_WARNING "timer doesn't work through the IO-APIC - disabling NMI Watchdog!\n");
1644                 nmi_watchdog = 0;
1645         }
1646
1647         apic_printk(APIC_VERBOSE, KERN_INFO "...trying to set up timer as Virtual Wire IRQ...");
1648
1649         disable_8259A_irq(0);
1650         irq_desc[0].handler = &lapic_irq_type;
1651         apic_write_around(APIC_LVT0, APIC_DM_FIXED | vector);   /* Fixed mode */
1652         enable_8259A_irq(0);
1653
1654         if (timer_irq_works()) {
1655                 apic_printk(APIC_QUIET, " works.\n");
1656                 return;
1657         }
1658         apic_write_around(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | vector);
1659         apic_printk(APIC_VERBOSE," failed.\n");
1660
1661         apic_printk(APIC_VERBOSE, KERN_INFO "...trying to set up timer as ExtINT IRQ...");
1662
1663         init_8259A(0);
1664         make_8259A_irq(0);
1665         apic_write_around(APIC_LVT0, APIC_DM_EXTINT);
1666
1667         unlock_ExtINT_logic();
1668
1669         if (timer_irq_works()) {
1670                 apic_printk(APIC_VERBOSE," works.\n");
1671                 return;
1672         }
1673         apic_printk(APIC_VERBOSE," failed :(.\n");
1674         panic("IO-APIC + timer doesn't work! Try using the 'noapic' kernel parameter\n");
1675 }
1676
1677 static int __init notimercheck(char *s)
1678 {
1679         no_timer_check = 1;
1680         return 1;
1681 }
1682 __setup("no_timer_check", notimercheck);
1683
1684 /*
1685  *
1686  * IRQ's that are handled by the PIC in the MPS IOAPIC case.
1687  * - IRQ2 is the cascade IRQ, and cannot be a io-apic IRQ.
1688  *   Linux doesn't really care, as it's not actually used
1689  *   for any interrupt handling anyway.
1690  */
1691 #define PIC_IRQS        (1<<2)
1692
1693 void __init setup_IO_APIC(void)
1694 {
1695         enable_IO_APIC();
1696
1697         if (acpi_ioapic)
1698                 io_apic_irqs = ~0;      /* all IRQs go through IOAPIC */
1699         else
1700                 io_apic_irqs = ~PIC_IRQS;
1701
1702         apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
1703
1704         /*
1705          * Set up the IO-APIC IRQ routing table.
1706          */
1707         if (!acpi_ioapic)
1708                 setup_ioapic_ids_from_mpc();
1709         sync_Arb_IDs();
1710         setup_IO_APIC_irqs();
1711         init_IO_APIC_traps();
1712         check_timer();
1713         if (!acpi_ioapic)
1714                 print_IO_APIC();
1715 }
1716
1717 struct sysfs_ioapic_data {
1718         struct sys_device dev;
1719         struct IO_APIC_route_entry entry[0];
1720 };
1721 static struct sysfs_ioapic_data * mp_ioapic_data[MAX_IO_APICS];
1722
1723 static int ioapic_suspend(struct sys_device *dev, pm_message_t state)
1724 {
1725         struct IO_APIC_route_entry *entry;
1726         struct sysfs_ioapic_data *data;
1727         unsigned long flags;
1728         int i;
1729
1730         data = container_of(dev, struct sysfs_ioapic_data, dev);
1731         entry = data->entry;
1732         spin_lock_irqsave(&ioapic_lock, flags);
1733         for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ ) {
1734                 *(((int *)entry) + 1) = io_apic_read(dev->id, 0x11 + 2 * i);
1735                 *(((int *)entry) + 0) = io_apic_read(dev->id, 0x10 + 2 * i);
1736         }
1737         spin_unlock_irqrestore(&ioapic_lock, flags);
1738
1739         return 0;
1740 }
1741
1742 static int ioapic_resume(struct sys_device *dev)
1743 {
1744         struct IO_APIC_route_entry *entry;
1745         struct sysfs_ioapic_data *data;
1746         unsigned long flags;
1747         union IO_APIC_reg_00 reg_00;
1748         int i;
1749
1750         data = container_of(dev, struct sysfs_ioapic_data, dev);
1751         entry = data->entry;
1752
1753         spin_lock_irqsave(&ioapic_lock, flags);
1754         reg_00.raw = io_apic_read(dev->id, 0);
1755         if (reg_00.bits.ID != mp_ioapics[dev->id].mpc_apicid) {
1756                 reg_00.bits.ID = mp_ioapics[dev->id].mpc_apicid;
1757                 io_apic_write(dev->id, 0, reg_00.raw);
1758         }
1759         for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ ) {
1760                 io_apic_write(dev->id, 0x11+2*i, *(((int *)entry)+1));
1761                 io_apic_write(dev->id, 0x10+2*i, *(((int *)entry)+0));
1762         }
1763         spin_unlock_irqrestore(&ioapic_lock, flags);
1764
1765         return 0;
1766 }
1767
1768 static struct sysdev_class ioapic_sysdev_class = {
1769         set_kset_name("ioapic"),
1770         .suspend = ioapic_suspend,
1771         .resume = ioapic_resume,
1772 };
1773
1774 static int __init ioapic_init_sysfs(void)
1775 {
1776         struct sys_device * dev;
1777         int i, size, error = 0;
1778
1779         error = sysdev_class_register(&ioapic_sysdev_class);
1780         if (error)
1781                 return error;
1782
1783         for (i = 0; i < nr_ioapics; i++ ) {
1784                 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
1785                         * sizeof(struct IO_APIC_route_entry);
1786                 mp_ioapic_data[i] = kmalloc(size, GFP_KERNEL);
1787                 if (!mp_ioapic_data[i]) {
1788                         printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
1789                         continue;
1790                 }
1791                 memset(mp_ioapic_data[i], 0, size);
1792                 dev = &mp_ioapic_data[i]->dev;
1793                 dev->id = i;
1794                 dev->cls = &ioapic_sysdev_class;
1795                 error = sysdev_register(dev);
1796                 if (error) {
1797                         kfree(mp_ioapic_data[i]);
1798                         mp_ioapic_data[i] = NULL;
1799                         printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
1800                         continue;
1801                 }
1802         }
1803
1804         return 0;
1805 }
1806
1807 device_initcall(ioapic_init_sysfs);
1808
1809 /* --------------------------------------------------------------------------
1810                           ACPI-based IOAPIC Configuration
1811    -------------------------------------------------------------------------- */
1812
1813 #ifdef CONFIG_ACPI_BOOT
1814
1815 #define IO_APIC_MAX_ID          0xFE
1816
1817 int __init io_apic_get_version (int ioapic)
1818 {
1819         union IO_APIC_reg_01    reg_01;
1820         unsigned long flags;
1821
1822         spin_lock_irqsave(&ioapic_lock, flags);
1823         reg_01.raw = io_apic_read(ioapic, 1);
1824         spin_unlock_irqrestore(&ioapic_lock, flags);
1825
1826         return reg_01.bits.version;
1827 }
1828
1829
1830 int __init io_apic_get_redir_entries (int ioapic)
1831 {
1832         union IO_APIC_reg_01    reg_01;
1833         unsigned long flags;
1834
1835         spin_lock_irqsave(&ioapic_lock, flags);
1836         reg_01.raw = io_apic_read(ioapic, 1);
1837         spin_unlock_irqrestore(&ioapic_lock, flags);
1838
1839         return reg_01.bits.entries;
1840 }
1841
1842
1843 int io_apic_set_pci_routing (int ioapic, int pin, int irq, int edge_level, int active_high_low)
1844 {
1845         struct IO_APIC_route_entry entry;
1846         unsigned long flags;
1847
1848         if (!IO_APIC_IRQ(irq)) {
1849                 apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
1850                         ioapic);
1851                 return -EINVAL;
1852         }
1853
1854         /*
1855          * Generate a PCI IRQ routing entry and program the IOAPIC accordingly.
1856          * Note that we mask (disable) IRQs now -- these get enabled when the
1857          * corresponding device driver registers for this IRQ.
1858          */
1859
1860         memset(&entry,0,sizeof(entry));
1861
1862         entry.delivery_mode = INT_DELIVERY_MODE;
1863         entry.dest_mode = INT_DEST_MODE;
1864         entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
1865         entry.trigger = edge_level;
1866         entry.polarity = active_high_low;
1867         entry.mask = 1;                                  /* Disabled (masked) */
1868
1869         /*
1870          * IRQs < 16 are already in the irq_2_pin[] map
1871          */
1872         if (irq >= 16)
1873                 add_pin_to_irq(irq, ioapic, pin);
1874
1875         entry.vector = assign_irq_vector(irq);
1876
1877         apic_printk(APIC_VERBOSE,KERN_DEBUG "IOAPIC[%d]: Set PCI routing entry (%d-%d -> 0x%x -> "
1878                 "IRQ %d Mode:%i Active:%i)\n", ioapic, 
1879                mp_ioapics[ioapic].mpc_apicid, pin, entry.vector, irq,
1880                edge_level, active_high_low);
1881
1882         ioapic_register_intr(irq, entry.vector, edge_level);
1883
1884         if (!ioapic && (irq < 16))
1885                 disable_8259A_irq(irq);
1886
1887         spin_lock_irqsave(&ioapic_lock, flags);
1888         io_apic_write(ioapic, 0x11+2*pin, *(((int *)&entry)+1));
1889         io_apic_write(ioapic, 0x10+2*pin, *(((int *)&entry)+0));
1890         spin_unlock_irqrestore(&ioapic_lock, flags);
1891
1892         return 0;
1893 }
1894
1895 #endif /*CONFIG_ACPI_BOOT*/
1896
1897
1898 /*
1899  * This function currently is only a helper for the i386 smp boot process where
1900  * we need to reprogram the ioredtbls to cater for the cpus which have come online
1901  * so mask in all cases should simply be TARGET_CPUS
1902  */
1903 void __init setup_ioapic_dest(void)
1904 {
1905         int pin, ioapic, irq, irq_entry;
1906
1907         if (skip_ioapic_setup == 1)
1908                 return;
1909
1910         for (ioapic = 0; ioapic < nr_ioapics; ioapic++) {
1911                 for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
1912                         irq_entry = find_irq_entry(ioapic, pin, mp_INT);
1913                         if (irq_entry == -1)
1914                                 continue;
1915                         irq = pin_2_irq(irq_entry, ioapic, pin);
1916                         set_ioapic_affinity_irq(irq, TARGET_CPUS);
1917                 }
1918
1919         }
1920 }