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
= 0;
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 static 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
];
91 #define __DO_ACTION(R, ACTION, FINAL) \
95 struct irq_pin_list *entry = irq_2_pin + irq; \
97 BUG_ON(irq >= NR_IRQS); \
103 reg = io_apic_read(entry->apic, 0x10 + R + pin*2); \
105 io_apic_modify(entry->apic, reg); \
108 entry = irq_2_pin + entry->next; \
114 struct { u32 w1
, w2
; };
115 struct IO_APIC_route_entry entry
;
118 static struct IO_APIC_route_entry
ioapic_read_entry(int apic
, int pin
)
120 union entry_union eu
;
122 spin_lock_irqsave(&ioapic_lock
, flags
);
123 eu
.w1
= io_apic_read(apic
, 0x10 + 2 * pin
);
124 eu
.w2
= io_apic_read(apic
, 0x11 + 2 * pin
);
125 spin_unlock_irqrestore(&ioapic_lock
, flags
);
129 static void ioapic_write_entry(int apic
, int pin
, struct IO_APIC_route_entry e
)
132 union entry_union eu
;
134 spin_lock_irqsave(&ioapic_lock
, flags
);
135 io_apic_write(apic
, 0x10 + 2*pin
, eu
.w1
);
136 io_apic_write(apic
, 0x11 + 2*pin
, eu
.w2
);
137 spin_unlock_irqrestore(&ioapic_lock
, flags
);
141 static void __target_IO_APIC_irq(unsigned int irq
, unsigned int dest
, u8 vector
)
144 struct irq_pin_list
*entry
= irq_2_pin
+ irq
;
146 BUG_ON(irq
>= NR_IRQS
);
153 io_apic_write(apic
, 0x11 + pin
*2, dest
);
154 reg
= io_apic_read(apic
, 0x10 + pin
*2);
157 io_apic_modify(apic
, reg
);
160 entry
= irq_2_pin
+ entry
->next
;
164 static void set_ioapic_affinity_irq(unsigned int irq
, cpumask_t mask
)
171 cpus_and(tmp
, mask
, cpu_online_map
);
175 cpus_and(mask
, tmp
, CPU_MASK_ALL
);
177 vector
= assign_irq_vector(irq
, mask
, &tmp
);
181 dest
= cpu_mask_to_apicid(tmp
);
184 * Only the high 8 bits are valid.
186 dest
= SET_APIC_LOGICAL_ID(dest
);
188 spin_lock_irqsave(&ioapic_lock
, flags
);
189 __target_IO_APIC_irq(irq
, dest
, vector
);
190 set_native_irq_info(irq
, mask
);
191 spin_unlock_irqrestore(&ioapic_lock
, flags
);
196 * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
197 * shared ISA-space IRQs, so we have to support them. We are super
198 * fast in the common case, and fast for shared ISA-space IRQs.
200 static void add_pin_to_irq(unsigned int irq
, int apic
, int pin
)
202 static int first_free_entry
= NR_IRQS
;
203 struct irq_pin_list
*entry
= irq_2_pin
+ irq
;
205 BUG_ON(irq
>= NR_IRQS
);
207 entry
= irq_2_pin
+ entry
->next
;
209 if (entry
->pin
!= -1) {
210 entry
->next
= first_free_entry
;
211 entry
= irq_2_pin
+ entry
->next
;
212 if (++first_free_entry
>= PIN_MAP_SIZE
)
213 panic("io_apic.c: ran out of irq_2_pin entries!");
220 #define DO_ACTION(name,R,ACTION, FINAL) \
222 static void name##_IO_APIC_irq (unsigned int irq) \
223 __DO_ACTION(R, ACTION, FINAL)
225 DO_ACTION( __mask
, 0, |= 0x00010000, io_apic_sync(entry
->apic
) )
227 DO_ACTION( __unmask
, 0, &= 0xfffeffff, )
230 static void mask_IO_APIC_irq (unsigned int irq
)
234 spin_lock_irqsave(&ioapic_lock
, flags
);
235 __mask_IO_APIC_irq(irq
);
236 spin_unlock_irqrestore(&ioapic_lock
, flags
);
239 static void unmask_IO_APIC_irq (unsigned int irq
)
243 spin_lock_irqsave(&ioapic_lock
, flags
);
244 __unmask_IO_APIC_irq(irq
);
245 spin_unlock_irqrestore(&ioapic_lock
, flags
);
248 static void clear_IO_APIC_pin(unsigned int apic
, unsigned int pin
)
250 struct IO_APIC_route_entry entry
;
252 /* Check delivery_mode to be sure we're not clearing an SMI pin */
253 entry
= ioapic_read_entry(apic
, pin
);
254 if (entry
.delivery_mode
== dest_SMI
)
257 * Disable it in the IO-APIC irq-routing table:
259 memset(&entry
, 0, sizeof(entry
));
261 ioapic_write_entry(apic
, pin
, entry
);
264 static void clear_IO_APIC (void)
268 for (apic
= 0; apic
< nr_ioapics
; apic
++)
269 for (pin
= 0; pin
< nr_ioapic_registers
[apic
]; pin
++)
270 clear_IO_APIC_pin(apic
, pin
);
273 int skip_ioapic_setup
;
276 /* dummy parsing: see setup.c */
278 static int __init
disable_ioapic_setup(char *str
)
280 skip_ioapic_setup
= 1;
283 early_param("noapic", disable_ioapic_setup
);
285 /* Actually the next is obsolete, but keep it for paranoid reasons -AK */
286 static int __init
disable_timer_pin_setup(char *arg
)
288 disable_timer_pin_1
= 1;
291 __setup("disable_timer_pin_1", disable_timer_pin_setup
);
293 static int __init
setup_disable_8254_timer(char *s
)
295 timer_over_8254
= -1;
298 static int __init
setup_enable_8254_timer(char *s
)
304 __setup("disable_8254_timer", setup_disable_8254_timer
);
305 __setup("enable_8254_timer", setup_enable_8254_timer
);
309 * Find the IRQ entry number of a certain pin.
311 static int find_irq_entry(int apic
, int pin
, int type
)
315 for (i
= 0; i
< mp_irq_entries
; i
++)
316 if (mp_irqs
[i
].mpc_irqtype
== type
&&
317 (mp_irqs
[i
].mpc_dstapic
== mp_ioapics
[apic
].mpc_apicid
||
318 mp_irqs
[i
].mpc_dstapic
== MP_APIC_ALL
) &&
319 mp_irqs
[i
].mpc_dstirq
== pin
)
326 * Find the pin to which IRQ[irq] (ISA) is connected
328 static int __init
find_isa_irq_pin(int irq
, int type
)
332 for (i
= 0; i
< mp_irq_entries
; i
++) {
333 int lbus
= mp_irqs
[i
].mpc_srcbus
;
335 if (test_bit(lbus
, mp_bus_not_pci
) &&
336 (mp_irqs
[i
].mpc_irqtype
== type
) &&
337 (mp_irqs
[i
].mpc_srcbusirq
== irq
))
339 return mp_irqs
[i
].mpc_dstirq
;
344 static int __init
find_isa_irq_apic(int irq
, int type
)
348 for (i
= 0; i
< mp_irq_entries
; i
++) {
349 int lbus
= mp_irqs
[i
].mpc_srcbus
;
351 if (test_bit(lbus
, mp_bus_not_pci
) &&
352 (mp_irqs
[i
].mpc_irqtype
== type
) &&
353 (mp_irqs
[i
].mpc_srcbusirq
== irq
))
356 if (i
< mp_irq_entries
) {
358 for(apic
= 0; apic
< nr_ioapics
; apic
++) {
359 if (mp_ioapics
[apic
].mpc_apicid
== mp_irqs
[i
].mpc_dstapic
)
368 * Find a specific PCI IRQ entry.
369 * Not an __init, possibly needed by modules
371 static int pin_2_irq(int idx
, int apic
, int pin
);
373 int IO_APIC_get_PCI_irq_vector(int bus
, int slot
, int pin
)
375 int apic
, i
, best_guess
= -1;
377 apic_printk(APIC_DEBUG
, "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
379 if (mp_bus_id_to_pci_bus
[bus
] == -1) {
380 apic_printk(APIC_VERBOSE
, "PCI BIOS passed nonexistent PCI bus %d!\n", bus
);
383 for (i
= 0; i
< mp_irq_entries
; i
++) {
384 int lbus
= mp_irqs
[i
].mpc_srcbus
;
386 for (apic
= 0; apic
< nr_ioapics
; apic
++)
387 if (mp_ioapics
[apic
].mpc_apicid
== mp_irqs
[i
].mpc_dstapic
||
388 mp_irqs
[i
].mpc_dstapic
== MP_APIC_ALL
)
391 if (!test_bit(lbus
, mp_bus_not_pci
) &&
392 !mp_irqs
[i
].mpc_irqtype
&&
394 (slot
== ((mp_irqs
[i
].mpc_srcbusirq
>> 2) & 0x1f))) {
395 int irq
= pin_2_irq(i
,apic
,mp_irqs
[i
].mpc_dstirq
);
397 if (!(apic
|| IO_APIC_IRQ(irq
)))
400 if (pin
== (mp_irqs
[i
].mpc_srcbusirq
& 3))
403 * Use the first all-but-pin matching entry as a
404 * best-guess fuzzy result for broken mptables.
410 BUG_ON(best_guess
>= NR_IRQS
);
414 /* ISA interrupts are always polarity zero edge triggered,
415 * when listed as conforming in the MP table. */
417 #define default_ISA_trigger(idx) (0)
418 #define default_ISA_polarity(idx) (0)
420 /* PCI interrupts are always polarity one level triggered,
421 * when listed as conforming in the MP table. */
423 #define default_PCI_trigger(idx) (1)
424 #define default_PCI_polarity(idx) (1)
426 static int __init
MPBIOS_polarity(int idx
)
428 int bus
= mp_irqs
[idx
].mpc_srcbus
;
432 * Determine IRQ line polarity (high active or low active):
434 switch (mp_irqs
[idx
].mpc_irqflag
& 3)
436 case 0: /* conforms, ie. bus-type dependent polarity */
437 if (test_bit(bus
, mp_bus_not_pci
))
438 polarity
= default_ISA_polarity(idx
);
440 polarity
= default_PCI_polarity(idx
);
442 case 1: /* high active */
447 case 2: /* reserved */
449 printk(KERN_WARNING
"broken BIOS!!\n");
453 case 3: /* low active */
458 default: /* invalid */
460 printk(KERN_WARNING
"broken BIOS!!\n");
468 static int MPBIOS_trigger(int idx
)
470 int bus
= mp_irqs
[idx
].mpc_srcbus
;
474 * Determine IRQ trigger mode (edge or level sensitive):
476 switch ((mp_irqs
[idx
].mpc_irqflag
>>2) & 3)
478 case 0: /* conforms, ie. bus-type dependent */
479 if (test_bit(bus
, mp_bus_not_pci
))
480 trigger
= default_ISA_trigger(idx
);
482 trigger
= default_PCI_trigger(idx
);
489 case 2: /* reserved */
491 printk(KERN_WARNING
"broken BIOS!!\n");
500 default: /* invalid */
502 printk(KERN_WARNING
"broken BIOS!!\n");
510 static inline int irq_polarity(int idx
)
512 return MPBIOS_polarity(idx
);
515 static inline int irq_trigger(int idx
)
517 return MPBIOS_trigger(idx
);
520 static int pin_2_irq(int idx
, int apic
, int pin
)
523 int bus
= mp_irqs
[idx
].mpc_srcbus
;
526 * Debugging check, we are in big trouble if this message pops up!
528 if (mp_irqs
[idx
].mpc_dstirq
!= pin
)
529 printk(KERN_ERR
"broken BIOS or MPTABLE parser, ayiee!!\n");
531 if (test_bit(bus
, mp_bus_not_pci
)) {
532 irq
= mp_irqs
[idx
].mpc_srcbusirq
;
535 * PCI IRQs are mapped in order
539 irq
+= nr_ioapic_registers
[i
++];
542 BUG_ON(irq
>= NR_IRQS
);
546 static inline int IO_APIC_irq_trigger(int irq
)
550 for (apic
= 0; apic
< nr_ioapics
; apic
++) {
551 for (pin
= 0; pin
< nr_ioapic_registers
[apic
]; pin
++) {
552 idx
= find_irq_entry(apic
,pin
,mp_INT
);
553 if ((idx
!= -1) && (irq
== pin_2_irq(idx
,apic
,pin
)))
554 return irq_trigger(idx
);
558 * nonexistent IRQs are edge default
563 /* irq_vectors is indexed by the sum of all RTEs in all I/O APICs. */
564 static u8 irq_vector
[NR_IRQ_VECTORS
] __read_mostly
= {
565 [0] = FIRST_EXTERNAL_VECTOR
+ 0,
566 [1] = FIRST_EXTERNAL_VECTOR
+ 1,
567 [2] = FIRST_EXTERNAL_VECTOR
+ 2,
568 [3] = FIRST_EXTERNAL_VECTOR
+ 3,
569 [4] = FIRST_EXTERNAL_VECTOR
+ 4,
570 [5] = FIRST_EXTERNAL_VECTOR
+ 5,
571 [6] = FIRST_EXTERNAL_VECTOR
+ 6,
572 [7] = FIRST_EXTERNAL_VECTOR
+ 7,
573 [8] = FIRST_EXTERNAL_VECTOR
+ 8,
574 [9] = FIRST_EXTERNAL_VECTOR
+ 9,
575 [10] = FIRST_EXTERNAL_VECTOR
+ 10,
576 [11] = FIRST_EXTERNAL_VECTOR
+ 11,
577 [12] = FIRST_EXTERNAL_VECTOR
+ 12,
578 [13] = FIRST_EXTERNAL_VECTOR
+ 13,
579 [14] = FIRST_EXTERNAL_VECTOR
+ 14,
580 [15] = FIRST_EXTERNAL_VECTOR
+ 15,
583 static cpumask_t irq_domain
[NR_IRQ_VECTORS
] __read_mostly
= {
602 static int __assign_irq_vector(int irq
, cpumask_t mask
, cpumask_t
*result
)
605 * NOTE! The local APIC isn't very good at handling
606 * multiple interrupts at the same interrupt level.
607 * As the interrupt level is determined by taking the
608 * vector number and shifting that right by 4, we
609 * want to spread these out a bit so that they don't
610 * all fall in the same interrupt level.
612 * Also, we've got to be careful not to trash gate
613 * 0x80, because int 0x80 is hm, kind of importantish. ;)
618 } pos
[NR_CPUS
] = { [ 0 ... NR_CPUS
- 1] = {FIRST_DEVICE_VECTOR
, 0} };
622 BUG_ON((unsigned)irq
>= NR_IRQ_VECTORS
);
624 if (irq_vector
[irq
] > 0)
625 old_vector
= irq_vector
[irq
];
626 if (old_vector
> 0) {
627 cpus_and(*result
, irq_domain
[irq
], mask
);
628 if (!cpus_empty(*result
))
632 for_each_cpu_mask(cpu
, mask
) {
637 domain
= vector_allocation_domain(cpu
);
638 first
= first_cpu(domain
);
640 vector
= pos
[first
].vector
;
641 offset
= pos
[first
].offset
;
644 if (vector
>= FIRST_SYSTEM_VECTOR
) {
645 /* If we run out of vectors on large boxen, must share them. */
646 offset
= (offset
+ 1) % 8;
647 vector
= FIRST_DEVICE_VECTOR
+ offset
;
649 if (unlikely(pos
[first
].vector
== vector
))
651 if (vector
== IA32_SYSCALL_VECTOR
)
653 for_each_cpu_mask(new_cpu
, domain
)
654 if (per_cpu(vector_irq
, cpu
)[vector
] != -1)
657 for_each_cpu_mask(new_cpu
, domain
) {
658 pos
[cpu
].vector
= vector
;
659 pos
[cpu
].offset
= offset
;
661 if (old_vector
>= 0) {
663 for_each_cpu_mask(old_cpu
, domain
)
664 per_cpu(vector_irq
, old_cpu
)[old_vector
] = -1;
666 for_each_cpu_mask(new_cpu
, domain
)
667 per_cpu(vector_irq
, new_cpu
)[vector
] = irq
;
668 irq_vector
[irq
] = vector
;
669 irq_domain
[irq
] = domain
;
670 cpus_and(*result
, domain
, mask
);
676 static int assign_irq_vector(int irq
, cpumask_t mask
, cpumask_t
*result
)
681 spin_lock_irqsave(&vector_lock
, flags
);
682 vector
= __assign_irq_vector(irq
, mask
, result
);
683 spin_unlock_irqrestore(&vector_lock
, flags
);
687 extern void (*interrupt
[NR_IRQS
])(void);
689 static struct irq_chip ioapic_chip
;
691 #define IOAPIC_AUTO -1
692 #define IOAPIC_EDGE 0
693 #define IOAPIC_LEVEL 1
695 static void ioapic_register_intr(int irq
, int vector
, unsigned long trigger
)
697 if ((trigger
== IOAPIC_AUTO
&& IO_APIC_irq_trigger(irq
)) ||
698 trigger
== IOAPIC_LEVEL
)
699 set_irq_chip_and_handler(irq
, &ioapic_chip
,
702 set_irq_chip_and_handler(irq
, &ioapic_chip
,
706 static void __init
setup_IO_APIC_irqs(void)
708 struct IO_APIC_route_entry entry
;
709 int apic
, pin
, idx
, irq
, first_notcon
= 1, vector
;
712 apic_printk(APIC_VERBOSE
, KERN_DEBUG
"init IO_APIC IRQs\n");
714 for (apic
= 0; apic
< nr_ioapics
; apic
++) {
715 for (pin
= 0; pin
< nr_ioapic_registers
[apic
]; pin
++) {
718 * add it to the IO-APIC irq-routing table:
720 memset(&entry
,0,sizeof(entry
));
722 entry
.delivery_mode
= INT_DELIVERY_MODE
;
723 entry
.dest_mode
= INT_DEST_MODE
;
724 entry
.mask
= 0; /* enable IRQ */
725 entry
.dest
.logical
.logical_dest
= cpu_mask_to_apicid(TARGET_CPUS
);
727 idx
= find_irq_entry(apic
,pin
,mp_INT
);
730 apic_printk(APIC_VERBOSE
, KERN_DEBUG
" IO-APIC (apicid-pin) %d-%d", mp_ioapics
[apic
].mpc_apicid
, pin
);
733 apic_printk(APIC_VERBOSE
, ", %d-%d", mp_ioapics
[apic
].mpc_apicid
, pin
);
737 entry
.trigger
= irq_trigger(idx
);
738 entry
.polarity
= irq_polarity(idx
);
740 if (irq_trigger(idx
)) {
743 entry
.dest
.logical
.logical_dest
= cpu_mask_to_apicid(TARGET_CPUS
);
746 irq
= pin_2_irq(idx
, apic
, pin
);
747 add_pin_to_irq(irq
, apic
, pin
);
749 if (!apic
&& !IO_APIC_IRQ(irq
))
752 if (IO_APIC_IRQ(irq
)) {
754 vector
= assign_irq_vector(irq
, TARGET_CPUS
, &mask
);
758 entry
.dest
.logical
.logical_dest
= cpu_mask_to_apicid(mask
);
759 entry
.vector
= vector
;
761 ioapic_register_intr(irq
, vector
, IOAPIC_AUTO
);
762 if (!apic
&& (irq
< 16))
763 disable_8259A_irq(irq
);
765 ioapic_write_entry(apic
, pin
, entry
);
767 spin_lock_irqsave(&ioapic_lock
, flags
);
768 set_native_irq_info(irq
, TARGET_CPUS
);
769 spin_unlock_irqrestore(&ioapic_lock
, flags
);
774 apic_printk(APIC_VERBOSE
," not connected.\n");
778 * Set up the 8259A-master output pin as broadcast to all
781 static void __init
setup_ExtINT_IRQ0_pin(unsigned int apic
, unsigned int pin
, int vector
)
783 struct IO_APIC_route_entry entry
;
786 memset(&entry
,0,sizeof(entry
));
788 disable_8259A_irq(0);
791 apic_write(APIC_LVT0
, APIC_LVT_MASKED
| APIC_DM_EXTINT
);
794 * We use logical delivery to get the timer IRQ
797 entry
.dest_mode
= INT_DEST_MODE
;
798 entry
.mask
= 0; /* unmask IRQ now */
799 entry
.dest
.logical
.logical_dest
= cpu_mask_to_apicid(TARGET_CPUS
);
800 entry
.delivery_mode
= INT_DELIVERY_MODE
;
803 entry
.vector
= vector
;
806 * The timer IRQ doesn't have to know that behind the
807 * scene we have a 8259A-master in AEOI mode ...
809 set_irq_chip_and_handler(0, &ioapic_chip
, handle_edge_irq
);
812 * Add it to the IO-APIC irq-routing table:
814 spin_lock_irqsave(&ioapic_lock
, flags
);
815 io_apic_write(apic
, 0x11+2*pin
, *(((int *)&entry
)+1));
816 io_apic_write(apic
, 0x10+2*pin
, *(((int *)&entry
)+0));
817 spin_unlock_irqrestore(&ioapic_lock
, flags
);
822 void __init
UNEXPECTED_IO_APIC(void)
826 void __apicdebuginit
print_IO_APIC(void)
829 union IO_APIC_reg_00 reg_00
;
830 union IO_APIC_reg_01 reg_01
;
831 union IO_APIC_reg_02 reg_02
;
834 if (apic_verbosity
== APIC_QUIET
)
837 printk(KERN_DEBUG
"number of MP IRQ sources: %d.\n", mp_irq_entries
);
838 for (i
= 0; i
< nr_ioapics
; i
++)
839 printk(KERN_DEBUG
"number of IO-APIC #%d registers: %d.\n",
840 mp_ioapics
[i
].mpc_apicid
, nr_ioapic_registers
[i
]);
843 * We are a bit conservative about what we expect. We have to
844 * know about every hardware change ASAP.
846 printk(KERN_INFO
"testing the IO APIC.......................\n");
848 for (apic
= 0; apic
< nr_ioapics
; apic
++) {
850 spin_lock_irqsave(&ioapic_lock
, flags
);
851 reg_00
.raw
= io_apic_read(apic
, 0);
852 reg_01
.raw
= io_apic_read(apic
, 1);
853 if (reg_01
.bits
.version
>= 0x10)
854 reg_02
.raw
= io_apic_read(apic
, 2);
855 spin_unlock_irqrestore(&ioapic_lock
, flags
);
858 printk(KERN_DEBUG
"IO APIC #%d......\n", mp_ioapics
[apic
].mpc_apicid
);
859 printk(KERN_DEBUG
".... register #00: %08X\n", reg_00
.raw
);
860 printk(KERN_DEBUG
"....... : physical APIC id: %02X\n", reg_00
.bits
.ID
);
861 if (reg_00
.bits
.__reserved_1
|| reg_00
.bits
.__reserved_2
)
862 UNEXPECTED_IO_APIC();
864 printk(KERN_DEBUG
".... register #01: %08X\n", *(int *)®_01
);
865 printk(KERN_DEBUG
"....... : max redirection entries: %04X\n", reg_01
.bits
.entries
);
866 if ( (reg_01
.bits
.entries
!= 0x0f) && /* older (Neptune) boards */
867 (reg_01
.bits
.entries
!= 0x17) && /* typical ISA+PCI boards */
868 (reg_01
.bits
.entries
!= 0x1b) && /* Compaq Proliant boards */
869 (reg_01
.bits
.entries
!= 0x1f) && /* dual Xeon boards */
870 (reg_01
.bits
.entries
!= 0x22) && /* bigger Xeon boards */
871 (reg_01
.bits
.entries
!= 0x2E) &&
872 (reg_01
.bits
.entries
!= 0x3F) &&
873 (reg_01
.bits
.entries
!= 0x03)
875 UNEXPECTED_IO_APIC();
877 printk(KERN_DEBUG
"....... : PRQ implemented: %X\n", reg_01
.bits
.PRQ
);
878 printk(KERN_DEBUG
"....... : IO APIC version: %04X\n", reg_01
.bits
.version
);
879 if ( (reg_01
.bits
.version
!= 0x01) && /* 82489DX IO-APICs */
880 (reg_01
.bits
.version
!= 0x02) && /* 82801BA IO-APICs (ICH2) */
881 (reg_01
.bits
.version
!= 0x10) && /* oldest IO-APICs */
882 (reg_01
.bits
.version
!= 0x11) && /* Pentium/Pro IO-APICs */
883 (reg_01
.bits
.version
!= 0x13) && /* Xeon IO-APICs */
884 (reg_01
.bits
.version
!= 0x20) /* Intel P64H (82806 AA) */
886 UNEXPECTED_IO_APIC();
887 if (reg_01
.bits
.__reserved_1
|| reg_01
.bits
.__reserved_2
)
888 UNEXPECTED_IO_APIC();
890 if (reg_01
.bits
.version
>= 0x10) {
891 printk(KERN_DEBUG
".... register #02: %08X\n", reg_02
.raw
);
892 printk(KERN_DEBUG
"....... : arbitration: %02X\n", reg_02
.bits
.arbitration
);
893 if (reg_02
.bits
.__reserved_1
|| reg_02
.bits
.__reserved_2
)
894 UNEXPECTED_IO_APIC();
897 printk(KERN_DEBUG
".... IRQ redirection table:\n");
899 printk(KERN_DEBUG
" NR Log Phy Mask Trig IRR Pol"
900 " Stat Dest Deli Vect: \n");
902 for (i
= 0; i
<= reg_01
.bits
.entries
; i
++) {
903 struct IO_APIC_route_entry entry
;
905 entry
= ioapic_read_entry(apic
, i
);
907 printk(KERN_DEBUG
" %02x %03X %02X ",
909 entry
.dest
.logical
.logical_dest
,
910 entry
.dest
.physical
.physical_dest
913 printk("%1d %1d %1d %1d %1d %1d %1d %02X\n",
918 entry
.delivery_status
,
925 printk(KERN_DEBUG
"IRQ to pin mappings:\n");
926 for (i
= 0; i
< NR_IRQS
; i
++) {
927 struct irq_pin_list
*entry
= irq_2_pin
+ i
;
930 printk(KERN_DEBUG
"IRQ%d ", i
);
932 printk("-> %d:%d", entry
->apic
, entry
->pin
);
935 entry
= irq_2_pin
+ entry
->next
;
940 printk(KERN_INFO
".................................... done.\n");
947 static __apicdebuginit
void print_APIC_bitfield (int base
)
952 if (apic_verbosity
== APIC_QUIET
)
955 printk(KERN_DEBUG
"0123456789abcdef0123456789abcdef\n" KERN_DEBUG
);
956 for (i
= 0; i
< 8; i
++) {
957 v
= apic_read(base
+ i
*0x10);
958 for (j
= 0; j
< 32; j
++) {
968 void __apicdebuginit
print_local_APIC(void * dummy
)
970 unsigned int v
, ver
, maxlvt
;
972 if (apic_verbosity
== APIC_QUIET
)
975 printk("\n" KERN_DEBUG
"printing local APIC contents on CPU#%d/%d:\n",
976 smp_processor_id(), hard_smp_processor_id());
977 v
= apic_read(APIC_ID
);
978 printk(KERN_INFO
"... APIC ID: %08x (%01x)\n", v
, GET_APIC_ID(v
));
979 v
= apic_read(APIC_LVR
);
980 printk(KERN_INFO
"... APIC VERSION: %08x\n", v
);
981 ver
= GET_APIC_VERSION(v
);
982 maxlvt
= get_maxlvt();
984 v
= apic_read(APIC_TASKPRI
);
985 printk(KERN_DEBUG
"... APIC TASKPRI: %08x (%02x)\n", v
, v
& APIC_TPRI_MASK
);
987 v
= apic_read(APIC_ARBPRI
);
988 printk(KERN_DEBUG
"... APIC ARBPRI: %08x (%02x)\n", v
,
989 v
& APIC_ARBPRI_MASK
);
990 v
= apic_read(APIC_PROCPRI
);
991 printk(KERN_DEBUG
"... APIC PROCPRI: %08x\n", v
);
993 v
= apic_read(APIC_EOI
);
994 printk(KERN_DEBUG
"... APIC EOI: %08x\n", v
);
995 v
= apic_read(APIC_RRR
);
996 printk(KERN_DEBUG
"... APIC RRR: %08x\n", v
);
997 v
= apic_read(APIC_LDR
);
998 printk(KERN_DEBUG
"... APIC LDR: %08x\n", v
);
999 v
= apic_read(APIC_DFR
);
1000 printk(KERN_DEBUG
"... APIC DFR: %08x\n", v
);
1001 v
= apic_read(APIC_SPIV
);
1002 printk(KERN_DEBUG
"... APIC SPIV: %08x\n", v
);
1004 printk(KERN_DEBUG
"... APIC ISR field:\n");
1005 print_APIC_bitfield(APIC_ISR
);
1006 printk(KERN_DEBUG
"... APIC TMR field:\n");
1007 print_APIC_bitfield(APIC_TMR
);
1008 printk(KERN_DEBUG
"... APIC IRR field:\n");
1009 print_APIC_bitfield(APIC_IRR
);
1011 v
= apic_read(APIC_ESR
);
1012 printk(KERN_DEBUG
"... APIC ESR: %08x\n", v
);
1014 v
= apic_read(APIC_ICR
);
1015 printk(KERN_DEBUG
"... APIC ICR: %08x\n", v
);
1016 v
= apic_read(APIC_ICR2
);
1017 printk(KERN_DEBUG
"... APIC ICR2: %08x\n", v
);
1019 v
= apic_read(APIC_LVTT
);
1020 printk(KERN_DEBUG
"... APIC LVTT: %08x\n", v
);
1022 if (maxlvt
> 3) { /* PC is LVT#4. */
1023 v
= apic_read(APIC_LVTPC
);
1024 printk(KERN_DEBUG
"... APIC LVTPC: %08x\n", v
);
1026 v
= apic_read(APIC_LVT0
);
1027 printk(KERN_DEBUG
"... APIC LVT0: %08x\n", v
);
1028 v
= apic_read(APIC_LVT1
);
1029 printk(KERN_DEBUG
"... APIC LVT1: %08x\n", v
);
1031 if (maxlvt
> 2) { /* ERR is LVT#3. */
1032 v
= apic_read(APIC_LVTERR
);
1033 printk(KERN_DEBUG
"... APIC LVTERR: %08x\n", v
);
1036 v
= apic_read(APIC_TMICT
);
1037 printk(KERN_DEBUG
"... APIC TMICT: %08x\n", v
);
1038 v
= apic_read(APIC_TMCCT
);
1039 printk(KERN_DEBUG
"... APIC TMCCT: %08x\n", v
);
1040 v
= apic_read(APIC_TDCR
);
1041 printk(KERN_DEBUG
"... APIC TDCR: %08x\n", v
);
1045 void print_all_local_APICs (void)
1047 on_each_cpu(print_local_APIC
, NULL
, 1, 1);
1050 void __apicdebuginit
print_PIC(void)
1053 unsigned long flags
;
1055 if (apic_verbosity
== APIC_QUIET
)
1058 printk(KERN_DEBUG
"\nprinting PIC contents\n");
1060 spin_lock_irqsave(&i8259A_lock
, flags
);
1062 v
= inb(0xa1) << 8 | inb(0x21);
1063 printk(KERN_DEBUG
"... PIC IMR: %04x\n", v
);
1065 v
= inb(0xa0) << 8 | inb(0x20);
1066 printk(KERN_DEBUG
"... PIC IRR: %04x\n", v
);
1070 v
= inb(0xa0) << 8 | inb(0x20);
1074 spin_unlock_irqrestore(&i8259A_lock
, flags
);
1076 printk(KERN_DEBUG
"... PIC ISR: %04x\n", v
);
1078 v
= inb(0x4d1) << 8 | inb(0x4d0);
1079 printk(KERN_DEBUG
"... PIC ELCR: %04x\n", v
);
1084 static void __init
enable_IO_APIC(void)
1086 union IO_APIC_reg_01 reg_01
;
1087 int i8259_apic
, i8259_pin
;
1089 unsigned long flags
;
1091 for (i
= 0; i
< PIN_MAP_SIZE
; i
++) {
1092 irq_2_pin
[i
].pin
= -1;
1093 irq_2_pin
[i
].next
= 0;
1097 * The number of IO-APIC IRQ registers (== #pins):
1099 for (apic
= 0; apic
< nr_ioapics
; apic
++) {
1100 spin_lock_irqsave(&ioapic_lock
, flags
);
1101 reg_01
.raw
= io_apic_read(apic
, 1);
1102 spin_unlock_irqrestore(&ioapic_lock
, flags
);
1103 nr_ioapic_registers
[apic
] = reg_01
.bits
.entries
+1;
1105 for(apic
= 0; apic
< nr_ioapics
; apic
++) {
1107 /* See if any of the pins is in ExtINT mode */
1108 for (pin
= 0; pin
< nr_ioapic_registers
[apic
]; pin
++) {
1109 struct IO_APIC_route_entry entry
;
1110 entry
= ioapic_read_entry(apic
, pin
);
1112 /* If the interrupt line is enabled and in ExtInt mode
1113 * I have found the pin where the i8259 is connected.
1115 if ((entry
.mask
== 0) && (entry
.delivery_mode
== dest_ExtINT
)) {
1116 ioapic_i8259
.apic
= apic
;
1117 ioapic_i8259
.pin
= pin
;
1123 /* Look to see what if the MP table has reported the ExtINT */
1124 i8259_pin
= find_isa_irq_pin(0, mp_ExtINT
);
1125 i8259_apic
= find_isa_irq_apic(0, mp_ExtINT
);
1126 /* Trust the MP table if nothing is setup in the hardware */
1127 if ((ioapic_i8259
.pin
== -1) && (i8259_pin
>= 0)) {
1128 printk(KERN_WARNING
"ExtINT not setup in hardware but reported by MP table\n");
1129 ioapic_i8259
.pin
= i8259_pin
;
1130 ioapic_i8259
.apic
= i8259_apic
;
1132 /* Complain if the MP table and the hardware disagree */
1133 if (((ioapic_i8259
.apic
!= i8259_apic
) || (ioapic_i8259
.pin
!= i8259_pin
)) &&
1134 (i8259_pin
>= 0) && (ioapic_i8259
.pin
>= 0))
1136 printk(KERN_WARNING
"ExtINT in hardware and MP table differ\n");
1140 * Do not trust the IO-APIC being empty at bootup
1146 * Not an __init, needed by the reboot code
1148 void disable_IO_APIC(void)
1151 * Clear the IO-APIC before rebooting:
1156 * If the i8259 is routed through an IOAPIC
1157 * Put that IOAPIC in virtual wire mode
1158 * so legacy interrupts can be delivered.
1160 if (ioapic_i8259
.pin
!= -1) {
1161 struct IO_APIC_route_entry entry
;
1163 memset(&entry
, 0, sizeof(entry
));
1164 entry
.mask
= 0; /* Enabled */
1165 entry
.trigger
= 0; /* Edge */
1167 entry
.polarity
= 0; /* High */
1168 entry
.delivery_status
= 0;
1169 entry
.dest_mode
= 0; /* Physical */
1170 entry
.delivery_mode
= dest_ExtINT
; /* ExtInt */
1172 entry
.dest
.physical
.physical_dest
=
1173 GET_APIC_ID(apic_read(APIC_ID
));
1176 * Add it to the IO-APIC irq-routing table:
1178 ioapic_write_entry(ioapic_i8259
.apic
, ioapic_i8259
.pin
, entry
);
1181 disconnect_bsp_APIC(ioapic_i8259
.pin
!= -1);
1185 * There is a nasty bug in some older SMP boards, their mptable lies
1186 * about the timer IRQ. We do the following to work around the situation:
1188 * - timer IRQ defaults to IO-APIC IRQ
1189 * - if this function detects that timer IRQs are defunct, then we fall
1190 * back to ISA timer IRQs
1192 static int __init
timer_irq_works(void)
1194 unsigned long t1
= jiffies
;
1197 /* Let ten ticks pass... */
1198 mdelay((10 * 1000) / HZ
);
1201 * Expect a few ticks at least, to be sure some possible
1202 * glue logic does not lock up after one or two first
1203 * ticks in a non-ExtINT mode. Also the local APIC
1204 * might have cached one ExtINT interrupt. Finally, at
1205 * least one tick may be lost due to delays.
1209 if (jiffies
- t1
> 4)
1215 * In the SMP+IOAPIC case it might happen that there are an unspecified
1216 * number of pending IRQ events unhandled. These cases are very rare,
1217 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
1218 * better to do it this way as thus we do not have to be aware of
1219 * 'pending' interrupts in the IRQ path, except at this point.
1222 * Edge triggered needs to resend any interrupt
1223 * that was delayed but this is now handled in the device
1228 * Starting up a edge-triggered IO-APIC interrupt is
1229 * nasty - we need to make sure that we get the edge.
1230 * If it is already asserted for some reason, we need
1231 * return 1 to indicate that is was pending.
1233 * This is not complete - we should be able to fake
1234 * an edge even if it isn't on the 8259A...
1237 static unsigned int startup_ioapic_irq(unsigned int irq
)
1239 int was_pending
= 0;
1240 unsigned long flags
;
1242 spin_lock_irqsave(&ioapic_lock
, flags
);
1244 disable_8259A_irq(irq
);
1245 if (i8259A_irq_pending(irq
))
1248 __unmask_IO_APIC_irq(irq
);
1249 spin_unlock_irqrestore(&ioapic_lock
, flags
);
1254 static int ioapic_retrigger_irq(unsigned int irq
)
1259 vector
= irq_vector
[irq
];
1261 cpu_set(vector
>> 8, mask
);
1263 send_IPI_mask(mask
, vector
& 0xff);
1269 * Level and edge triggered IO-APIC interrupts need different handling,
1270 * so we use two separate IRQ descriptors. Edge triggered IRQs can be
1271 * handled with the level-triggered descriptor, but that one has slightly
1272 * more overhead. Level-triggered interrupts cannot be handled with the
1273 * edge-triggered handler, without risking IRQ storms and other ugly
1277 static void ack_apic_edge(unsigned int irq
)
1279 move_native_irq(irq
);
1283 static void ack_apic_level(unsigned int irq
)
1285 int do_unmask_irq
= 0;
1287 #if defined(CONFIG_GENERIC_PENDING_IRQ) || defined(CONFIG_IRQBALANCE)
1288 /* If we are moving the irq we need to mask it */
1289 if (unlikely(irq_desc
[irq
].status
& IRQ_MOVE_PENDING
)) {
1291 mask_IO_APIC_irq(irq
);
1296 * We must acknowledge the irq before we move it or the acknowledge will
1297 * not propogate properly.
1301 /* Now we can move and renable the irq */
1302 move_masked_irq(irq
);
1303 if (unlikely(do_unmask_irq
))
1304 unmask_IO_APIC_irq(irq
);
1307 static struct irq_chip ioapic_chip __read_mostly
= {
1309 .startup
= startup_ioapic_irq
,
1310 .mask
= mask_IO_APIC_irq
,
1311 .unmask
= unmask_IO_APIC_irq
,
1312 .ack
= ack_apic_edge
,
1313 .eoi
= ack_apic_level
,
1315 .set_affinity
= set_ioapic_affinity_irq
,
1317 .retrigger
= ioapic_retrigger_irq
,
1320 static inline void init_IO_APIC_traps(void)
1325 * NOTE! The local APIC isn't very good at handling
1326 * multiple interrupts at the same interrupt level.
1327 * As the interrupt level is determined by taking the
1328 * vector number and shifting that right by 4, we
1329 * want to spread these out a bit so that they don't
1330 * all fall in the same interrupt level.
1332 * Also, we've got to be careful not to trash gate
1333 * 0x80, because int 0x80 is hm, kind of importantish. ;)
1335 for (irq
= 0; irq
< NR_IRQS
; irq
++) {
1337 if (IO_APIC_IRQ(tmp
) && !irq_vector
[tmp
]) {
1339 * Hmm.. We don't have an entry for this,
1340 * so default to an old-fashioned 8259
1341 * interrupt if we can..
1344 make_8259A_irq(irq
);
1346 /* Strange. Oh, well.. */
1347 irq_desc
[irq
].chip
= &no_irq_chip
;
1352 static void enable_lapic_irq (unsigned int irq
)
1356 v
= apic_read(APIC_LVT0
);
1357 apic_write(APIC_LVT0
, v
& ~APIC_LVT_MASKED
);
1360 static void disable_lapic_irq (unsigned int irq
)
1364 v
= apic_read(APIC_LVT0
);
1365 apic_write(APIC_LVT0
, v
| APIC_LVT_MASKED
);
1368 static void ack_lapic_irq (unsigned int irq
)
1373 static void end_lapic_irq (unsigned int i
) { /* nothing */ }
1375 static struct hw_interrupt_type lapic_irq_type __read_mostly
= {
1376 .typename
= "local-APIC-edge",
1377 .startup
= NULL
, /* startup_irq() not used for IRQ0 */
1378 .shutdown
= NULL
, /* shutdown_irq() not used for IRQ0 */
1379 .enable
= enable_lapic_irq
,
1380 .disable
= disable_lapic_irq
,
1381 .ack
= ack_lapic_irq
,
1382 .end
= end_lapic_irq
,
1385 static void setup_nmi (void)
1388 * Dirty trick to enable the NMI watchdog ...
1389 * We put the 8259A master into AEOI mode and
1390 * unmask on all local APICs LVT0 as NMI.
1392 * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
1393 * is from Maciej W. Rozycki - so we do not have to EOI from
1394 * the NMI handler or the timer interrupt.
1396 printk(KERN_INFO
"activating NMI Watchdog ...");
1398 enable_NMI_through_LVT0(NULL
);
1404 * This looks a bit hackish but it's about the only one way of sending
1405 * a few INTA cycles to 8259As and any associated glue logic. ICR does
1406 * not support the ExtINT mode, unfortunately. We need to send these
1407 * cycles as some i82489DX-based boards have glue logic that keeps the
1408 * 8259A interrupt line asserted until INTA. --macro
1410 static inline void unlock_ExtINT_logic(void)
1413 struct IO_APIC_route_entry entry0
, entry1
;
1414 unsigned char save_control
, save_freq_select
;
1415 unsigned long flags
;
1417 pin
= find_isa_irq_pin(8, mp_INT
);
1418 apic
= find_isa_irq_apic(8, mp_INT
);
1422 spin_lock_irqsave(&ioapic_lock
, flags
);
1423 *(((int *)&entry0
) + 1) = io_apic_read(apic
, 0x11 + 2 * pin
);
1424 *(((int *)&entry0
) + 0) = io_apic_read(apic
, 0x10 + 2 * pin
);
1425 spin_unlock_irqrestore(&ioapic_lock
, flags
);
1426 clear_IO_APIC_pin(apic
, pin
);
1428 memset(&entry1
, 0, sizeof(entry1
));
1430 entry1
.dest_mode
= 0; /* physical delivery */
1431 entry1
.mask
= 0; /* unmask IRQ now */
1432 entry1
.dest
.physical
.physical_dest
= hard_smp_processor_id();
1433 entry1
.delivery_mode
= dest_ExtINT
;
1434 entry1
.polarity
= entry0
.polarity
;
1438 spin_lock_irqsave(&ioapic_lock
, flags
);
1439 io_apic_write(apic
, 0x11 + 2 * pin
, *(((int *)&entry1
) + 1));
1440 io_apic_write(apic
, 0x10 + 2 * pin
, *(((int *)&entry1
) + 0));
1441 spin_unlock_irqrestore(&ioapic_lock
, flags
);
1443 save_control
= CMOS_READ(RTC_CONTROL
);
1444 save_freq_select
= CMOS_READ(RTC_FREQ_SELECT
);
1445 CMOS_WRITE((save_freq_select
& ~RTC_RATE_SELECT
) | 0x6,
1447 CMOS_WRITE(save_control
| RTC_PIE
, RTC_CONTROL
);
1452 if ((CMOS_READ(RTC_INTR_FLAGS
) & RTC_PF
) == RTC_PF
)
1456 CMOS_WRITE(save_control
, RTC_CONTROL
);
1457 CMOS_WRITE(save_freq_select
, RTC_FREQ_SELECT
);
1458 clear_IO_APIC_pin(apic
, pin
);
1460 spin_lock_irqsave(&ioapic_lock
, flags
);
1461 io_apic_write(apic
, 0x11 + 2 * pin
, *(((int *)&entry0
) + 1));
1462 io_apic_write(apic
, 0x10 + 2 * pin
, *(((int *)&entry0
) + 0));
1463 spin_unlock_irqrestore(&ioapic_lock
, flags
);
1467 * This code may look a bit paranoid, but it's supposed to cooperate with
1468 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ
1469 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast
1470 * fanatically on his truly buggy board.
1472 * FIXME: really need to revamp this for modern platforms only.
1474 static inline void check_timer(void)
1476 int apic1
, pin1
, apic2
, pin2
;
1481 * get/set the timer IRQ vector:
1483 disable_8259A_irq(0);
1484 vector
= assign_irq_vector(0, TARGET_CPUS
, &mask
);
1487 * Subtle, code in do_timer_interrupt() expects an AEOI
1488 * mode for the 8259A whenever interrupts are routed
1489 * through I/O APICs. Also IRQ0 has to be enabled in
1490 * the 8259A which implies the virtual wire has to be
1491 * disabled in the local APIC.
1493 apic_write(APIC_LVT0
, APIC_LVT_MASKED
| APIC_DM_EXTINT
);
1495 if (timer_over_8254
> 0)
1496 enable_8259A_irq(0);
1498 pin1
= find_isa_irq_pin(0, mp_INT
);
1499 apic1
= find_isa_irq_apic(0, mp_INT
);
1500 pin2
= ioapic_i8259
.pin
;
1501 apic2
= ioapic_i8259
.apic
;
1503 apic_printk(APIC_VERBOSE
,KERN_INFO
"..TIMER: vector=0x%02X apic1=%d pin1=%d apic2=%d pin2=%d\n",
1504 vector
, apic1
, pin1
, apic2
, pin2
);
1508 * Ok, does IRQ0 through the IOAPIC work?
1510 unmask_IO_APIC_irq(0);
1511 if (!no_timer_check
&& timer_irq_works()) {
1512 nmi_watchdog_default();
1513 if (nmi_watchdog
== NMI_IO_APIC
) {
1514 disable_8259A_irq(0);
1516 enable_8259A_irq(0);
1518 if (disable_timer_pin_1
> 0)
1519 clear_IO_APIC_pin(0, pin1
);
1522 clear_IO_APIC_pin(apic1
, pin1
);
1523 apic_printk(APIC_QUIET
,KERN_ERR
"..MP-BIOS bug: 8254 timer not "
1524 "connected to IO-APIC\n");
1527 apic_printk(APIC_VERBOSE
,KERN_INFO
"...trying to set up timer (IRQ0) "
1528 "through the 8259A ... ");
1530 apic_printk(APIC_VERBOSE
,"\n..... (found apic %d pin %d) ...",
1533 * legacy devices should be connected to IO APIC #0
1535 setup_ExtINT_IRQ0_pin(apic2
, pin2
, vector
);
1536 if (timer_irq_works()) {
1537 apic_printk(APIC_VERBOSE
," works.\n");
1538 nmi_watchdog_default();
1539 if (nmi_watchdog
== NMI_IO_APIC
) {
1545 * Cleanup, just in case ...
1547 clear_IO_APIC_pin(apic2
, pin2
);
1549 apic_printk(APIC_VERBOSE
," failed.\n");
1551 if (nmi_watchdog
== NMI_IO_APIC
) {
1552 printk(KERN_WARNING
"timer doesn't work through the IO-APIC - disabling NMI Watchdog!\n");
1556 apic_printk(APIC_VERBOSE
, KERN_INFO
"...trying to set up timer as Virtual Wire IRQ...");
1558 disable_8259A_irq(0);
1559 irq_desc
[0].chip
= &lapic_irq_type
;
1560 apic_write(APIC_LVT0
, APIC_DM_FIXED
| vector
); /* Fixed mode */
1561 enable_8259A_irq(0);
1563 if (timer_irq_works()) {
1564 apic_printk(APIC_VERBOSE
," works.\n");
1567 apic_write(APIC_LVT0
, APIC_LVT_MASKED
| APIC_DM_FIXED
| vector
);
1568 apic_printk(APIC_VERBOSE
," failed.\n");
1570 apic_printk(APIC_VERBOSE
, KERN_INFO
"...trying to set up timer as ExtINT IRQ...");
1574 apic_write(APIC_LVT0
, APIC_DM_EXTINT
);
1576 unlock_ExtINT_logic();
1578 if (timer_irq_works()) {
1579 apic_printk(APIC_VERBOSE
," works.\n");
1582 apic_printk(APIC_VERBOSE
," failed :(.\n");
1583 panic("IO-APIC + timer doesn't work! Try using the 'noapic' kernel parameter\n");
1586 static int __init
notimercheck(char *s
)
1591 __setup("no_timer_check", notimercheck
);
1595 * IRQ's that are handled by the PIC in the MPS IOAPIC case.
1596 * - IRQ2 is the cascade IRQ, and cannot be a io-apic IRQ.
1597 * Linux doesn't really care, as it's not actually used
1598 * for any interrupt handling anyway.
1600 #define PIC_IRQS (1<<2)
1602 void __init
setup_IO_APIC(void)
1607 io_apic_irqs
= ~0; /* all IRQs go through IOAPIC */
1609 io_apic_irqs
= ~PIC_IRQS
;
1611 apic_printk(APIC_VERBOSE
, "ENABLING IO-APIC IRQs\n");
1614 setup_IO_APIC_irqs();
1615 init_IO_APIC_traps();
1621 struct sysfs_ioapic_data
{
1622 struct sys_device dev
;
1623 struct IO_APIC_route_entry entry
[0];
1625 static struct sysfs_ioapic_data
* mp_ioapic_data
[MAX_IO_APICS
];
1627 static int ioapic_suspend(struct sys_device
*dev
, pm_message_t state
)
1629 struct IO_APIC_route_entry
*entry
;
1630 struct sysfs_ioapic_data
*data
;
1633 data
= container_of(dev
, struct sysfs_ioapic_data
, dev
);
1634 entry
= data
->entry
;
1635 for (i
= 0; i
< nr_ioapic_registers
[dev
->id
]; i
++, entry
++ )
1636 *entry
= ioapic_read_entry(dev
->id
, i
);
1641 static int ioapic_resume(struct sys_device
*dev
)
1643 struct IO_APIC_route_entry
*entry
;
1644 struct sysfs_ioapic_data
*data
;
1645 unsigned long flags
;
1646 union IO_APIC_reg_00 reg_00
;
1649 data
= container_of(dev
, struct sysfs_ioapic_data
, dev
);
1650 entry
= data
->entry
;
1652 spin_lock_irqsave(&ioapic_lock
, flags
);
1653 reg_00
.raw
= io_apic_read(dev
->id
, 0);
1654 if (reg_00
.bits
.ID
!= mp_ioapics
[dev
->id
].mpc_apicid
) {
1655 reg_00
.bits
.ID
= mp_ioapics
[dev
->id
].mpc_apicid
;
1656 io_apic_write(dev
->id
, 0, reg_00
.raw
);
1658 spin_unlock_irqrestore(&ioapic_lock
, flags
);
1659 for (i
= 0; i
< nr_ioapic_registers
[dev
->id
]; i
++)
1660 ioapic_write_entry(dev
->id
, i
, entry
[i
]);
1665 static struct sysdev_class ioapic_sysdev_class
= {
1666 set_kset_name("ioapic"),
1667 .suspend
= ioapic_suspend
,
1668 .resume
= ioapic_resume
,
1671 static int __init
ioapic_init_sysfs(void)
1673 struct sys_device
* dev
;
1674 int i
, size
, error
= 0;
1676 error
= sysdev_class_register(&ioapic_sysdev_class
);
1680 for (i
= 0; i
< nr_ioapics
; i
++ ) {
1681 size
= sizeof(struct sys_device
) + nr_ioapic_registers
[i
]
1682 * sizeof(struct IO_APIC_route_entry
);
1683 mp_ioapic_data
[i
] = kmalloc(size
, GFP_KERNEL
);
1684 if (!mp_ioapic_data
[i
]) {
1685 printk(KERN_ERR
"Can't suspend/resume IOAPIC %d\n", i
);
1688 memset(mp_ioapic_data
[i
], 0, size
);
1689 dev
= &mp_ioapic_data
[i
]->dev
;
1691 dev
->cls
= &ioapic_sysdev_class
;
1692 error
= sysdev_register(dev
);
1694 kfree(mp_ioapic_data
[i
]);
1695 mp_ioapic_data
[i
] = NULL
;
1696 printk(KERN_ERR
"Can't suspend/resume IOAPIC %d\n", i
);
1704 device_initcall(ioapic_init_sysfs
);
1707 * Dynamic irq allocate and deallocation
1709 int create_irq(void)
1711 /* Allocate an unused irq */
1715 unsigned long flags
;
1719 spin_lock_irqsave(&vector_lock
, flags
);
1720 for (new = (NR_IRQS
- 1); new >= 0; new--) {
1721 if (platform_legacy_irq(new))
1723 if (irq_vector
[new] != 0)
1725 vector
= __assign_irq_vector(new, TARGET_CPUS
, &mask
);
1726 if (likely(vector
> 0))
1730 spin_unlock_irqrestore(&vector_lock
, flags
);
1733 dynamic_irq_init(irq
);
1738 void destroy_irq(unsigned int irq
)
1740 unsigned long flags
;
1742 dynamic_irq_cleanup(irq
);
1744 spin_lock_irqsave(&vector_lock
, flags
);
1745 irq_vector
[irq
] = 0;
1746 spin_unlock_irqrestore(&vector_lock
, flags
);
1750 * MSI mesage composition
1752 #ifdef CONFIG_PCI_MSI
1753 static int msi_compose_msg(struct pci_dev
*pdev
, unsigned int irq
, struct msi_msg
*msg
)
1759 vector
= assign_irq_vector(irq
, TARGET_CPUS
, &tmp
);
1761 dest
= cpu_mask_to_apicid(tmp
);
1763 msg
->address_hi
= MSI_ADDR_BASE_HI
;
1766 ((INT_DEST_MODE
== 0) ?
1767 MSI_ADDR_DEST_MODE_PHYSICAL
:
1768 MSI_ADDR_DEST_MODE_LOGICAL
) |
1769 ((INT_DELIVERY_MODE
!= dest_LowestPrio
) ?
1770 MSI_ADDR_REDIRECTION_CPU
:
1771 MSI_ADDR_REDIRECTION_LOWPRI
) |
1772 MSI_ADDR_DEST_ID(dest
);
1775 MSI_DATA_TRIGGER_EDGE
|
1776 MSI_DATA_LEVEL_ASSERT
|
1777 ((INT_DELIVERY_MODE
!= dest_LowestPrio
) ?
1778 MSI_DATA_DELIVERY_FIXED
:
1779 MSI_DATA_DELIVERY_LOWPRI
) |
1780 MSI_DATA_VECTOR(vector
);
1786 static void set_msi_irq_affinity(unsigned int irq
, cpumask_t mask
)
1793 cpus_and(tmp
, mask
, cpu_online_map
);
1794 if (cpus_empty(tmp
))
1797 cpus_and(mask
, tmp
, CPU_MASK_ALL
);
1799 vector
= assign_irq_vector(irq
, mask
, &tmp
);
1803 dest
= cpu_mask_to_apicid(tmp
);
1805 read_msi_msg(irq
, &msg
);
1807 msg
.data
&= ~MSI_DATA_VECTOR_MASK
;
1808 msg
.data
|= MSI_DATA_VECTOR(vector
);
1809 msg
.address_lo
&= ~MSI_ADDR_DEST_ID_MASK
;
1810 msg
.address_lo
|= MSI_ADDR_DEST_ID(dest
);
1812 write_msi_msg(irq
, &msg
);
1813 set_native_irq_info(irq
, mask
);
1815 #endif /* CONFIG_SMP */
1818 * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
1819 * which implement the MSI or MSI-X Capability Structure.
1821 static struct irq_chip msi_chip
= {
1823 .unmask
= unmask_msi_irq
,
1824 .mask
= mask_msi_irq
,
1825 .ack
= ack_apic_edge
,
1827 .set_affinity
= set_msi_irq_affinity
,
1829 .retrigger
= ioapic_retrigger_irq
,
1832 int arch_setup_msi_irq(unsigned int irq
, struct pci_dev
*dev
)
1836 ret
= msi_compose_msg(dev
, irq
, &msg
);
1840 write_msi_msg(irq
, &msg
);
1842 set_irq_chip_and_handler(irq
, &msi_chip
, handle_edge_irq
);
1847 void arch_teardown_msi_irq(unsigned int irq
)
1852 #endif /* CONFIG_PCI_MSI */
1855 * Hypertransport interrupt support
1857 #ifdef CONFIG_HT_IRQ
1861 static void target_ht_irq(unsigned int irq
, unsigned int dest
, u8 vector
)
1864 low
= read_ht_irq_low(irq
);
1865 high
= read_ht_irq_high(irq
);
1867 low
&= ~(HT_IRQ_LOW_VECTOR_MASK
| HT_IRQ_LOW_DEST_ID_MASK
);
1868 high
&= ~(HT_IRQ_HIGH_DEST_ID_MASK
);
1870 low
|= HT_IRQ_LOW_VECTOR(vector
) | HT_IRQ_LOW_DEST_ID(dest
);
1871 high
|= HT_IRQ_HIGH_DEST_ID(dest
);
1873 write_ht_irq_low(irq
, low
);
1874 write_ht_irq_high(irq
, high
);
1877 static void set_ht_irq_affinity(unsigned int irq
, cpumask_t mask
)
1883 cpus_and(tmp
, mask
, cpu_online_map
);
1884 if (cpus_empty(tmp
))
1887 cpus_and(mask
, tmp
, CPU_MASK_ALL
);
1889 vector
= assign_irq_vector(irq
, mask
, &tmp
);
1893 dest
= cpu_mask_to_apicid(tmp
);
1895 target_ht_irq(irq
, dest
, vector
& 0xff);
1896 set_native_irq_info(irq
, mask
);
1900 static struct hw_interrupt_type ht_irq_chip
= {
1902 .mask
= mask_ht_irq
,
1903 .unmask
= unmask_ht_irq
,
1904 .ack
= ack_apic_edge
,
1906 .set_affinity
= set_ht_irq_affinity
,
1908 .retrigger
= ioapic_retrigger_irq
,
1911 int arch_setup_ht_irq(unsigned int irq
, struct pci_dev
*dev
)
1916 vector
= assign_irq_vector(irq
, TARGET_CPUS
, &tmp
);
1921 dest
= cpu_mask_to_apicid(tmp
);
1923 high
= HT_IRQ_HIGH_DEST_ID(dest
);
1925 low
= HT_IRQ_LOW_BASE
|
1926 HT_IRQ_LOW_DEST_ID(dest
) |
1927 HT_IRQ_LOW_VECTOR(vector
) |
1928 ((INT_DEST_MODE
== 0) ?
1929 HT_IRQ_LOW_DM_PHYSICAL
:
1930 HT_IRQ_LOW_DM_LOGICAL
) |
1931 HT_IRQ_LOW_RQEOI_EDGE
|
1932 ((INT_DELIVERY_MODE
!= dest_LowestPrio
) ?
1933 HT_IRQ_LOW_MT_FIXED
:
1934 HT_IRQ_LOW_MT_ARBITRATED
);
1936 write_ht_irq_low(irq
, low
);
1937 write_ht_irq_high(irq
, high
);
1939 set_irq_chip_and_handler(irq
, &ht_irq_chip
, handle_edge_irq
);
1943 #endif /* CONFIG_HT_IRQ */
1945 /* --------------------------------------------------------------------------
1946 ACPI-based IOAPIC Configuration
1947 -------------------------------------------------------------------------- */
1951 #define IO_APIC_MAX_ID 0xFE
1953 int __init
io_apic_get_redir_entries (int ioapic
)
1955 union IO_APIC_reg_01 reg_01
;
1956 unsigned long flags
;
1958 spin_lock_irqsave(&ioapic_lock
, flags
);
1959 reg_01
.raw
= io_apic_read(ioapic
, 1);
1960 spin_unlock_irqrestore(&ioapic_lock
, flags
);
1962 return reg_01
.bits
.entries
;
1966 int io_apic_set_pci_routing (int ioapic
, int pin
, int irq
, int triggering
, int polarity
)
1968 struct IO_APIC_route_entry entry
;
1969 unsigned long flags
;
1973 if (!IO_APIC_IRQ(irq
)) {
1974 apic_printk(APIC_QUIET
,KERN_ERR
"IOAPIC[%d]: Invalid reference to IRQ 0\n",
1980 * IRQs < 16 are already in the irq_2_pin[] map
1983 add_pin_to_irq(irq
, ioapic
, pin
);
1986 vector
= assign_irq_vector(irq
, TARGET_CPUS
, &mask
);
1991 * Generate a PCI IRQ routing entry and program the IOAPIC accordingly.
1992 * Note that we mask (disable) IRQs now -- these get enabled when the
1993 * corresponding device driver registers for this IRQ.
1996 memset(&entry
,0,sizeof(entry
));
1998 entry
.delivery_mode
= INT_DELIVERY_MODE
;
1999 entry
.dest_mode
= INT_DEST_MODE
;
2000 entry
.dest
.logical
.logical_dest
= cpu_mask_to_apicid(mask
);
2001 entry
.trigger
= triggering
;
2002 entry
.polarity
= polarity
;
2003 entry
.mask
= 1; /* Disabled (masked) */
2004 entry
.vector
= vector
& 0xff;
2006 apic_printk(APIC_VERBOSE
,KERN_DEBUG
"IOAPIC[%d]: Set PCI routing entry (%d-%d -> 0x%x -> "
2007 "IRQ %d Mode:%i Active:%i)\n", ioapic
,
2008 mp_ioapics
[ioapic
].mpc_apicid
, pin
, entry
.vector
, irq
,
2009 triggering
, polarity
);
2011 ioapic_register_intr(irq
, entry
.vector
, triggering
);
2013 if (!ioapic
&& (irq
< 16))
2014 disable_8259A_irq(irq
);
2016 ioapic_write_entry(ioapic
, pin
, entry
);
2018 spin_lock_irqsave(&ioapic_lock
, flags
);
2019 set_native_irq_info(irq
, TARGET_CPUS
);
2020 spin_unlock_irqrestore(&ioapic_lock
, flags
);
2025 #endif /* CONFIG_ACPI */
2029 * This function currently is only a helper for the i386 smp boot process where
2030 * we need to reprogram the ioredtbls to cater for the cpus which have come online
2031 * so mask in all cases should simply be TARGET_CPUS
2034 void __init
setup_ioapic_dest(void)
2036 int pin
, ioapic
, irq
, irq_entry
;
2038 if (skip_ioapic_setup
== 1)
2041 for (ioapic
= 0; ioapic
< nr_ioapics
; ioapic
++) {
2042 for (pin
= 0; pin
< nr_ioapic_registers
[ioapic
]; pin
++) {
2043 irq_entry
= find_irq_entry(ioapic
, pin
, mp_INT
);
2044 if (irq_entry
== -1)
2046 irq
= pin_2_irq(irq_entry
, ioapic
, pin
);
2047 set_ioapic_affinity_irq(irq
, TARGET_CPUS
);