2 * Intel IO-APIC support for multi-Pentium hosts.
4 * Copyright (C) 1997, 1998, 1999, 2000 Ingo Molnar, Hajnalka Szabo
6 * Many thanks to Stig Venaas for trying out countless experimental
7 * patches and reporting/debugging problems patiently!
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>
16 * Maciej W. Rozycki : Bits for genuine 82489DX APICs;
17 * thanks to Eric Gilmore
19 * for testing these extensively
20 * Paul Diefenbaugh : Added full ACPI support
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>
36 #include <acpi/acpi_bus.h>
42 #include <asm/proto.h>
43 #include <asm/mach_apic.h>
47 #include <asm/msidef.h>
48 #include <asm/hypertransport.h>
50 static int assign_irq_vector(int irq
, cpumask_t mask
, cpumask_t
*result
);
52 #define __apicdebuginit __init
54 int sis_apic_bug
; /* not actually supported, dummy for compile */
56 static int no_timer_check
;
58 static int disable_timer_pin_1 __initdata
;
60 int timer_over_8254 __initdata
= 1;
62 /* Where if anywhere is the i8259 connect in external int mode */
63 static struct { int pin
, apic
; } ioapic_i8259
= { -1, -1 };
65 static DEFINE_SPINLOCK(ioapic_lock
);
66 DEFINE_SPINLOCK(vector_lock
);
69 * # of IRQ routing registers
71 int nr_ioapic_registers
[MAX_IO_APICS
];
74 * Rough estimation of how many shared IRQs there are, can
77 #define MAX_PLUS_SHARED_IRQS NR_IRQ_VECTORS
78 #define PIN_MAP_SIZE (MAX_PLUS_SHARED_IRQS + NR_IRQS)
81 * This is performance-critical, we want to do it O(1)
83 * the indexing order of this array favors 1:1 mappings
84 * between pins and IRQs.
87 static struct irq_pin_list
{
88 short apic
, pin
, next
;
89 } irq_2_pin
[PIN_MAP_SIZE
];
93 unsigned int unused
[3];
97 static __attribute_const__
struct io_apic __iomem
*io_apic_base(int idx
)
99 return (void __iomem
*) __fix_to_virt(FIX_IO_APIC_BASE_0
+ idx
)
100 + (mp_ioapics
[idx
].mpc_apicaddr
& ~PAGE_MASK
);
103 static inline unsigned int io_apic_read(unsigned int apic
, unsigned int reg
)
105 struct io_apic __iomem
*io_apic
= io_apic_base(apic
);
106 writel(reg
, &io_apic
->index
);
107 return readl(&io_apic
->data
);
110 static inline void io_apic_write(unsigned int apic
, unsigned int reg
, unsigned int value
)
112 struct io_apic __iomem
*io_apic
= io_apic_base(apic
);
113 writel(reg
, &io_apic
->index
);
114 writel(value
, &io_apic
->data
);
118 * Re-write a value: to be used for read-modify-write
119 * cycles where the read already set up the index register.
121 static inline void io_apic_modify(unsigned int apic
, unsigned int value
)
123 struct io_apic __iomem
*io_apic
= io_apic_base(apic
);
124 writel(value
, &io_apic
->data
);
128 * Synchronize the IO-APIC and the CPU by doing
129 * a dummy read from the IO-APIC
131 static inline void io_apic_sync(unsigned int apic
)
133 struct io_apic __iomem
*io_apic
= io_apic_base(apic
);
134 readl(&io_apic
->data
);
137 #define __DO_ACTION(R, ACTION, FINAL) \
141 struct irq_pin_list *entry = irq_2_pin + irq; \
143 BUG_ON(irq >= NR_IRQS); \
149 reg = io_apic_read(entry->apic, 0x10 + R + pin*2); \
151 io_apic_modify(entry->apic, reg); \
154 entry = irq_2_pin + entry->next; \
160 struct { u32 w1
, w2
; };
161 struct IO_APIC_route_entry entry
;
164 static struct IO_APIC_route_entry
ioapic_read_entry(int apic
, int pin
)
166 union entry_union eu
;
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
);
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.
181 static void ioapic_write_entry(int apic
, int pin
, struct IO_APIC_route_entry e
)
184 union entry_union eu
;
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
);
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
197 static void ioapic_mask_entry(int apic
, int pin
)
200 union entry_union eu
= { .entry
.mask
= 1 };
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
);
209 static void __target_IO_APIC_irq(unsigned int irq
, unsigned int dest
, u8 vector
)
212 struct irq_pin_list
*entry
= irq_2_pin
+ irq
;
214 BUG_ON(irq
>= NR_IRQS
);
221 io_apic_write(apic
, 0x11 + pin
*2, dest
);
222 reg
= io_apic_read(apic
, 0x10 + pin
*2);
225 io_apic_modify(apic
, reg
);
228 entry
= irq_2_pin
+ entry
->next
;
232 static void set_ioapic_affinity_irq(unsigned int irq
, cpumask_t mask
)
239 cpus_and(tmp
, mask
, cpu_online_map
);
243 cpus_and(mask
, tmp
, CPU_MASK_ALL
);
245 vector
= assign_irq_vector(irq
, mask
, &tmp
);
249 dest
= cpu_mask_to_apicid(tmp
);
252 * Only the high 8 bits are valid.
254 dest
= SET_APIC_LOGICAL_ID(dest
);
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
);
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.
268 static void add_pin_to_irq(unsigned int irq
, int apic
, int pin
)
270 static int first_free_entry
= NR_IRQS
;
271 struct irq_pin_list
*entry
= irq_2_pin
+ irq
;
273 BUG_ON(irq
>= NR_IRQS
);
275 entry
= irq_2_pin
+ entry
->next
;
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!");
288 #define DO_ACTION(name,R,ACTION, FINAL) \
290 static void name##_IO_APIC_irq (unsigned int irq) \
291 __DO_ACTION(R, ACTION, FINAL)
293 DO_ACTION( __mask
, 0, |= 0x00010000, io_apic_sync(entry
->apic
) )
295 DO_ACTION( __unmask
, 0, &= 0xfffeffff, )
298 static void mask_IO_APIC_irq (unsigned int irq
)
302 spin_lock_irqsave(&ioapic_lock
, flags
);
303 __mask_IO_APIC_irq(irq
);
304 spin_unlock_irqrestore(&ioapic_lock
, flags
);
307 static void unmask_IO_APIC_irq (unsigned int irq
)
311 spin_lock_irqsave(&ioapic_lock
, flags
);
312 __unmask_IO_APIC_irq(irq
);
313 spin_unlock_irqrestore(&ioapic_lock
, flags
);
316 static void clear_IO_APIC_pin(unsigned int apic
, unsigned int pin
)
318 struct IO_APIC_route_entry entry
;
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
)
325 * Disable it in the IO-APIC irq-routing table:
327 ioapic_mask_entry(apic
, pin
);
330 static void clear_IO_APIC (void)
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
);
339 int skip_ioapic_setup
;
342 /* dummy parsing: see setup.c */
344 static int __init
disable_ioapic_setup(char *str
)
346 skip_ioapic_setup
= 1;
349 early_param("noapic", disable_ioapic_setup
);
351 /* Actually the next is obsolete, but keep it for paranoid reasons -AK */
352 static int __init
disable_timer_pin_setup(char *arg
)
354 disable_timer_pin_1
= 1;
357 __setup("disable_timer_pin_1", disable_timer_pin_setup
);
359 static int __init
setup_disable_8254_timer(char *s
)
361 timer_over_8254
= -1;
364 static int __init
setup_enable_8254_timer(char *s
)
370 __setup("disable_8254_timer", setup_disable_8254_timer
);
371 __setup("enable_8254_timer", setup_enable_8254_timer
);
375 * Find the IRQ entry number of a certain pin.
377 static int find_irq_entry(int apic
, int pin
, int type
)
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
)
392 * Find the pin to which IRQ[irq] (ISA) is connected
394 static int __init
find_isa_irq_pin(int irq
, int type
)
398 for (i
= 0; i
< mp_irq_entries
; i
++) {
399 int lbus
= mp_irqs
[i
].mpc_srcbus
;
401 if (test_bit(lbus
, mp_bus_not_pci
) &&
402 (mp_irqs
[i
].mpc_irqtype
== type
) &&
403 (mp_irqs
[i
].mpc_srcbusirq
== irq
))
405 return mp_irqs
[i
].mpc_dstirq
;
410 static int __init
find_isa_irq_apic(int irq
, int type
)
414 for (i
= 0; i
< mp_irq_entries
; i
++) {
415 int lbus
= mp_irqs
[i
].mpc_srcbus
;
417 if (test_bit(lbus
, mp_bus_not_pci
) &&
418 (mp_irqs
[i
].mpc_irqtype
== type
) &&
419 (mp_irqs
[i
].mpc_srcbusirq
== irq
))
422 if (i
< mp_irq_entries
) {
424 for(apic
= 0; apic
< nr_ioapics
; apic
++) {
425 if (mp_ioapics
[apic
].mpc_apicid
== mp_irqs
[i
].mpc_dstapic
)
434 * Find a specific PCI IRQ entry.
435 * Not an __init, possibly needed by modules
437 static int pin_2_irq(int idx
, int apic
, int pin
);
439 int IO_APIC_get_PCI_irq_vector(int bus
, int slot
, int pin
)
441 int apic
, i
, best_guess
= -1;
443 apic_printk(APIC_DEBUG
, "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
445 if (mp_bus_id_to_pci_bus
[bus
] == -1) {
446 apic_printk(APIC_VERBOSE
, "PCI BIOS passed nonexistent PCI bus %d!\n", bus
);
449 for (i
= 0; i
< mp_irq_entries
; i
++) {
450 int lbus
= mp_irqs
[i
].mpc_srcbus
;
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
)
457 if (!test_bit(lbus
, mp_bus_not_pci
) &&
458 !mp_irqs
[i
].mpc_irqtype
&&
460 (slot
== ((mp_irqs
[i
].mpc_srcbusirq
>> 2) & 0x1f))) {
461 int irq
= pin_2_irq(i
,apic
,mp_irqs
[i
].mpc_dstirq
);
463 if (!(apic
|| IO_APIC_IRQ(irq
)))
466 if (pin
== (mp_irqs
[i
].mpc_srcbusirq
& 3))
469 * Use the first all-but-pin matching entry as a
470 * best-guess fuzzy result for broken mptables.
476 BUG_ON(best_guess
>= NR_IRQS
);
480 /* ISA interrupts are always polarity zero edge triggered,
481 * when listed as conforming in the MP table. */
483 #define default_ISA_trigger(idx) (0)
484 #define default_ISA_polarity(idx) (0)
486 /* PCI interrupts are always polarity one level triggered,
487 * when listed as conforming in the MP table. */
489 #define default_PCI_trigger(idx) (1)
490 #define default_PCI_polarity(idx) (1)
492 static int __init
MPBIOS_polarity(int idx
)
494 int bus
= mp_irqs
[idx
].mpc_srcbus
;
498 * Determine IRQ line polarity (high active or low active):
500 switch (mp_irqs
[idx
].mpc_irqflag
& 3)
502 case 0: /* conforms, ie. bus-type dependent polarity */
503 if (test_bit(bus
, mp_bus_not_pci
))
504 polarity
= default_ISA_polarity(idx
);
506 polarity
= default_PCI_polarity(idx
);
508 case 1: /* high active */
513 case 2: /* reserved */
515 printk(KERN_WARNING
"broken BIOS!!\n");
519 case 3: /* low active */
524 default: /* invalid */
526 printk(KERN_WARNING
"broken BIOS!!\n");
534 static int MPBIOS_trigger(int idx
)
536 int bus
= mp_irqs
[idx
].mpc_srcbus
;
540 * Determine IRQ trigger mode (edge or level sensitive):
542 switch ((mp_irqs
[idx
].mpc_irqflag
>>2) & 3)
544 case 0: /* conforms, ie. bus-type dependent */
545 if (test_bit(bus
, mp_bus_not_pci
))
546 trigger
= default_ISA_trigger(idx
);
548 trigger
= default_PCI_trigger(idx
);
555 case 2: /* reserved */
557 printk(KERN_WARNING
"broken BIOS!!\n");
566 default: /* invalid */
568 printk(KERN_WARNING
"broken BIOS!!\n");
576 static inline int irq_polarity(int idx
)
578 return MPBIOS_polarity(idx
);
581 static inline int irq_trigger(int idx
)
583 return MPBIOS_trigger(idx
);
586 static int pin_2_irq(int idx
, int apic
, int pin
)
589 int bus
= mp_irqs
[idx
].mpc_srcbus
;
592 * Debugging check, we are in big trouble if this message pops up!
594 if (mp_irqs
[idx
].mpc_dstirq
!= pin
)
595 printk(KERN_ERR
"broken BIOS or MPTABLE parser, ayiee!!\n");
597 if (test_bit(bus
, mp_bus_not_pci
)) {
598 irq
= mp_irqs
[idx
].mpc_srcbusirq
;
601 * PCI IRQs are mapped in order
605 irq
+= nr_ioapic_registers
[i
++];
608 BUG_ON(irq
>= NR_IRQS
);
612 static inline int IO_APIC_irq_trigger(int irq
)
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
);
624 * nonexistent IRQs are edge default
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,
649 static cpumask_t irq_domain
[NR_IRQ_VECTORS
] __read_mostly
= {
668 static int __assign_irq_vector(int irq
, cpumask_t mask
, cpumask_t
*result
)
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.
678 * Also, we've got to be careful not to trash gate
679 * 0x80, because int 0x80 is hm, kind of importantish. ;)
681 static int current_vector
= FIRST_DEVICE_VECTOR
, current_offset
= 0;
685 BUG_ON((unsigned)irq
>= NR_IRQ_VECTORS
);
687 /* Only try and allocate irqs on cpus that are present */
688 cpus_and(mask
, mask
, cpu_online_map
);
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
))
698 for_each_cpu_mask(cpu
, mask
) {
699 cpumask_t domain
, new_mask
;
703 domain
= vector_allocation_domain(cpu
);
704 cpus_and(new_mask
, domain
, cpu_online_map
);
706 vector
= current_vector
;
707 offset
= current_offset
;
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
;
715 if (unlikely(current_vector
== vector
))
717 if (vector
== IA32_SYSCALL_VECTOR
)
719 for_each_cpu_mask(new_cpu
, new_mask
)
720 if (per_cpu(vector_irq
, new_cpu
)[vector
] != -1)
723 current_vector
= vector
;
724 current_offset
= offset
;
725 if (old_vector
>= 0) {
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;
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
);
742 static int assign_irq_vector(int irq
, cpumask_t mask
, cpumask_t
*result
)
747 spin_lock_irqsave(&vector_lock
, flags
);
748 vector
= __assign_irq_vector(irq
, mask
, result
);
749 spin_unlock_irqrestore(&vector_lock
, flags
);
753 void __setup_vector_irq(int cpu
)
755 /* Initialize vector_irq on a new cpu */
756 /* This function must be called with vector_lock held */
759 /* Mark the inuse vectors */
760 for (irq
= 0; irq
< NR_IRQ_VECTORS
; ++irq
) {
761 if (!cpu_isset(cpu
, irq_domain
[irq
]))
763 vector
= irq_vector
[irq
];
764 per_cpu(vector_irq
, cpu
)[vector
] = irq
;
766 /* Mark the free vectors */
767 for (vector
= 0; vector
< NR_VECTORS
; ++vector
) {
768 irq
= per_cpu(vector_irq
, cpu
)[vector
];
771 if (!cpu_isset(cpu
, irq_domain
[irq
]))
772 per_cpu(vector_irq
, cpu
)[vector
] = -1;
777 extern void (*interrupt
[NR_IRQS
])(void);
779 static struct irq_chip ioapic_chip
;
781 #define IOAPIC_AUTO -1
782 #define IOAPIC_EDGE 0
783 #define IOAPIC_LEVEL 1
785 static void ioapic_register_intr(int irq
, int vector
, unsigned long trigger
)
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");
792 irq_desc
[irq
].status
|= IRQ_DELAYED_DISABLE
;
793 set_irq_chip_and_handler_name(irq
, &ioapic_chip
,
794 handle_edge_irq
, "edge");
798 static void __init
setup_IO_APIC_irqs(void)
800 struct IO_APIC_route_entry entry
;
801 int apic
, pin
, idx
, irq
, first_notcon
= 1, vector
;
804 apic_printk(APIC_VERBOSE
, KERN_DEBUG
"init IO_APIC IRQs\n");
806 for (apic
= 0; apic
< nr_ioapics
; apic
++) {
807 for (pin
= 0; pin
< nr_ioapic_registers
[apic
]; pin
++) {
810 * add it to the IO-APIC irq-routing table:
812 memset(&entry
,0,sizeof(entry
));
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
);
819 idx
= find_irq_entry(apic
,pin
,mp_INT
);
822 apic_printk(APIC_VERBOSE
, KERN_DEBUG
" IO-APIC (apicid-pin) %d-%d", mp_ioapics
[apic
].mpc_apicid
, pin
);
825 apic_printk(APIC_VERBOSE
, ", %d-%d", mp_ioapics
[apic
].mpc_apicid
, pin
);
829 entry
.trigger
= irq_trigger(idx
);
830 entry
.polarity
= irq_polarity(idx
);
832 if (irq_trigger(idx
)) {
835 entry
.dest
.logical
.logical_dest
= cpu_mask_to_apicid(TARGET_CPUS
);
838 irq
= pin_2_irq(idx
, apic
, pin
);
839 add_pin_to_irq(irq
, apic
, pin
);
841 if (!apic
&& !IO_APIC_IRQ(irq
))
844 if (IO_APIC_IRQ(irq
)) {
846 vector
= assign_irq_vector(irq
, TARGET_CPUS
, &mask
);
850 entry
.dest
.logical
.logical_dest
= cpu_mask_to_apicid(mask
);
851 entry
.vector
= vector
;
853 ioapic_register_intr(irq
, vector
, IOAPIC_AUTO
);
854 if (!apic
&& (irq
< 16))
855 disable_8259A_irq(irq
);
857 ioapic_write_entry(apic
, pin
, entry
);
859 spin_lock_irqsave(&ioapic_lock
, flags
);
860 set_native_irq_info(irq
, TARGET_CPUS
);
861 spin_unlock_irqrestore(&ioapic_lock
, flags
);
866 apic_printk(APIC_VERBOSE
," not connected.\n");
870 * Set up the 8259A-master output pin as broadcast to all
873 static void __init
setup_ExtINT_IRQ0_pin(unsigned int apic
, unsigned int pin
, int vector
)
875 struct IO_APIC_route_entry entry
;
878 memset(&entry
,0,sizeof(entry
));
880 disable_8259A_irq(0);
883 apic_write(APIC_LVT0
, APIC_LVT_MASKED
| APIC_DM_EXTINT
);
886 * We use logical delivery to get the timer IRQ
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
;
895 entry
.vector
= vector
;
898 * The timer IRQ doesn't have to know that behind the
899 * scene we have a 8259A-master in AEOI mode ...
901 set_irq_chip_and_handler_name(0, &ioapic_chip
, handle_edge_irq
, "edge");
904 * Add it to the IO-APIC irq-routing table:
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
);
914 void __init
UNEXPECTED_IO_APIC(void)
918 void __apicdebuginit
print_IO_APIC(void)
921 union IO_APIC_reg_00 reg_00
;
922 union IO_APIC_reg_01 reg_01
;
923 union IO_APIC_reg_02 reg_02
;
926 if (apic_verbosity
== APIC_QUIET
)
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
]);
935 * We are a bit conservative about what we expect. We have to
936 * know about every hardware change ASAP.
938 printk(KERN_INFO
"testing the IO APIC.......................\n");
940 for (apic
= 0; apic
< nr_ioapics
; apic
++) {
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
);
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();
956 printk(KERN_DEBUG
".... register #01: %08X\n", *(int *)®_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)
967 UNEXPECTED_IO_APIC();
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) */
978 UNEXPECTED_IO_APIC();
979 if (reg_01
.bits
.__reserved_1
|| reg_01
.bits
.__reserved_2
)
980 UNEXPECTED_IO_APIC();
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();
989 printk(KERN_DEBUG
".... IRQ redirection table:\n");
991 printk(KERN_DEBUG
" NR Log Phy Mask Trig IRR Pol"
992 " Stat Dest Deli Vect: \n");
994 for (i
= 0; i
<= reg_01
.bits
.entries
; i
++) {
995 struct IO_APIC_route_entry entry
;
997 entry
= ioapic_read_entry(apic
, i
);
999 printk(KERN_DEBUG
" %02x %03X %02X ",
1001 entry
.dest
.logical
.logical_dest
,
1002 entry
.dest
.physical
.physical_dest
1005 printk("%1d %1d %1d %1d %1d %1d %1d %02X\n",
1010 entry
.delivery_status
,
1012 entry
.delivery_mode
,
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
;
1022 printk(KERN_DEBUG
"IRQ%d ", i
);
1024 printk("-> %d:%d", entry
->apic
, entry
->pin
);
1027 entry
= irq_2_pin
+ entry
->next
;
1032 printk(KERN_INFO
".................................... done.\n");
1039 static __apicdebuginit
void print_APIC_bitfield (int base
)
1044 if (apic_verbosity
== APIC_QUIET
)
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
++) {
1060 void __apicdebuginit
print_local_APIC(void * dummy
)
1062 unsigned int v
, ver
, maxlvt
;
1064 if (apic_verbosity
== APIC_QUIET
)
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();
1076 v
= apic_read(APIC_TASKPRI
);
1077 printk(KERN_DEBUG
"... APIC TASKPRI: %08x (%02x)\n", v
, v
& APIC_TPRI_MASK
);
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
);
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
);
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
);
1103 v
= apic_read(APIC_ESR
);
1104 printk(KERN_DEBUG
"... APIC ESR: %08x\n", v
);
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
);
1111 v
= apic_read(APIC_LVTT
);
1112 printk(KERN_DEBUG
"... APIC LVTT: %08x\n", v
);
1114 if (maxlvt
> 3) { /* PC is LVT#4. */
1115 v
= apic_read(APIC_LVTPC
);
1116 printk(KERN_DEBUG
"... APIC LVTPC: %08x\n", v
);
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
);
1123 if (maxlvt
> 2) { /* ERR is LVT#3. */
1124 v
= apic_read(APIC_LVTERR
);
1125 printk(KERN_DEBUG
"... APIC LVTERR: %08x\n", v
);
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
);
1137 void print_all_local_APICs (void)
1139 on_each_cpu(print_local_APIC
, NULL
, 1, 1);
1142 void __apicdebuginit
print_PIC(void)
1145 unsigned long flags
;
1147 if (apic_verbosity
== APIC_QUIET
)
1150 printk(KERN_DEBUG
"\nprinting PIC contents\n");
1152 spin_lock_irqsave(&i8259A_lock
, flags
);
1154 v
= inb(0xa1) << 8 | inb(0x21);
1155 printk(KERN_DEBUG
"... PIC IMR: %04x\n", v
);
1157 v
= inb(0xa0) << 8 | inb(0x20);
1158 printk(KERN_DEBUG
"... PIC IRR: %04x\n", v
);
1162 v
= inb(0xa0) << 8 | inb(0x20);
1166 spin_unlock_irqrestore(&i8259A_lock
, flags
);
1168 printk(KERN_DEBUG
"... PIC ISR: %04x\n", v
);
1170 v
= inb(0x4d1) << 8 | inb(0x4d0);
1171 printk(KERN_DEBUG
"... PIC ELCR: %04x\n", v
);
1176 static void __init
enable_IO_APIC(void)
1178 union IO_APIC_reg_01 reg_01
;
1179 int i8259_apic
, i8259_pin
;
1181 unsigned long flags
;
1183 for (i
= 0; i
< PIN_MAP_SIZE
; i
++) {
1184 irq_2_pin
[i
].pin
= -1;
1185 irq_2_pin
[i
].next
= 0;
1189 * The number of IO-APIC IRQ registers (== #pins):
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;
1197 for(apic
= 0; apic
< nr_ioapics
; apic
++) {
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
);
1204 /* If the interrupt line is enabled and in ExtInt mode
1205 * I have found the pin where the i8259 is connected.
1207 if ((entry
.mask
== 0) && (entry
.delivery_mode
== dest_ExtINT
)) {
1208 ioapic_i8259
.apic
= apic
;
1209 ioapic_i8259
.pin
= pin
;
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
;
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))
1228 printk(KERN_WARNING
"ExtINT in hardware and MP table differ\n");
1232 * Do not trust the IO-APIC being empty at bootup
1238 * Not an __init, needed by the reboot code
1240 void disable_IO_APIC(void)
1243 * Clear the IO-APIC before rebooting:
1248 * If the i8259 is routed through an IOAPIC
1249 * Put that IOAPIC in virtual wire mode
1250 * so legacy interrupts can be delivered.
1252 if (ioapic_i8259
.pin
!= -1) {
1253 struct IO_APIC_route_entry entry
;
1255 memset(&entry
, 0, sizeof(entry
));
1256 entry
.mask
= 0; /* Enabled */
1257 entry
.trigger
= 0; /* Edge */
1259 entry
.polarity
= 0; /* High */
1260 entry
.delivery_status
= 0;
1261 entry
.dest_mode
= 0; /* Physical */
1262 entry
.delivery_mode
= dest_ExtINT
; /* ExtInt */
1264 entry
.dest
.physical
.physical_dest
=
1265 GET_APIC_ID(apic_read(APIC_ID
));
1268 * Add it to the IO-APIC irq-routing table:
1270 ioapic_write_entry(ioapic_i8259
.apic
, ioapic_i8259
.pin
, entry
);
1273 disconnect_bsp_APIC(ioapic_i8259
.pin
!= -1);
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:
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
1284 static int __init
timer_irq_works(void)
1286 unsigned long t1
= jiffies
;
1289 /* Let ten ticks pass... */
1290 mdelay((10 * 1000) / HZ
);
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.
1301 if (jiffies
- t1
> 4)
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.
1314 * Edge triggered needs to resend any interrupt
1315 * that was delayed but this is now handled in the device
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.
1325 * This is not complete - we should be able to fake
1326 * an edge even if it isn't on the 8259A...
1329 static unsigned int startup_ioapic_irq(unsigned int irq
)
1331 int was_pending
= 0;
1332 unsigned long flags
;
1334 spin_lock_irqsave(&ioapic_lock
, flags
);
1336 disable_8259A_irq(irq
);
1337 if (i8259A_irq_pending(irq
))
1340 __unmask_IO_APIC_irq(irq
);
1341 spin_unlock_irqrestore(&ioapic_lock
, flags
);
1346 static int ioapic_retrigger_irq(unsigned int irq
)
1350 unsigned long flags
;
1352 spin_lock_irqsave(&vector_lock
, flags
);
1353 vector
= irq_vector
[irq
];
1355 cpu_set(first_cpu(irq_domain
[irq
]), mask
);
1357 send_IPI_mask(mask
, vector
);
1358 spin_unlock_irqrestore(&vector_lock
, flags
);
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
1372 static void ack_apic_edge(unsigned int irq
)
1374 move_native_irq(irq
);
1378 static void ack_apic_level(unsigned int irq
)
1380 int do_unmask_irq
= 0;
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
)) {
1386 mask_IO_APIC_irq(irq
);
1391 * We must acknowledge the irq before we move it or the acknowledge will
1392 * not propogate properly.
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
);
1402 static struct irq_chip ioapic_chip __read_mostly
= {
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
,
1410 .set_affinity
= set_ioapic_affinity_irq
,
1412 .retrigger
= ioapic_retrigger_irq
,
1415 static inline void init_IO_APIC_traps(void)
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.
1427 * Also, we've got to be careful not to trash gate
1428 * 0x80, because int 0x80 is hm, kind of importantish. ;)
1430 for (irq
= 0; irq
< NR_IRQS
; irq
++) {
1432 if (IO_APIC_IRQ(tmp
) && !irq_vector
[tmp
]) {
1434 * Hmm.. We don't have an entry for this,
1435 * so default to an old-fashioned 8259
1436 * interrupt if we can..
1439 make_8259A_irq(irq
);
1441 /* Strange. Oh, well.. */
1442 irq_desc
[irq
].chip
= &no_irq_chip
;
1447 static void enable_lapic_irq (unsigned int irq
)
1451 v
= apic_read(APIC_LVT0
);
1452 apic_write(APIC_LVT0
, v
& ~APIC_LVT_MASKED
);
1455 static void disable_lapic_irq (unsigned int irq
)
1459 v
= apic_read(APIC_LVT0
);
1460 apic_write(APIC_LVT0
, v
| APIC_LVT_MASKED
);
1463 static void ack_lapic_irq (unsigned int irq
)
1468 static void end_lapic_irq (unsigned int i
) { /* nothing */ }
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
,
1480 static void setup_nmi (void)
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.
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.
1491 printk(KERN_INFO
"activating NMI Watchdog ...");
1493 enable_NMI_through_LVT0(NULL
);
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
1505 static inline void unlock_ExtINT_logic(void)
1508 struct IO_APIC_route_entry entry0
, entry1
;
1509 unsigned char save_control
, save_freq_select
;
1510 unsigned long flags
;
1512 pin
= find_isa_irq_pin(8, mp_INT
);
1513 apic
= find_isa_irq_apic(8, mp_INT
);
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
);
1523 memset(&entry1
, 0, sizeof(entry1
));
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
;
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
);
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,
1542 CMOS_WRITE(save_control
| RTC_PIE
, RTC_CONTROL
);
1547 if ((CMOS_READ(RTC_INTR_FLAGS
) & RTC_PF
) == RTC_PF
)
1551 CMOS_WRITE(save_control
, RTC_CONTROL
);
1552 CMOS_WRITE(save_freq_select
, RTC_FREQ_SELECT
);
1553 clear_IO_APIC_pin(apic
, pin
);
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
);
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.
1567 * FIXME: really need to revamp this for modern platforms only.
1569 static inline void check_timer(void)
1571 int apic1
, pin1
, apic2
, pin2
;
1576 * get/set the timer IRQ vector:
1578 disable_8259A_irq(0);
1579 vector
= assign_irq_vector(0, TARGET_CPUS
, &mask
);
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.
1588 apic_write(APIC_LVT0
, APIC_LVT_MASKED
| APIC_DM_EXTINT
);
1590 if (timer_over_8254
> 0)
1591 enable_8259A_irq(0);
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
;
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
);
1603 * Ok, does IRQ0 through the IOAPIC work?
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);
1611 enable_8259A_irq(0);
1613 if (disable_timer_pin_1
> 0)
1614 clear_IO_APIC_pin(0, pin1
);
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");
1622 apic_printk(APIC_VERBOSE
,KERN_INFO
"...trying to set up timer (IRQ0) "
1623 "through the 8259A ... ");
1625 apic_printk(APIC_VERBOSE
,"\n..... (found apic %d pin %d) ...",
1628 * legacy devices should be connected to IO APIC #0
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
) {
1640 * Cleanup, just in case ...
1642 clear_IO_APIC_pin(apic2
, pin2
);
1644 apic_printk(APIC_VERBOSE
," failed.\n");
1646 if (nmi_watchdog
== NMI_IO_APIC
) {
1647 printk(KERN_WARNING
"timer doesn't work through the IO-APIC - disabling NMI Watchdog!\n");
1651 apic_printk(APIC_VERBOSE
, KERN_INFO
"...trying to set up timer as Virtual Wire IRQ...");
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);
1658 if (timer_irq_works()) {
1659 apic_printk(APIC_VERBOSE
," works.\n");
1662 apic_write(APIC_LVT0
, APIC_LVT_MASKED
| APIC_DM_FIXED
| vector
);
1663 apic_printk(APIC_VERBOSE
," failed.\n");
1665 apic_printk(APIC_VERBOSE
, KERN_INFO
"...trying to set up timer as ExtINT IRQ...");
1669 apic_write(APIC_LVT0
, APIC_DM_EXTINT
);
1671 unlock_ExtINT_logic();
1673 if (timer_irq_works()) {
1674 apic_printk(APIC_VERBOSE
," works.\n");
1677 apic_printk(APIC_VERBOSE
," failed :(.\n");
1678 panic("IO-APIC + timer doesn't work! Try using the 'noapic' kernel parameter\n");
1681 static int __init
notimercheck(char *s
)
1686 __setup("no_timer_check", notimercheck
);
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.
1695 #define PIC_IRQS (1<<2)
1697 void __init
setup_IO_APIC(void)
1702 io_apic_irqs
= ~0; /* all IRQs go through IOAPIC */
1704 io_apic_irqs
= ~PIC_IRQS
;
1706 apic_printk(APIC_VERBOSE
, "ENABLING IO-APIC IRQs\n");
1709 setup_IO_APIC_irqs();
1710 init_IO_APIC_traps();
1716 struct sysfs_ioapic_data
{
1717 struct sys_device dev
;
1718 struct IO_APIC_route_entry entry
[0];
1720 static struct sysfs_ioapic_data
* mp_ioapic_data
[MAX_IO_APICS
];
1722 static int ioapic_suspend(struct sys_device
*dev
, pm_message_t state
)
1724 struct IO_APIC_route_entry
*entry
;
1725 struct sysfs_ioapic_data
*data
;
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
);
1736 static int ioapic_resume(struct sys_device
*dev
)
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
;
1744 data
= container_of(dev
, struct sysfs_ioapic_data
, dev
);
1745 entry
= data
->entry
;
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
);
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
]);
1760 static struct sysdev_class ioapic_sysdev_class
= {
1761 set_kset_name("ioapic"),
1762 .suspend
= ioapic_suspend
,
1763 .resume
= ioapic_resume
,
1766 static int __init
ioapic_init_sysfs(void)
1768 struct sys_device
* dev
;
1769 int i
, size
, error
= 0;
1771 error
= sysdev_class_register(&ioapic_sysdev_class
);
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
);
1783 memset(mp_ioapic_data
[i
], 0, size
);
1784 dev
= &mp_ioapic_data
[i
]->dev
;
1786 dev
->cls
= &ioapic_sysdev_class
;
1787 error
= sysdev_register(dev
);
1789 kfree(mp_ioapic_data
[i
]);
1790 mp_ioapic_data
[i
] = NULL
;
1791 printk(KERN_ERR
"Can't suspend/resume IOAPIC %d\n", i
);
1799 device_initcall(ioapic_init_sysfs
);
1802 * Dynamic irq allocate and deallocation
1804 int create_irq(void)
1806 /* Allocate an unused irq */
1810 unsigned long flags
;
1814 spin_lock_irqsave(&vector_lock
, flags
);
1815 for (new = (NR_IRQS
- 1); new >= 0; new--) {
1816 if (platform_legacy_irq(new))
1818 if (irq_vector
[new] != 0)
1820 vector
= __assign_irq_vector(new, TARGET_CPUS
, &mask
);
1821 if (likely(vector
> 0))
1825 spin_unlock_irqrestore(&vector_lock
, flags
);
1828 dynamic_irq_init(irq
);
1833 void destroy_irq(unsigned int irq
)
1835 unsigned long flags
;
1837 dynamic_irq_cleanup(irq
);
1839 spin_lock_irqsave(&vector_lock
, flags
);
1840 irq_vector
[irq
] = 0;
1841 spin_unlock_irqrestore(&vector_lock
, flags
);
1845 * MSI mesage composition
1847 #ifdef CONFIG_PCI_MSI
1848 static int msi_compose_msg(struct pci_dev
*pdev
, unsigned int irq
, struct msi_msg
*msg
)
1854 vector
= assign_irq_vector(irq
, TARGET_CPUS
, &tmp
);
1856 dest
= cpu_mask_to_apicid(tmp
);
1858 msg
->address_hi
= MSI_ADDR_BASE_HI
;
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
);
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
);
1881 static void set_msi_irq_affinity(unsigned int irq
, cpumask_t mask
)
1888 cpus_and(tmp
, mask
, cpu_online_map
);
1889 if (cpus_empty(tmp
))
1892 cpus_and(mask
, tmp
, CPU_MASK_ALL
);
1894 vector
= assign_irq_vector(irq
, mask
, &tmp
);
1898 dest
= cpu_mask_to_apicid(tmp
);
1900 read_msi_msg(irq
, &msg
);
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
);
1907 write_msi_msg(irq
, &msg
);
1908 set_native_irq_info(irq
, mask
);
1910 #endif /* CONFIG_SMP */
1913 * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
1914 * which implement the MSI or MSI-X Capability Structure.
1916 static struct irq_chip msi_chip
= {
1918 .unmask
= unmask_msi_irq
,
1919 .mask
= mask_msi_irq
,
1920 .ack
= ack_apic_edge
,
1922 .set_affinity
= set_msi_irq_affinity
,
1924 .retrigger
= ioapic_retrigger_irq
,
1927 int arch_setup_msi_irq(unsigned int irq
, struct pci_dev
*dev
)
1931 ret
= msi_compose_msg(dev
, irq
, &msg
);
1935 write_msi_msg(irq
, &msg
);
1937 set_irq_chip_and_handler_name(irq
, &msi_chip
, handle_edge_irq
, "edge");
1942 void arch_teardown_msi_irq(unsigned int irq
)
1947 #endif /* CONFIG_PCI_MSI */
1950 * Hypertransport interrupt support
1952 #ifdef CONFIG_HT_IRQ
1956 static void target_ht_irq(unsigned int irq
, unsigned int dest
, u8 vector
)
1958 struct ht_irq_msg msg
;
1959 fetch_ht_irq_msg(irq
, &msg
);
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
);
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
);
1967 write_ht_irq_msg(irq
, &msg
);
1970 static void set_ht_irq_affinity(unsigned int irq
, cpumask_t mask
)
1976 cpus_and(tmp
, mask
, cpu_online_map
);
1977 if (cpus_empty(tmp
))
1980 cpus_and(mask
, tmp
, CPU_MASK_ALL
);
1982 vector
= assign_irq_vector(irq
, mask
, &tmp
);
1986 dest
= cpu_mask_to_apicid(tmp
);
1988 target_ht_irq(irq
, dest
, vector
);
1989 set_native_irq_info(irq
, mask
);
1993 static struct irq_chip ht_irq_chip
= {
1995 .mask
= mask_ht_irq
,
1996 .unmask
= unmask_ht_irq
,
1997 .ack
= ack_apic_edge
,
1999 .set_affinity
= set_ht_irq_affinity
,
2001 .retrigger
= ioapic_retrigger_irq
,
2004 int arch_setup_ht_irq(unsigned int irq
, struct pci_dev
*dev
)
2009 vector
= assign_irq_vector(irq
, TARGET_CPUS
, &tmp
);
2011 struct ht_irq_msg msg
;
2014 dest
= cpu_mask_to_apicid(tmp
);
2016 msg
.address_hi
= HT_IRQ_HIGH_DEST_ID(dest
);
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
;
2031 write_ht_irq_msg(irq
, &msg
);
2033 set_irq_chip_and_handler_name(irq
, &ht_irq_chip
,
2034 handle_edge_irq
, "edge");
2038 #endif /* CONFIG_HT_IRQ */
2040 /* --------------------------------------------------------------------------
2041 ACPI-based IOAPIC Configuration
2042 -------------------------------------------------------------------------- */
2046 #define IO_APIC_MAX_ID 0xFE
2048 int __init
io_apic_get_redir_entries (int ioapic
)
2050 union IO_APIC_reg_01 reg_01
;
2051 unsigned long flags
;
2053 spin_lock_irqsave(&ioapic_lock
, flags
);
2054 reg_01
.raw
= io_apic_read(ioapic
, 1);
2055 spin_unlock_irqrestore(&ioapic_lock
, flags
);
2057 return reg_01
.bits
.entries
;
2061 int io_apic_set_pci_routing (int ioapic
, int pin
, int irq
, int triggering
, int polarity
)
2063 struct IO_APIC_route_entry entry
;
2064 unsigned long flags
;
2068 if (!IO_APIC_IRQ(irq
)) {
2069 apic_printk(APIC_QUIET
,KERN_ERR
"IOAPIC[%d]: Invalid reference to IRQ 0\n",
2075 * IRQs < 16 are already in the irq_2_pin[] map
2078 add_pin_to_irq(irq
, ioapic
, pin
);
2081 vector
= assign_irq_vector(irq
, TARGET_CPUS
, &mask
);
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.
2091 memset(&entry
,0,sizeof(entry
));
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;
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
);
2106 ioapic_register_intr(irq
, entry
.vector
, triggering
);
2108 if (!ioapic
&& (irq
< 16))
2109 disable_8259A_irq(irq
);
2111 ioapic_write_entry(ioapic
, pin
, entry
);
2113 spin_lock_irqsave(&ioapic_lock
, flags
);
2114 set_native_irq_info(irq
, TARGET_CPUS
);
2115 spin_unlock_irqrestore(&ioapic_lock
, flags
);
2120 #endif /* CONFIG_ACPI */
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
2129 void __init
setup_ioapic_dest(void)
2131 int pin
, ioapic
, irq
, irq_entry
;
2133 if (skip_ioapic_setup
== 1)
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)
2141 irq
= pin_2_irq(irq_entry
, ioapic
, pin
);
2142 set_ioapic_affinity_irq(irq
, TARGET_CPUS
);