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 /* Where if anywhere is the i8259 connect in external int mode */
59 static struct { int pin
, apic
; } ioapic_i8259
= { -1, -1 };
61 static DEFINE_SPINLOCK(ioapic_lock
);
62 DEFINE_SPINLOCK(vector_lock
);
65 * # of IRQ routing registers
67 int nr_ioapic_registers
[MAX_IO_APICS
];
70 * Rough estimation of how many shared IRQs there are, can
73 #define MAX_PLUS_SHARED_IRQS NR_IRQ_VECTORS
74 #define PIN_MAP_SIZE (MAX_PLUS_SHARED_IRQS + NR_IRQS)
77 * This is performance-critical, we want to do it O(1)
79 * the indexing order of this array favors 1:1 mappings
80 * between pins and IRQs.
83 static struct irq_pin_list
{
84 short apic
, pin
, next
;
85 } irq_2_pin
[PIN_MAP_SIZE
];
89 unsigned int unused
[3];
93 static __attribute_const__
struct io_apic __iomem
*io_apic_base(int idx
)
95 return (void __iomem
*) __fix_to_virt(FIX_IO_APIC_BASE_0
+ idx
)
96 + (mp_ioapics
[idx
].mpc_apicaddr
& ~PAGE_MASK
);
99 static inline unsigned int io_apic_read(unsigned int apic
, unsigned int reg
)
101 struct io_apic __iomem
*io_apic
= io_apic_base(apic
);
102 writel(reg
, &io_apic
->index
);
103 return readl(&io_apic
->data
);
106 static inline void io_apic_write(unsigned int apic
, unsigned int reg
, unsigned int value
)
108 struct io_apic __iomem
*io_apic
= io_apic_base(apic
);
109 writel(reg
, &io_apic
->index
);
110 writel(value
, &io_apic
->data
);
114 * Re-write a value: to be used for read-modify-write
115 * cycles where the read already set up the index register.
117 static inline void io_apic_modify(unsigned int apic
, unsigned int value
)
119 struct io_apic __iomem
*io_apic
= io_apic_base(apic
);
120 writel(value
, &io_apic
->data
);
124 * Synchronize the IO-APIC and the CPU by doing
125 * a dummy read from the IO-APIC
127 static inline void io_apic_sync(unsigned int apic
)
129 struct io_apic __iomem
*io_apic
= io_apic_base(apic
);
130 readl(&io_apic
->data
);
133 #define __DO_ACTION(R, ACTION, FINAL) \
137 struct irq_pin_list *entry = irq_2_pin + irq; \
139 BUG_ON(irq >= NR_IRQS); \
145 reg = io_apic_read(entry->apic, 0x10 + R + pin*2); \
147 io_apic_modify(entry->apic, reg); \
150 entry = irq_2_pin + entry->next; \
156 struct { u32 w1
, w2
; };
157 struct IO_APIC_route_entry entry
;
160 static struct IO_APIC_route_entry
ioapic_read_entry(int apic
, int pin
)
162 union entry_union eu
;
164 spin_lock_irqsave(&ioapic_lock
, flags
);
165 eu
.w1
= io_apic_read(apic
, 0x10 + 2 * pin
);
166 eu
.w2
= io_apic_read(apic
, 0x11 + 2 * pin
);
167 spin_unlock_irqrestore(&ioapic_lock
, flags
);
172 * When we write a new IO APIC routing entry, we need to write the high
173 * word first! If the mask bit in the low word is clear, we will enable
174 * the interrupt, and we need to make sure the entry is fully populated
175 * before that happens.
178 __ioapic_write_entry(int apic
, int pin
, struct IO_APIC_route_entry e
)
180 union entry_union eu
;
182 io_apic_write(apic
, 0x11 + 2*pin
, eu
.w2
);
183 io_apic_write(apic
, 0x10 + 2*pin
, eu
.w1
);
186 static void ioapic_write_entry(int apic
, int pin
, struct IO_APIC_route_entry e
)
189 spin_lock_irqsave(&ioapic_lock
, flags
);
190 __ioapic_write_entry(apic
, pin
, e
);
191 spin_unlock_irqrestore(&ioapic_lock
, flags
);
195 * When we mask an IO APIC routing entry, we need to write the low
196 * word first, in order to set the mask bit before we change the
199 static void ioapic_mask_entry(int apic
, int pin
)
202 union entry_union eu
= { .entry
.mask
= 1 };
204 spin_lock_irqsave(&ioapic_lock
, flags
);
205 io_apic_write(apic
, 0x10 + 2*pin
, eu
.w1
);
206 io_apic_write(apic
, 0x11 + 2*pin
, eu
.w2
);
207 spin_unlock_irqrestore(&ioapic_lock
, flags
);
211 static void __target_IO_APIC_irq(unsigned int irq
, unsigned int dest
, u8 vector
)
214 struct irq_pin_list
*entry
= irq_2_pin
+ irq
;
216 BUG_ON(irq
>= NR_IRQS
);
223 io_apic_write(apic
, 0x11 + pin
*2, dest
);
224 reg
= io_apic_read(apic
, 0x10 + pin
*2);
227 io_apic_modify(apic
, reg
);
230 entry
= irq_2_pin
+ entry
->next
;
234 static void set_ioapic_affinity_irq(unsigned int irq
, cpumask_t mask
)
241 cpus_and(tmp
, mask
, cpu_online_map
);
245 cpus_and(mask
, tmp
, CPU_MASK_ALL
);
247 vector
= assign_irq_vector(irq
, mask
, &tmp
);
251 dest
= cpu_mask_to_apicid(tmp
);
254 * Only the high 8 bits are valid.
256 dest
= SET_APIC_LOGICAL_ID(dest
);
258 spin_lock_irqsave(&ioapic_lock
, flags
);
259 __target_IO_APIC_irq(irq
, dest
, vector
);
260 set_native_irq_info(irq
, mask
);
261 spin_unlock_irqrestore(&ioapic_lock
, flags
);
266 * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
267 * shared ISA-space IRQs, so we have to support them. We are super
268 * fast in the common case, and fast for shared ISA-space IRQs.
270 static void add_pin_to_irq(unsigned int irq
, int apic
, int pin
)
272 static int first_free_entry
= NR_IRQS
;
273 struct irq_pin_list
*entry
= irq_2_pin
+ irq
;
275 BUG_ON(irq
>= NR_IRQS
);
277 entry
= irq_2_pin
+ entry
->next
;
279 if (entry
->pin
!= -1) {
280 entry
->next
= first_free_entry
;
281 entry
= irq_2_pin
+ entry
->next
;
282 if (++first_free_entry
>= PIN_MAP_SIZE
)
283 panic("io_apic.c: ran out of irq_2_pin entries!");
290 #define DO_ACTION(name,R,ACTION, FINAL) \
292 static void name##_IO_APIC_irq (unsigned int irq) \
293 __DO_ACTION(R, ACTION, FINAL)
295 DO_ACTION( __mask
, 0, |= 0x00010000, io_apic_sync(entry
->apic
) )
297 DO_ACTION( __unmask
, 0, &= 0xfffeffff, )
300 static void mask_IO_APIC_irq (unsigned int irq
)
304 spin_lock_irqsave(&ioapic_lock
, flags
);
305 __mask_IO_APIC_irq(irq
);
306 spin_unlock_irqrestore(&ioapic_lock
, flags
);
309 static void unmask_IO_APIC_irq (unsigned int irq
)
313 spin_lock_irqsave(&ioapic_lock
, flags
);
314 __unmask_IO_APIC_irq(irq
);
315 spin_unlock_irqrestore(&ioapic_lock
, flags
);
318 static void clear_IO_APIC_pin(unsigned int apic
, unsigned int pin
)
320 struct IO_APIC_route_entry entry
;
322 /* Check delivery_mode to be sure we're not clearing an SMI pin */
323 entry
= ioapic_read_entry(apic
, pin
);
324 if (entry
.delivery_mode
== dest_SMI
)
327 * Disable it in the IO-APIC irq-routing table:
329 ioapic_mask_entry(apic
, pin
);
332 static void clear_IO_APIC (void)
336 for (apic
= 0; apic
< nr_ioapics
; apic
++)
337 for (pin
= 0; pin
< nr_ioapic_registers
[apic
]; pin
++)
338 clear_IO_APIC_pin(apic
, pin
);
341 int skip_ioapic_setup
;
344 /* dummy parsing: see setup.c */
346 static int __init
disable_ioapic_setup(char *str
)
348 skip_ioapic_setup
= 1;
351 early_param("noapic", disable_ioapic_setup
);
354 * Find the IRQ entry number of a certain pin.
356 static int find_irq_entry(int apic
, int pin
, int type
)
360 for (i
= 0; i
< mp_irq_entries
; i
++)
361 if (mp_irqs
[i
].mpc_irqtype
== type
&&
362 (mp_irqs
[i
].mpc_dstapic
== mp_ioapics
[apic
].mpc_apicid
||
363 mp_irqs
[i
].mpc_dstapic
== MP_APIC_ALL
) &&
364 mp_irqs
[i
].mpc_dstirq
== pin
)
371 * Find the pin to which IRQ[irq] (ISA) is connected
373 static int __init
find_isa_irq_pin(int irq
, int type
)
377 for (i
= 0; i
< mp_irq_entries
; i
++) {
378 int lbus
= mp_irqs
[i
].mpc_srcbus
;
380 if (test_bit(lbus
, mp_bus_not_pci
) &&
381 (mp_irqs
[i
].mpc_irqtype
== type
) &&
382 (mp_irqs
[i
].mpc_srcbusirq
== irq
))
384 return mp_irqs
[i
].mpc_dstirq
;
389 static int __init
find_isa_irq_apic(int irq
, int type
)
393 for (i
= 0; i
< mp_irq_entries
; i
++) {
394 int lbus
= mp_irqs
[i
].mpc_srcbus
;
396 if (test_bit(lbus
, mp_bus_not_pci
) &&
397 (mp_irqs
[i
].mpc_irqtype
== type
) &&
398 (mp_irqs
[i
].mpc_srcbusirq
== irq
))
401 if (i
< mp_irq_entries
) {
403 for(apic
= 0; apic
< nr_ioapics
; apic
++) {
404 if (mp_ioapics
[apic
].mpc_apicid
== mp_irqs
[i
].mpc_dstapic
)
413 * Find a specific PCI IRQ entry.
414 * Not an __init, possibly needed by modules
416 static int pin_2_irq(int idx
, int apic
, int pin
);
418 int IO_APIC_get_PCI_irq_vector(int bus
, int slot
, int pin
)
420 int apic
, i
, best_guess
= -1;
422 apic_printk(APIC_DEBUG
, "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
424 if (mp_bus_id_to_pci_bus
[bus
] == -1) {
425 apic_printk(APIC_VERBOSE
, "PCI BIOS passed nonexistent PCI bus %d!\n", bus
);
428 for (i
= 0; i
< mp_irq_entries
; i
++) {
429 int lbus
= mp_irqs
[i
].mpc_srcbus
;
431 for (apic
= 0; apic
< nr_ioapics
; apic
++)
432 if (mp_ioapics
[apic
].mpc_apicid
== mp_irqs
[i
].mpc_dstapic
||
433 mp_irqs
[i
].mpc_dstapic
== MP_APIC_ALL
)
436 if (!test_bit(lbus
, mp_bus_not_pci
) &&
437 !mp_irqs
[i
].mpc_irqtype
&&
439 (slot
== ((mp_irqs
[i
].mpc_srcbusirq
>> 2) & 0x1f))) {
440 int irq
= pin_2_irq(i
,apic
,mp_irqs
[i
].mpc_dstirq
);
442 if (!(apic
|| IO_APIC_IRQ(irq
)))
445 if (pin
== (mp_irqs
[i
].mpc_srcbusirq
& 3))
448 * Use the first all-but-pin matching entry as a
449 * best-guess fuzzy result for broken mptables.
455 BUG_ON(best_guess
>= NR_IRQS
);
459 /* ISA interrupts are always polarity zero edge triggered,
460 * when listed as conforming in the MP table. */
462 #define default_ISA_trigger(idx) (0)
463 #define default_ISA_polarity(idx) (0)
465 /* PCI interrupts are always polarity one level triggered,
466 * when listed as conforming in the MP table. */
468 #define default_PCI_trigger(idx) (1)
469 #define default_PCI_polarity(idx) (1)
471 static int __init
MPBIOS_polarity(int idx
)
473 int bus
= mp_irqs
[idx
].mpc_srcbus
;
477 * Determine IRQ line polarity (high active or low active):
479 switch (mp_irqs
[idx
].mpc_irqflag
& 3)
481 case 0: /* conforms, ie. bus-type dependent polarity */
482 if (test_bit(bus
, mp_bus_not_pci
))
483 polarity
= default_ISA_polarity(idx
);
485 polarity
= default_PCI_polarity(idx
);
487 case 1: /* high active */
492 case 2: /* reserved */
494 printk(KERN_WARNING
"broken BIOS!!\n");
498 case 3: /* low active */
503 default: /* invalid */
505 printk(KERN_WARNING
"broken BIOS!!\n");
513 static int MPBIOS_trigger(int idx
)
515 int bus
= mp_irqs
[idx
].mpc_srcbus
;
519 * Determine IRQ trigger mode (edge or level sensitive):
521 switch ((mp_irqs
[idx
].mpc_irqflag
>>2) & 3)
523 case 0: /* conforms, ie. bus-type dependent */
524 if (test_bit(bus
, mp_bus_not_pci
))
525 trigger
= default_ISA_trigger(idx
);
527 trigger
= default_PCI_trigger(idx
);
534 case 2: /* reserved */
536 printk(KERN_WARNING
"broken BIOS!!\n");
545 default: /* invalid */
547 printk(KERN_WARNING
"broken BIOS!!\n");
555 static inline int irq_polarity(int idx
)
557 return MPBIOS_polarity(idx
);
560 static inline int irq_trigger(int idx
)
562 return MPBIOS_trigger(idx
);
565 static int pin_2_irq(int idx
, int apic
, int pin
)
568 int bus
= mp_irqs
[idx
].mpc_srcbus
;
571 * Debugging check, we are in big trouble if this message pops up!
573 if (mp_irqs
[idx
].mpc_dstirq
!= pin
)
574 printk(KERN_ERR
"broken BIOS or MPTABLE parser, ayiee!!\n");
576 if (test_bit(bus
, mp_bus_not_pci
)) {
577 irq
= mp_irqs
[idx
].mpc_srcbusirq
;
580 * PCI IRQs are mapped in order
584 irq
+= nr_ioapic_registers
[i
++];
587 BUG_ON(irq
>= NR_IRQS
);
591 static inline int IO_APIC_irq_trigger(int irq
)
595 for (apic
= 0; apic
< nr_ioapics
; apic
++) {
596 for (pin
= 0; pin
< nr_ioapic_registers
[apic
]; pin
++) {
597 idx
= find_irq_entry(apic
,pin
,mp_INT
);
598 if ((idx
!= -1) && (irq
== pin_2_irq(idx
,apic
,pin
)))
599 return irq_trigger(idx
);
603 * nonexistent IRQs are edge default
608 /* irq_vectors is indexed by the sum of all RTEs in all I/O APICs. */
609 static u8 irq_vector
[NR_IRQ_VECTORS
] __read_mostly
= {
610 [0] = FIRST_EXTERNAL_VECTOR
+ 0,
611 [1] = FIRST_EXTERNAL_VECTOR
+ 1,
612 [2] = FIRST_EXTERNAL_VECTOR
+ 2,
613 [3] = FIRST_EXTERNAL_VECTOR
+ 3,
614 [4] = FIRST_EXTERNAL_VECTOR
+ 4,
615 [5] = FIRST_EXTERNAL_VECTOR
+ 5,
616 [6] = FIRST_EXTERNAL_VECTOR
+ 6,
617 [7] = FIRST_EXTERNAL_VECTOR
+ 7,
618 [8] = FIRST_EXTERNAL_VECTOR
+ 8,
619 [9] = FIRST_EXTERNAL_VECTOR
+ 9,
620 [10] = FIRST_EXTERNAL_VECTOR
+ 10,
621 [11] = FIRST_EXTERNAL_VECTOR
+ 11,
622 [12] = FIRST_EXTERNAL_VECTOR
+ 12,
623 [13] = FIRST_EXTERNAL_VECTOR
+ 13,
624 [14] = FIRST_EXTERNAL_VECTOR
+ 14,
625 [15] = FIRST_EXTERNAL_VECTOR
+ 15,
628 static cpumask_t irq_domain
[NR_IRQ_VECTORS
] __read_mostly
= {
647 static int __assign_irq_vector(int irq
, cpumask_t mask
, cpumask_t
*result
)
650 * NOTE! The local APIC isn't very good at handling
651 * multiple interrupts at the same interrupt level.
652 * As the interrupt level is determined by taking the
653 * vector number and shifting that right by 4, we
654 * want to spread these out a bit so that they don't
655 * all fall in the same interrupt level.
657 * Also, we've got to be careful not to trash gate
658 * 0x80, because int 0x80 is hm, kind of importantish. ;)
660 static int current_vector
= FIRST_DEVICE_VECTOR
, current_offset
= 0;
664 BUG_ON((unsigned)irq
>= NR_IRQ_VECTORS
);
666 /* Only try and allocate irqs on cpus that are present */
667 cpus_and(mask
, mask
, cpu_online_map
);
669 if (irq_vector
[irq
] > 0)
670 old_vector
= irq_vector
[irq
];
671 if (old_vector
> 0) {
672 cpus_and(*result
, irq_domain
[irq
], mask
);
673 if (!cpus_empty(*result
))
677 for_each_cpu_mask(cpu
, mask
) {
678 cpumask_t domain
, new_mask
;
682 domain
= vector_allocation_domain(cpu
);
683 cpus_and(new_mask
, domain
, cpu_online_map
);
685 vector
= current_vector
;
686 offset
= current_offset
;
689 if (vector
>= FIRST_SYSTEM_VECTOR
) {
690 /* If we run out of vectors on large boxen, must share them. */
691 offset
= (offset
+ 1) % 8;
692 vector
= FIRST_DEVICE_VECTOR
+ offset
;
694 if (unlikely(current_vector
== vector
))
696 if (vector
== IA32_SYSCALL_VECTOR
)
698 for_each_cpu_mask(new_cpu
, new_mask
)
699 if (per_cpu(vector_irq
, new_cpu
)[vector
] != -1)
702 current_vector
= vector
;
703 current_offset
= offset
;
704 if (old_vector
>= 0) {
707 cpus_and(old_mask
, irq_domain
[irq
], cpu_online_map
);
708 for_each_cpu_mask(old_cpu
, old_mask
)
709 per_cpu(vector_irq
, old_cpu
)[old_vector
] = -1;
711 for_each_cpu_mask(new_cpu
, new_mask
)
712 per_cpu(vector_irq
, new_cpu
)[vector
] = irq
;
713 irq_vector
[irq
] = vector
;
714 irq_domain
[irq
] = domain
;
715 cpus_and(*result
, domain
, mask
);
721 static int assign_irq_vector(int irq
, cpumask_t mask
, cpumask_t
*result
)
726 spin_lock_irqsave(&vector_lock
, flags
);
727 vector
= __assign_irq_vector(irq
, mask
, result
);
728 spin_unlock_irqrestore(&vector_lock
, flags
);
732 static void __clear_irq_vector(int irq
)
737 BUG_ON(!irq_vector
[irq
]);
739 vector
= irq_vector
[irq
];
740 cpus_and(mask
, irq_domain
[irq
], cpu_online_map
);
741 for_each_cpu_mask(cpu
, mask
)
742 per_cpu(vector_irq
, cpu
)[vector
] = -1;
745 irq_domain
[irq
] = CPU_MASK_NONE
;
748 void __setup_vector_irq(int cpu
)
750 /* Initialize vector_irq on a new cpu */
751 /* This function must be called with vector_lock held */
754 /* Mark the inuse vectors */
755 for (irq
= 0; irq
< NR_IRQ_VECTORS
; ++irq
) {
756 if (!cpu_isset(cpu
, irq_domain
[irq
]))
758 vector
= irq_vector
[irq
];
759 per_cpu(vector_irq
, cpu
)[vector
] = irq
;
761 /* Mark the free vectors */
762 for (vector
= 0; vector
< NR_VECTORS
; ++vector
) {
763 irq
= per_cpu(vector_irq
, cpu
)[vector
];
766 if (!cpu_isset(cpu
, irq_domain
[irq
]))
767 per_cpu(vector_irq
, cpu
)[vector
] = -1;
772 extern void (*interrupt
[NR_IRQS
])(void);
774 static struct irq_chip ioapic_chip
;
776 #define IOAPIC_AUTO -1
777 #define IOAPIC_EDGE 0
778 #define IOAPIC_LEVEL 1
780 static void ioapic_register_intr(int irq
, int vector
, unsigned long trigger
)
782 if ((trigger
== IOAPIC_AUTO
&& IO_APIC_irq_trigger(irq
)) ||
783 trigger
== IOAPIC_LEVEL
)
784 set_irq_chip_and_handler_name(irq
, &ioapic_chip
,
785 handle_fasteoi_irq
, "fasteoi");
787 irq_desc
[irq
].status
|= IRQ_DELAYED_DISABLE
;
788 set_irq_chip_and_handler_name(irq
, &ioapic_chip
,
789 handle_edge_irq
, "edge");
792 static void __init
setup_IO_APIC_irq(int apic
, int pin
, int idx
, int irq
)
794 struct IO_APIC_route_entry entry
;
800 * add it to the IO-APIC irq-routing table:
802 memset(&entry
,0,sizeof(entry
));
804 entry
.delivery_mode
= INT_DELIVERY_MODE
;
805 entry
.dest_mode
= INT_DEST_MODE
;
806 entry
.mask
= 0; /* enable IRQ */
807 entry
.dest
.logical
.logical_dest
= cpu_mask_to_apicid(TARGET_CPUS
);
809 entry
.trigger
= irq_trigger(idx
);
810 entry
.polarity
= irq_polarity(idx
);
812 if (irq_trigger(idx
)) {
815 entry
.dest
.logical
.logical_dest
= cpu_mask_to_apicid(TARGET_CPUS
);
818 if (!apic
&& !IO_APIC_IRQ(irq
))
821 if (IO_APIC_IRQ(irq
)) {
823 vector
= assign_irq_vector(irq
, TARGET_CPUS
, &mask
);
827 entry
.dest
.logical
.logical_dest
= cpu_mask_to_apicid(mask
);
828 entry
.vector
= vector
;
830 ioapic_register_intr(irq
, vector
, IOAPIC_AUTO
);
831 if (!apic
&& (irq
< 16))
832 disable_8259A_irq(irq
);
835 ioapic_write_entry(apic
, pin
, entry
);
837 spin_lock_irqsave(&ioapic_lock
, flags
);
838 set_native_irq_info(irq
, TARGET_CPUS
);
839 spin_unlock_irqrestore(&ioapic_lock
, flags
);
843 static void __init
setup_IO_APIC_irqs(void)
845 int apic
, pin
, idx
, irq
, first_notcon
= 1;
847 apic_printk(APIC_VERBOSE
, KERN_DEBUG
"init IO_APIC IRQs\n");
849 for (apic
= 0; apic
< nr_ioapics
; apic
++) {
850 for (pin
= 0; pin
< nr_ioapic_registers
[apic
]; pin
++) {
852 idx
= find_irq_entry(apic
,pin
,mp_INT
);
855 apic_printk(APIC_VERBOSE
, KERN_DEBUG
" IO-APIC (apicid-pin) %d-%d", mp_ioapics
[apic
].mpc_apicid
, pin
);
858 apic_printk(APIC_VERBOSE
, ", %d-%d", mp_ioapics
[apic
].mpc_apicid
, pin
);
862 irq
= pin_2_irq(idx
, apic
, pin
);
863 add_pin_to_irq(irq
, apic
, pin
);
865 setup_IO_APIC_irq(apic
, pin
, idx
, irq
);
871 apic_printk(APIC_VERBOSE
," not connected.\n");
875 * Set up the 8259A-master output pin as broadcast to all
878 static void __init
setup_ExtINT_IRQ0_pin(unsigned int apic
, unsigned int pin
, int vector
)
880 struct IO_APIC_route_entry entry
;
883 memset(&entry
,0,sizeof(entry
));
885 disable_8259A_irq(0);
888 apic_write(APIC_LVT0
, APIC_LVT_MASKED
| APIC_DM_EXTINT
);
891 * We use logical delivery to get the timer IRQ
894 entry
.dest_mode
= INT_DEST_MODE
;
895 entry
.mask
= 0; /* unmask IRQ now */
896 entry
.dest
.logical
.logical_dest
= cpu_mask_to_apicid(TARGET_CPUS
);
897 entry
.delivery_mode
= INT_DELIVERY_MODE
;
900 entry
.vector
= vector
;
903 * The timer IRQ doesn't have to know that behind the
904 * scene we have a 8259A-master in AEOI mode ...
906 set_irq_chip_and_handler_name(0, &ioapic_chip
, handle_edge_irq
, "edge");
909 * Add it to the IO-APIC irq-routing table:
911 spin_lock_irqsave(&ioapic_lock
, flags
);
912 io_apic_write(apic
, 0x11+2*pin
, *(((int *)&entry
)+1));
913 io_apic_write(apic
, 0x10+2*pin
, *(((int *)&entry
)+0));
914 spin_unlock_irqrestore(&ioapic_lock
, flags
);
919 void __init
UNEXPECTED_IO_APIC(void)
923 void __apicdebuginit
print_IO_APIC(void)
926 union IO_APIC_reg_00 reg_00
;
927 union IO_APIC_reg_01 reg_01
;
928 union IO_APIC_reg_02 reg_02
;
931 if (apic_verbosity
== APIC_QUIET
)
934 printk(KERN_DEBUG
"number of MP IRQ sources: %d.\n", mp_irq_entries
);
935 for (i
= 0; i
< nr_ioapics
; i
++)
936 printk(KERN_DEBUG
"number of IO-APIC #%d registers: %d.\n",
937 mp_ioapics
[i
].mpc_apicid
, nr_ioapic_registers
[i
]);
940 * We are a bit conservative about what we expect. We have to
941 * know about every hardware change ASAP.
943 printk(KERN_INFO
"testing the IO APIC.......................\n");
945 for (apic
= 0; apic
< nr_ioapics
; apic
++) {
947 spin_lock_irqsave(&ioapic_lock
, flags
);
948 reg_00
.raw
= io_apic_read(apic
, 0);
949 reg_01
.raw
= io_apic_read(apic
, 1);
950 if (reg_01
.bits
.version
>= 0x10)
951 reg_02
.raw
= io_apic_read(apic
, 2);
952 spin_unlock_irqrestore(&ioapic_lock
, flags
);
955 printk(KERN_DEBUG
"IO APIC #%d......\n", mp_ioapics
[apic
].mpc_apicid
);
956 printk(KERN_DEBUG
".... register #00: %08X\n", reg_00
.raw
);
957 printk(KERN_DEBUG
"....... : physical APIC id: %02X\n", reg_00
.bits
.ID
);
958 if (reg_00
.bits
.__reserved_1
|| reg_00
.bits
.__reserved_2
)
959 UNEXPECTED_IO_APIC();
961 printk(KERN_DEBUG
".... register #01: %08X\n", *(int *)®_01
);
962 printk(KERN_DEBUG
"....... : max redirection entries: %04X\n", reg_01
.bits
.entries
);
963 if ( (reg_01
.bits
.entries
!= 0x0f) && /* older (Neptune) boards */
964 (reg_01
.bits
.entries
!= 0x17) && /* typical ISA+PCI boards */
965 (reg_01
.bits
.entries
!= 0x1b) && /* Compaq Proliant boards */
966 (reg_01
.bits
.entries
!= 0x1f) && /* dual Xeon boards */
967 (reg_01
.bits
.entries
!= 0x22) && /* bigger Xeon boards */
968 (reg_01
.bits
.entries
!= 0x2E) &&
969 (reg_01
.bits
.entries
!= 0x3F) &&
970 (reg_01
.bits
.entries
!= 0x03)
972 UNEXPECTED_IO_APIC();
974 printk(KERN_DEBUG
"....... : PRQ implemented: %X\n", reg_01
.bits
.PRQ
);
975 printk(KERN_DEBUG
"....... : IO APIC version: %04X\n", reg_01
.bits
.version
);
976 if ( (reg_01
.bits
.version
!= 0x01) && /* 82489DX IO-APICs */
977 (reg_01
.bits
.version
!= 0x02) && /* 82801BA IO-APICs (ICH2) */
978 (reg_01
.bits
.version
!= 0x10) && /* oldest IO-APICs */
979 (reg_01
.bits
.version
!= 0x11) && /* Pentium/Pro IO-APICs */
980 (reg_01
.bits
.version
!= 0x13) && /* Xeon IO-APICs */
981 (reg_01
.bits
.version
!= 0x20) /* Intel P64H (82806 AA) */
983 UNEXPECTED_IO_APIC();
984 if (reg_01
.bits
.__reserved_1
|| reg_01
.bits
.__reserved_2
)
985 UNEXPECTED_IO_APIC();
987 if (reg_01
.bits
.version
>= 0x10) {
988 printk(KERN_DEBUG
".... register #02: %08X\n", reg_02
.raw
);
989 printk(KERN_DEBUG
"....... : arbitration: %02X\n", reg_02
.bits
.arbitration
);
990 if (reg_02
.bits
.__reserved_1
|| reg_02
.bits
.__reserved_2
)
991 UNEXPECTED_IO_APIC();
994 printk(KERN_DEBUG
".... IRQ redirection table:\n");
996 printk(KERN_DEBUG
" NR Log Phy Mask Trig IRR Pol"
997 " Stat Dest Deli Vect: \n");
999 for (i
= 0; i
<= reg_01
.bits
.entries
; i
++) {
1000 struct IO_APIC_route_entry entry
;
1002 entry
= ioapic_read_entry(apic
, i
);
1004 printk(KERN_DEBUG
" %02x %03X %02X ",
1006 entry
.dest
.logical
.logical_dest
,
1007 entry
.dest
.physical
.physical_dest
1010 printk("%1d %1d %1d %1d %1d %1d %1d %02X\n",
1015 entry
.delivery_status
,
1017 entry
.delivery_mode
,
1022 printk(KERN_DEBUG
"IRQ to pin mappings:\n");
1023 for (i
= 0; i
< NR_IRQS
; i
++) {
1024 struct irq_pin_list
*entry
= irq_2_pin
+ i
;
1027 printk(KERN_DEBUG
"IRQ%d ", i
);
1029 printk("-> %d:%d", entry
->apic
, entry
->pin
);
1032 entry
= irq_2_pin
+ entry
->next
;
1037 printk(KERN_INFO
".................................... done.\n");
1044 static __apicdebuginit
void print_APIC_bitfield (int base
)
1049 if (apic_verbosity
== APIC_QUIET
)
1052 printk(KERN_DEBUG
"0123456789abcdef0123456789abcdef\n" KERN_DEBUG
);
1053 for (i
= 0; i
< 8; i
++) {
1054 v
= apic_read(base
+ i
*0x10);
1055 for (j
= 0; j
< 32; j
++) {
1065 void __apicdebuginit
print_local_APIC(void * dummy
)
1067 unsigned int v
, ver
, maxlvt
;
1069 if (apic_verbosity
== APIC_QUIET
)
1072 printk("\n" KERN_DEBUG
"printing local APIC contents on CPU#%d/%d:\n",
1073 smp_processor_id(), hard_smp_processor_id());
1074 v
= apic_read(APIC_ID
);
1075 printk(KERN_INFO
"... APIC ID: %08x (%01x)\n", v
, GET_APIC_ID(v
));
1076 v
= apic_read(APIC_LVR
);
1077 printk(KERN_INFO
"... APIC VERSION: %08x\n", v
);
1078 ver
= GET_APIC_VERSION(v
);
1079 maxlvt
= get_maxlvt();
1081 v
= apic_read(APIC_TASKPRI
);
1082 printk(KERN_DEBUG
"... APIC TASKPRI: %08x (%02x)\n", v
, v
& APIC_TPRI_MASK
);
1084 v
= apic_read(APIC_ARBPRI
);
1085 printk(KERN_DEBUG
"... APIC ARBPRI: %08x (%02x)\n", v
,
1086 v
& APIC_ARBPRI_MASK
);
1087 v
= apic_read(APIC_PROCPRI
);
1088 printk(KERN_DEBUG
"... APIC PROCPRI: %08x\n", v
);
1090 v
= apic_read(APIC_EOI
);
1091 printk(KERN_DEBUG
"... APIC EOI: %08x\n", v
);
1092 v
= apic_read(APIC_RRR
);
1093 printk(KERN_DEBUG
"... APIC RRR: %08x\n", v
);
1094 v
= apic_read(APIC_LDR
);
1095 printk(KERN_DEBUG
"... APIC LDR: %08x\n", v
);
1096 v
= apic_read(APIC_DFR
);
1097 printk(KERN_DEBUG
"... APIC DFR: %08x\n", v
);
1098 v
= apic_read(APIC_SPIV
);
1099 printk(KERN_DEBUG
"... APIC SPIV: %08x\n", v
);
1101 printk(KERN_DEBUG
"... APIC ISR field:\n");
1102 print_APIC_bitfield(APIC_ISR
);
1103 printk(KERN_DEBUG
"... APIC TMR field:\n");
1104 print_APIC_bitfield(APIC_TMR
);
1105 printk(KERN_DEBUG
"... APIC IRR field:\n");
1106 print_APIC_bitfield(APIC_IRR
);
1108 v
= apic_read(APIC_ESR
);
1109 printk(KERN_DEBUG
"... APIC ESR: %08x\n", v
);
1111 v
= apic_read(APIC_ICR
);
1112 printk(KERN_DEBUG
"... APIC ICR: %08x\n", v
);
1113 v
= apic_read(APIC_ICR2
);
1114 printk(KERN_DEBUG
"... APIC ICR2: %08x\n", v
);
1116 v
= apic_read(APIC_LVTT
);
1117 printk(KERN_DEBUG
"... APIC LVTT: %08x\n", v
);
1119 if (maxlvt
> 3) { /* PC is LVT#4. */
1120 v
= apic_read(APIC_LVTPC
);
1121 printk(KERN_DEBUG
"... APIC LVTPC: %08x\n", v
);
1123 v
= apic_read(APIC_LVT0
);
1124 printk(KERN_DEBUG
"... APIC LVT0: %08x\n", v
);
1125 v
= apic_read(APIC_LVT1
);
1126 printk(KERN_DEBUG
"... APIC LVT1: %08x\n", v
);
1128 if (maxlvt
> 2) { /* ERR is LVT#3. */
1129 v
= apic_read(APIC_LVTERR
);
1130 printk(KERN_DEBUG
"... APIC LVTERR: %08x\n", v
);
1133 v
= apic_read(APIC_TMICT
);
1134 printk(KERN_DEBUG
"... APIC TMICT: %08x\n", v
);
1135 v
= apic_read(APIC_TMCCT
);
1136 printk(KERN_DEBUG
"... APIC TMCCT: %08x\n", v
);
1137 v
= apic_read(APIC_TDCR
);
1138 printk(KERN_DEBUG
"... APIC TDCR: %08x\n", v
);
1142 void print_all_local_APICs (void)
1144 on_each_cpu(print_local_APIC
, NULL
, 1, 1);
1147 void __apicdebuginit
print_PIC(void)
1150 unsigned long flags
;
1152 if (apic_verbosity
== APIC_QUIET
)
1155 printk(KERN_DEBUG
"\nprinting PIC contents\n");
1157 spin_lock_irqsave(&i8259A_lock
, flags
);
1159 v
= inb(0xa1) << 8 | inb(0x21);
1160 printk(KERN_DEBUG
"... PIC IMR: %04x\n", v
);
1162 v
= inb(0xa0) << 8 | inb(0x20);
1163 printk(KERN_DEBUG
"... PIC IRR: %04x\n", v
);
1167 v
= inb(0xa0) << 8 | inb(0x20);
1171 spin_unlock_irqrestore(&i8259A_lock
, flags
);
1173 printk(KERN_DEBUG
"... PIC ISR: %04x\n", v
);
1175 v
= inb(0x4d1) << 8 | inb(0x4d0);
1176 printk(KERN_DEBUG
"... PIC ELCR: %04x\n", v
);
1181 static void __init
enable_IO_APIC(void)
1183 union IO_APIC_reg_01 reg_01
;
1184 int i8259_apic
, i8259_pin
;
1186 unsigned long flags
;
1188 for (i
= 0; i
< PIN_MAP_SIZE
; i
++) {
1189 irq_2_pin
[i
].pin
= -1;
1190 irq_2_pin
[i
].next
= 0;
1194 * The number of IO-APIC IRQ registers (== #pins):
1196 for (apic
= 0; apic
< nr_ioapics
; apic
++) {
1197 spin_lock_irqsave(&ioapic_lock
, flags
);
1198 reg_01
.raw
= io_apic_read(apic
, 1);
1199 spin_unlock_irqrestore(&ioapic_lock
, flags
);
1200 nr_ioapic_registers
[apic
] = reg_01
.bits
.entries
+1;
1202 for(apic
= 0; apic
< nr_ioapics
; apic
++) {
1204 /* See if any of the pins is in ExtINT mode */
1205 for (pin
= 0; pin
< nr_ioapic_registers
[apic
]; pin
++) {
1206 struct IO_APIC_route_entry entry
;
1207 entry
= ioapic_read_entry(apic
, pin
);
1209 /* If the interrupt line is enabled and in ExtInt mode
1210 * I have found the pin where the i8259 is connected.
1212 if ((entry
.mask
== 0) && (entry
.delivery_mode
== dest_ExtINT
)) {
1213 ioapic_i8259
.apic
= apic
;
1214 ioapic_i8259
.pin
= pin
;
1220 /* Look to see what if the MP table has reported the ExtINT */
1221 i8259_pin
= find_isa_irq_pin(0, mp_ExtINT
);
1222 i8259_apic
= find_isa_irq_apic(0, mp_ExtINT
);
1223 /* Trust the MP table if nothing is setup in the hardware */
1224 if ((ioapic_i8259
.pin
== -1) && (i8259_pin
>= 0)) {
1225 printk(KERN_WARNING
"ExtINT not setup in hardware but reported by MP table\n");
1226 ioapic_i8259
.pin
= i8259_pin
;
1227 ioapic_i8259
.apic
= i8259_apic
;
1229 /* Complain if the MP table and the hardware disagree */
1230 if (((ioapic_i8259
.apic
!= i8259_apic
) || (ioapic_i8259
.pin
!= i8259_pin
)) &&
1231 (i8259_pin
>= 0) && (ioapic_i8259
.pin
>= 0))
1233 printk(KERN_WARNING
"ExtINT in hardware and MP table differ\n");
1237 * Do not trust the IO-APIC being empty at bootup
1243 * Not an __init, needed by the reboot code
1245 void disable_IO_APIC(void)
1248 * Clear the IO-APIC before rebooting:
1253 * If the i8259 is routed through an IOAPIC
1254 * Put that IOAPIC in virtual wire mode
1255 * so legacy interrupts can be delivered.
1257 if (ioapic_i8259
.pin
!= -1) {
1258 struct IO_APIC_route_entry entry
;
1260 memset(&entry
, 0, sizeof(entry
));
1261 entry
.mask
= 0; /* Enabled */
1262 entry
.trigger
= 0; /* Edge */
1264 entry
.polarity
= 0; /* High */
1265 entry
.delivery_status
= 0;
1266 entry
.dest_mode
= 0; /* Physical */
1267 entry
.delivery_mode
= dest_ExtINT
; /* ExtInt */
1269 entry
.dest
.physical
.physical_dest
=
1270 GET_APIC_ID(apic_read(APIC_ID
));
1273 * Add it to the IO-APIC irq-routing table:
1275 ioapic_write_entry(ioapic_i8259
.apic
, ioapic_i8259
.pin
, entry
);
1278 disconnect_bsp_APIC(ioapic_i8259
.pin
!= -1);
1282 * There is a nasty bug in some older SMP boards, their mptable lies
1283 * about the timer IRQ. We do the following to work around the situation:
1285 * - timer IRQ defaults to IO-APIC IRQ
1286 * - if this function detects that timer IRQs are defunct, then we fall
1287 * back to ISA timer IRQs
1289 static int __init
timer_irq_works(void)
1291 unsigned long t1
= jiffies
;
1294 /* Let ten ticks pass... */
1295 mdelay((10 * 1000) / HZ
);
1298 * Expect a few ticks at least, to be sure some possible
1299 * glue logic does not lock up after one or two first
1300 * ticks in a non-ExtINT mode. Also the local APIC
1301 * might have cached one ExtINT interrupt. Finally, at
1302 * least one tick may be lost due to delays.
1306 if (jiffies
- t1
> 4)
1312 * In the SMP+IOAPIC case it might happen that there are an unspecified
1313 * number of pending IRQ events unhandled. These cases are very rare,
1314 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
1315 * better to do it this way as thus we do not have to be aware of
1316 * 'pending' interrupts in the IRQ path, except at this point.
1319 * Edge triggered needs to resend any interrupt
1320 * that was delayed but this is now handled in the device
1325 * Starting up a edge-triggered IO-APIC interrupt is
1326 * nasty - we need to make sure that we get the edge.
1327 * If it is already asserted for some reason, we need
1328 * return 1 to indicate that is was pending.
1330 * This is not complete - we should be able to fake
1331 * an edge even if it isn't on the 8259A...
1334 static unsigned int startup_ioapic_irq(unsigned int irq
)
1336 int was_pending
= 0;
1337 unsigned long flags
;
1339 spin_lock_irqsave(&ioapic_lock
, flags
);
1341 disable_8259A_irq(irq
);
1342 if (i8259A_irq_pending(irq
))
1345 __unmask_IO_APIC_irq(irq
);
1346 spin_unlock_irqrestore(&ioapic_lock
, flags
);
1351 static int ioapic_retrigger_irq(unsigned int irq
)
1355 unsigned long flags
;
1357 spin_lock_irqsave(&vector_lock
, flags
);
1358 vector
= irq_vector
[irq
];
1360 cpu_set(first_cpu(irq_domain
[irq
]), mask
);
1362 send_IPI_mask(mask
, vector
);
1363 spin_unlock_irqrestore(&vector_lock
, flags
);
1369 * Level and edge triggered IO-APIC interrupts need different handling,
1370 * so we use two separate IRQ descriptors. Edge triggered IRQs can be
1371 * handled with the level-triggered descriptor, but that one has slightly
1372 * more overhead. Level-triggered interrupts cannot be handled with the
1373 * edge-triggered handler, without risking IRQ storms and other ugly
1377 static void ack_apic_edge(unsigned int irq
)
1379 move_native_irq(irq
);
1383 static void ack_apic_level(unsigned int irq
)
1385 int do_unmask_irq
= 0;
1387 #if defined(CONFIG_GENERIC_PENDING_IRQ) || defined(CONFIG_IRQBALANCE)
1388 /* If we are moving the irq we need to mask it */
1389 if (unlikely(irq_desc
[irq
].status
& IRQ_MOVE_PENDING
)) {
1391 mask_IO_APIC_irq(irq
);
1396 * We must acknowledge the irq before we move it or the acknowledge will
1397 * not propogate properly.
1401 /* Now we can move and renable the irq */
1402 move_masked_irq(irq
);
1403 if (unlikely(do_unmask_irq
))
1404 unmask_IO_APIC_irq(irq
);
1407 static struct irq_chip ioapic_chip __read_mostly
= {
1409 .startup
= startup_ioapic_irq
,
1410 .mask
= mask_IO_APIC_irq
,
1411 .unmask
= unmask_IO_APIC_irq
,
1412 .ack
= ack_apic_edge
,
1413 .eoi
= ack_apic_level
,
1415 .set_affinity
= set_ioapic_affinity_irq
,
1417 .retrigger
= ioapic_retrigger_irq
,
1420 static inline void init_IO_APIC_traps(void)
1425 * NOTE! The local APIC isn't very good at handling
1426 * multiple interrupts at the same interrupt level.
1427 * As the interrupt level is determined by taking the
1428 * vector number and shifting that right by 4, we
1429 * want to spread these out a bit so that they don't
1430 * all fall in the same interrupt level.
1432 * Also, we've got to be careful not to trash gate
1433 * 0x80, because int 0x80 is hm, kind of importantish. ;)
1435 for (irq
= 0; irq
< NR_IRQS
; irq
++) {
1437 if (IO_APIC_IRQ(tmp
) && !irq_vector
[tmp
]) {
1439 * Hmm.. We don't have an entry for this,
1440 * so default to an old-fashioned 8259
1441 * interrupt if we can..
1444 make_8259A_irq(irq
);
1446 /* Strange. Oh, well.. */
1447 irq_desc
[irq
].chip
= &no_irq_chip
;
1452 static void enable_lapic_irq (unsigned int irq
)
1456 v
= apic_read(APIC_LVT0
);
1457 apic_write(APIC_LVT0
, v
& ~APIC_LVT_MASKED
);
1460 static void disable_lapic_irq (unsigned int irq
)
1464 v
= apic_read(APIC_LVT0
);
1465 apic_write(APIC_LVT0
, v
| APIC_LVT_MASKED
);
1468 static void ack_lapic_irq (unsigned int irq
)
1473 static void end_lapic_irq (unsigned int i
) { /* nothing */ }
1475 static struct hw_interrupt_type lapic_irq_type __read_mostly
= {
1476 .typename
= "local-APIC-edge",
1477 .startup
= NULL
, /* startup_irq() not used for IRQ0 */
1478 .shutdown
= NULL
, /* shutdown_irq() not used for IRQ0 */
1479 .enable
= enable_lapic_irq
,
1480 .disable
= disable_lapic_irq
,
1481 .ack
= ack_lapic_irq
,
1482 .end
= end_lapic_irq
,
1485 static void setup_nmi (void)
1488 * Dirty trick to enable the NMI watchdog ...
1489 * We put the 8259A master into AEOI mode and
1490 * unmask on all local APICs LVT0 as NMI.
1492 * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
1493 * is from Maciej W. Rozycki - so we do not have to EOI from
1494 * the NMI handler or the timer interrupt.
1496 printk(KERN_INFO
"activating NMI Watchdog ...");
1498 enable_NMI_through_LVT0(NULL
);
1504 * This looks a bit hackish but it's about the only one way of sending
1505 * a few INTA cycles to 8259As and any associated glue logic. ICR does
1506 * not support the ExtINT mode, unfortunately. We need to send these
1507 * cycles as some i82489DX-based boards have glue logic that keeps the
1508 * 8259A interrupt line asserted until INTA. --macro
1510 static inline void unlock_ExtINT_logic(void)
1513 struct IO_APIC_route_entry entry0
, entry1
;
1514 unsigned char save_control
, save_freq_select
;
1515 unsigned long flags
;
1517 pin
= find_isa_irq_pin(8, mp_INT
);
1518 apic
= find_isa_irq_apic(8, mp_INT
);
1522 spin_lock_irqsave(&ioapic_lock
, flags
);
1523 *(((int *)&entry0
) + 1) = io_apic_read(apic
, 0x11 + 2 * pin
);
1524 *(((int *)&entry0
) + 0) = io_apic_read(apic
, 0x10 + 2 * pin
);
1525 spin_unlock_irqrestore(&ioapic_lock
, flags
);
1526 clear_IO_APIC_pin(apic
, pin
);
1528 memset(&entry1
, 0, sizeof(entry1
));
1530 entry1
.dest_mode
= 0; /* physical delivery */
1531 entry1
.mask
= 0; /* unmask IRQ now */
1532 entry1
.dest
.physical
.physical_dest
= hard_smp_processor_id();
1533 entry1
.delivery_mode
= dest_ExtINT
;
1534 entry1
.polarity
= entry0
.polarity
;
1538 spin_lock_irqsave(&ioapic_lock
, flags
);
1539 io_apic_write(apic
, 0x11 + 2 * pin
, *(((int *)&entry1
) + 1));
1540 io_apic_write(apic
, 0x10 + 2 * pin
, *(((int *)&entry1
) + 0));
1541 spin_unlock_irqrestore(&ioapic_lock
, flags
);
1543 save_control
= CMOS_READ(RTC_CONTROL
);
1544 save_freq_select
= CMOS_READ(RTC_FREQ_SELECT
);
1545 CMOS_WRITE((save_freq_select
& ~RTC_RATE_SELECT
) | 0x6,
1547 CMOS_WRITE(save_control
| RTC_PIE
, RTC_CONTROL
);
1552 if ((CMOS_READ(RTC_INTR_FLAGS
) & RTC_PF
) == RTC_PF
)
1556 CMOS_WRITE(save_control
, RTC_CONTROL
);
1557 CMOS_WRITE(save_freq_select
, RTC_FREQ_SELECT
);
1558 clear_IO_APIC_pin(apic
, pin
);
1560 spin_lock_irqsave(&ioapic_lock
, flags
);
1561 io_apic_write(apic
, 0x11 + 2 * pin
, *(((int *)&entry0
) + 1));
1562 io_apic_write(apic
, 0x10 + 2 * pin
, *(((int *)&entry0
) + 0));
1563 spin_unlock_irqrestore(&ioapic_lock
, flags
);
1567 * This code may look a bit paranoid, but it's supposed to cooperate with
1568 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ
1569 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast
1570 * fanatically on his truly buggy board.
1573 static int try_apic_pin(int apic
, int pin
, char *msg
)
1575 apic_printk(APIC_VERBOSE
, KERN_INFO
1576 "..TIMER: trying IO-APIC=%d PIN=%d %s",
1580 * Ok, does IRQ0 through the IOAPIC work?
1582 if (!no_timer_check
&& timer_irq_works()) {
1583 nmi_watchdog_default();
1584 if (nmi_watchdog
== NMI_IO_APIC
) {
1585 disable_8259A_irq(0);
1587 enable_8259A_irq(0);
1591 clear_IO_APIC_pin(apic
, pin
);
1592 apic_printk(APIC_QUIET
, KERN_ERR
" .. failed\n");
1596 /* The function from hell */
1597 static void check_timer(void)
1599 int apic1
, pin1
, apic2
, pin2
;
1604 * get/set the timer IRQ vector:
1606 disable_8259A_irq(0);
1607 vector
= assign_irq_vector(0, TARGET_CPUS
, &mask
);
1610 * Subtle, code in do_timer_interrupt() expects an AEOI
1611 * mode for the 8259A whenever interrupts are routed
1612 * through I/O APICs. Also IRQ0 has to be enabled in
1613 * the 8259A which implies the virtual wire has to be
1614 * disabled in the local APIC.
1616 apic_write(APIC_LVT0
, APIC_LVT_MASKED
| APIC_DM_EXTINT
);
1619 pin1
= find_isa_irq_pin(0, mp_INT
);
1620 apic1
= find_isa_irq_apic(0, mp_INT
);
1621 pin2
= ioapic_i8259
.pin
;
1622 apic2
= ioapic_i8259
.apic
;
1624 /* Do this first, otherwise we get double interrupts on ATI boards */
1625 if ((pin1
!= -1) && try_apic_pin(apic1
, pin1
,"with 8259 IRQ0 disabled"))
1628 /* Now try again with IRQ0 8259A enabled.
1629 Assumes timer is on IO-APIC 0 ?!? */
1630 enable_8259A_irq(0);
1631 unmask_IO_APIC_irq(0);
1632 if (try_apic_pin(apic1
, pin1
, "with 8259 IRQ0 enabled"))
1634 disable_8259A_irq(0);
1636 /* Always try pin0 and pin2 on APIC 0 to handle buggy timer overrides
1638 if (!(apic1
== 0 && pin1
== 0) &&
1639 try_apic_pin(0, 0, "fallback with 8259 IRQ0 disabled"))
1641 if (!(apic1
== 0 && pin1
== 2) &&
1642 try_apic_pin(0, 2, "fallback with 8259 IRQ0 disabled"))
1645 /* Then try pure 8259A routing on the 8259 as reported by BIOS*/
1646 enable_8259A_irq(0);
1648 setup_ExtINT_IRQ0_pin(apic2
, pin2
, vector
);
1649 if (try_apic_pin(apic2
,pin2
,"8259A broadcast ExtINT from BIOS"))
1653 /* Tried all possibilities to go through the IO-APIC. Now come the
1654 really cheesy fallbacks. */
1656 if (nmi_watchdog
== NMI_IO_APIC
) {
1657 printk(KERN_WARNING
"timer doesn't work through the IO-APIC - disabling NMI Watchdog!\n");
1661 apic_printk(APIC_VERBOSE
, KERN_INFO
"...trying to set up timer as Virtual Wire IRQ...");
1663 disable_8259A_irq(0);
1664 irq_desc
[0].chip
= &lapic_irq_type
;
1665 apic_write(APIC_LVT0
, APIC_DM_FIXED
| vector
); /* Fixed mode */
1666 enable_8259A_irq(0);
1668 if (timer_irq_works()) {
1669 apic_printk(APIC_VERBOSE
," works.\n");
1672 apic_write(APIC_LVT0
, APIC_LVT_MASKED
| APIC_DM_FIXED
| vector
);
1673 apic_printk(APIC_VERBOSE
," failed.\n");
1675 apic_printk(APIC_VERBOSE
, KERN_INFO
"...trying to set up timer as ExtINT IRQ...");
1679 apic_write(APIC_LVT0
, APIC_DM_EXTINT
);
1681 unlock_ExtINT_logic();
1683 if (timer_irq_works()) {
1684 apic_printk(APIC_VERBOSE
," works.\n");
1687 apic_printk(APIC_VERBOSE
," failed :(.\n");
1688 panic("IO-APIC + timer doesn't work! Try using the 'noapic' kernel parameter\n");
1691 static int __init
notimercheck(char *s
)
1696 __setup("no_timer_check", notimercheck
);
1700 * IRQ's that are handled by the PIC in the MPS IOAPIC case.
1701 * - IRQ2 is the cascade IRQ, and cannot be a io-apic IRQ.
1702 * Linux doesn't really care, as it's not actually used
1703 * for any interrupt handling anyway.
1705 #define PIC_IRQS (1<<2)
1707 void __init
setup_IO_APIC(void)
1712 io_apic_irqs
= ~0; /* all IRQs go through IOAPIC */
1714 io_apic_irqs
= ~PIC_IRQS
;
1716 apic_printk(APIC_VERBOSE
, "ENABLING IO-APIC IRQs\n");
1719 setup_IO_APIC_irqs();
1720 init_IO_APIC_traps();
1726 struct sysfs_ioapic_data
{
1727 struct sys_device dev
;
1728 struct IO_APIC_route_entry entry
[0];
1730 static struct sysfs_ioapic_data
* mp_ioapic_data
[MAX_IO_APICS
];
1732 static int ioapic_suspend(struct sys_device
*dev
, pm_message_t state
)
1734 struct IO_APIC_route_entry
*entry
;
1735 struct sysfs_ioapic_data
*data
;
1738 data
= container_of(dev
, struct sysfs_ioapic_data
, dev
);
1739 entry
= data
->entry
;
1740 for (i
= 0; i
< nr_ioapic_registers
[dev
->id
]; i
++, entry
++ )
1741 *entry
= ioapic_read_entry(dev
->id
, i
);
1746 static int ioapic_resume(struct sys_device
*dev
)
1748 struct IO_APIC_route_entry
*entry
;
1749 struct sysfs_ioapic_data
*data
;
1750 unsigned long flags
;
1751 union IO_APIC_reg_00 reg_00
;
1754 data
= container_of(dev
, struct sysfs_ioapic_data
, dev
);
1755 entry
= data
->entry
;
1757 spin_lock_irqsave(&ioapic_lock
, flags
);
1758 reg_00
.raw
= io_apic_read(dev
->id
, 0);
1759 if (reg_00
.bits
.ID
!= mp_ioapics
[dev
->id
].mpc_apicid
) {
1760 reg_00
.bits
.ID
= mp_ioapics
[dev
->id
].mpc_apicid
;
1761 io_apic_write(dev
->id
, 0, reg_00
.raw
);
1763 spin_unlock_irqrestore(&ioapic_lock
, flags
);
1764 for (i
= 0; i
< nr_ioapic_registers
[dev
->id
]; i
++)
1765 ioapic_write_entry(dev
->id
, i
, entry
[i
]);
1770 static struct sysdev_class ioapic_sysdev_class
= {
1771 set_kset_name("ioapic"),
1772 .suspend
= ioapic_suspend
,
1773 .resume
= ioapic_resume
,
1776 static int __init
ioapic_init_sysfs(void)
1778 struct sys_device
* dev
;
1779 int i
, size
, error
= 0;
1781 error
= sysdev_class_register(&ioapic_sysdev_class
);
1785 for (i
= 0; i
< nr_ioapics
; i
++ ) {
1786 size
= sizeof(struct sys_device
) + nr_ioapic_registers
[i
]
1787 * sizeof(struct IO_APIC_route_entry
);
1788 mp_ioapic_data
[i
] = kmalloc(size
, GFP_KERNEL
);
1789 if (!mp_ioapic_data
[i
]) {
1790 printk(KERN_ERR
"Can't suspend/resume IOAPIC %d\n", i
);
1793 memset(mp_ioapic_data
[i
], 0, size
);
1794 dev
= &mp_ioapic_data
[i
]->dev
;
1796 dev
->cls
= &ioapic_sysdev_class
;
1797 error
= sysdev_register(dev
);
1799 kfree(mp_ioapic_data
[i
]);
1800 mp_ioapic_data
[i
] = NULL
;
1801 printk(KERN_ERR
"Can't suspend/resume IOAPIC %d\n", i
);
1809 device_initcall(ioapic_init_sysfs
);
1812 * Dynamic irq allocate and deallocation
1814 int create_irq(void)
1816 /* Allocate an unused irq */
1820 unsigned long flags
;
1824 spin_lock_irqsave(&vector_lock
, flags
);
1825 for (new = (NR_IRQS
- 1); new >= 0; new--) {
1826 if (platform_legacy_irq(new))
1828 if (irq_vector
[new] != 0)
1830 vector
= __assign_irq_vector(new, TARGET_CPUS
, &mask
);
1831 if (likely(vector
> 0))
1835 spin_unlock_irqrestore(&vector_lock
, flags
);
1838 dynamic_irq_init(irq
);
1843 void destroy_irq(unsigned int irq
)
1845 unsigned long flags
;
1847 dynamic_irq_cleanup(irq
);
1849 spin_lock_irqsave(&vector_lock
, flags
);
1850 __clear_irq_vector(irq
);
1851 spin_unlock_irqrestore(&vector_lock
, flags
);
1855 * MSI mesage composition
1857 #ifdef CONFIG_PCI_MSI
1858 static int msi_compose_msg(struct pci_dev
*pdev
, unsigned int irq
, struct msi_msg
*msg
)
1864 vector
= assign_irq_vector(irq
, TARGET_CPUS
, &tmp
);
1866 dest
= cpu_mask_to_apicid(tmp
);
1868 msg
->address_hi
= MSI_ADDR_BASE_HI
;
1871 ((INT_DEST_MODE
== 0) ?
1872 MSI_ADDR_DEST_MODE_PHYSICAL
:
1873 MSI_ADDR_DEST_MODE_LOGICAL
) |
1874 ((INT_DELIVERY_MODE
!= dest_LowestPrio
) ?
1875 MSI_ADDR_REDIRECTION_CPU
:
1876 MSI_ADDR_REDIRECTION_LOWPRI
) |
1877 MSI_ADDR_DEST_ID(dest
);
1880 MSI_DATA_TRIGGER_EDGE
|
1881 MSI_DATA_LEVEL_ASSERT
|
1882 ((INT_DELIVERY_MODE
!= dest_LowestPrio
) ?
1883 MSI_DATA_DELIVERY_FIXED
:
1884 MSI_DATA_DELIVERY_LOWPRI
) |
1885 MSI_DATA_VECTOR(vector
);
1891 static void set_msi_irq_affinity(unsigned int irq
, cpumask_t mask
)
1898 cpus_and(tmp
, mask
, cpu_online_map
);
1899 if (cpus_empty(tmp
))
1902 cpus_and(mask
, tmp
, CPU_MASK_ALL
);
1904 vector
= assign_irq_vector(irq
, mask
, &tmp
);
1908 dest
= cpu_mask_to_apicid(tmp
);
1910 read_msi_msg(irq
, &msg
);
1912 msg
.data
&= ~MSI_DATA_VECTOR_MASK
;
1913 msg
.data
|= MSI_DATA_VECTOR(vector
);
1914 msg
.address_lo
&= ~MSI_ADDR_DEST_ID_MASK
;
1915 msg
.address_lo
|= MSI_ADDR_DEST_ID(dest
);
1917 write_msi_msg(irq
, &msg
);
1918 set_native_irq_info(irq
, mask
);
1920 #endif /* CONFIG_SMP */
1923 * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
1924 * which implement the MSI or MSI-X Capability Structure.
1926 static struct irq_chip msi_chip
= {
1928 .unmask
= unmask_msi_irq
,
1929 .mask
= mask_msi_irq
,
1930 .ack
= ack_apic_edge
,
1932 .set_affinity
= set_msi_irq_affinity
,
1934 .retrigger
= ioapic_retrigger_irq
,
1937 int arch_setup_msi_irq(unsigned int irq
, struct pci_dev
*dev
)
1941 ret
= msi_compose_msg(dev
, irq
, &msg
);
1945 write_msi_msg(irq
, &msg
);
1947 set_irq_chip_and_handler_name(irq
, &msi_chip
, handle_edge_irq
, "edge");
1952 void arch_teardown_msi_irq(unsigned int irq
)
1957 #endif /* CONFIG_PCI_MSI */
1960 * Hypertransport interrupt support
1962 #ifdef CONFIG_HT_IRQ
1966 static void target_ht_irq(unsigned int irq
, unsigned int dest
, u8 vector
)
1968 struct ht_irq_msg msg
;
1969 fetch_ht_irq_msg(irq
, &msg
);
1971 msg
.address_lo
&= ~(HT_IRQ_LOW_VECTOR_MASK
| HT_IRQ_LOW_DEST_ID_MASK
);
1972 msg
.address_hi
&= ~(HT_IRQ_HIGH_DEST_ID_MASK
);
1974 msg
.address_lo
|= HT_IRQ_LOW_VECTOR(vector
) | HT_IRQ_LOW_DEST_ID(dest
);
1975 msg
.address_hi
|= HT_IRQ_HIGH_DEST_ID(dest
);
1977 write_ht_irq_msg(irq
, &msg
);
1980 static void set_ht_irq_affinity(unsigned int irq
, cpumask_t mask
)
1986 cpus_and(tmp
, mask
, cpu_online_map
);
1987 if (cpus_empty(tmp
))
1990 cpus_and(mask
, tmp
, CPU_MASK_ALL
);
1992 vector
= assign_irq_vector(irq
, mask
, &tmp
);
1996 dest
= cpu_mask_to_apicid(tmp
);
1998 target_ht_irq(irq
, dest
, vector
);
1999 set_native_irq_info(irq
, mask
);
2003 static struct irq_chip ht_irq_chip
= {
2005 .mask
= mask_ht_irq
,
2006 .unmask
= unmask_ht_irq
,
2007 .ack
= ack_apic_edge
,
2009 .set_affinity
= set_ht_irq_affinity
,
2011 .retrigger
= ioapic_retrigger_irq
,
2014 int arch_setup_ht_irq(unsigned int irq
, struct pci_dev
*dev
)
2019 vector
= assign_irq_vector(irq
, TARGET_CPUS
, &tmp
);
2021 struct ht_irq_msg msg
;
2024 dest
= cpu_mask_to_apicid(tmp
);
2026 msg
.address_hi
= HT_IRQ_HIGH_DEST_ID(dest
);
2030 HT_IRQ_LOW_DEST_ID(dest
) |
2031 HT_IRQ_LOW_VECTOR(vector
) |
2032 ((INT_DEST_MODE
== 0) ?
2033 HT_IRQ_LOW_DM_PHYSICAL
:
2034 HT_IRQ_LOW_DM_LOGICAL
) |
2035 HT_IRQ_LOW_RQEOI_EDGE
|
2036 ((INT_DELIVERY_MODE
!= dest_LowestPrio
) ?
2037 HT_IRQ_LOW_MT_FIXED
:
2038 HT_IRQ_LOW_MT_ARBITRATED
) |
2039 HT_IRQ_LOW_IRQ_MASKED
;
2041 write_ht_irq_msg(irq
, &msg
);
2043 set_irq_chip_and_handler_name(irq
, &ht_irq_chip
,
2044 handle_edge_irq
, "edge");
2048 #endif /* CONFIG_HT_IRQ */
2050 /* --------------------------------------------------------------------------
2051 ACPI-based IOAPIC Configuration
2052 -------------------------------------------------------------------------- */
2056 #define IO_APIC_MAX_ID 0xFE
2058 int __init
io_apic_get_redir_entries (int ioapic
)
2060 union IO_APIC_reg_01 reg_01
;
2061 unsigned long flags
;
2063 spin_lock_irqsave(&ioapic_lock
, flags
);
2064 reg_01
.raw
= io_apic_read(ioapic
, 1);
2065 spin_unlock_irqrestore(&ioapic_lock
, flags
);
2067 return reg_01
.bits
.entries
;
2071 int io_apic_set_pci_routing (int ioapic
, int pin
, int irq
, int triggering
, int polarity
)
2073 struct IO_APIC_route_entry entry
;
2074 unsigned long flags
;
2078 if (!IO_APIC_IRQ(irq
)) {
2079 apic_printk(APIC_QUIET
,KERN_ERR
"IOAPIC[%d]: Invalid reference to IRQ 0\n",
2085 * IRQs < 16 are already in the irq_2_pin[] map
2088 add_pin_to_irq(irq
, ioapic
, pin
);
2091 vector
= assign_irq_vector(irq
, TARGET_CPUS
, &mask
);
2096 * Generate a PCI IRQ routing entry and program the IOAPIC accordingly.
2097 * Note that we mask (disable) IRQs now -- these get enabled when the
2098 * corresponding device driver registers for this IRQ.
2101 memset(&entry
,0,sizeof(entry
));
2103 entry
.delivery_mode
= INT_DELIVERY_MODE
;
2104 entry
.dest_mode
= INT_DEST_MODE
;
2105 entry
.dest
.logical
.logical_dest
= cpu_mask_to_apicid(mask
);
2106 entry
.trigger
= triggering
;
2107 entry
.polarity
= polarity
;
2108 entry
.mask
= 1; /* Disabled (masked) */
2109 entry
.vector
= vector
& 0xff;
2111 apic_printk(APIC_VERBOSE
,KERN_DEBUG
"IOAPIC[%d]: Set PCI routing entry (%d-%d -> 0x%x -> "
2112 "IRQ %d Mode:%i Active:%i)\n", ioapic
,
2113 mp_ioapics
[ioapic
].mpc_apicid
, pin
, entry
.vector
, irq
,
2114 triggering
, polarity
);
2116 ioapic_register_intr(irq
, entry
.vector
, triggering
);
2118 if (!ioapic
&& (irq
< 16))
2119 disable_8259A_irq(irq
);
2121 ioapic_write_entry(ioapic
, pin
, entry
);
2123 spin_lock_irqsave(&ioapic_lock
, flags
);
2124 set_native_irq_info(irq
, TARGET_CPUS
);
2125 spin_unlock_irqrestore(&ioapic_lock
, flags
);
2130 #endif /* CONFIG_ACPI */
2134 * This function currently is only a helper for the i386 smp boot process where
2135 * we need to reprogram the ioredtbls to cater for the cpus which have come online
2136 * so mask in all cases should simply be TARGET_CPUS
2139 void __init
setup_ioapic_dest(void)
2141 int pin
, ioapic
, irq
, irq_entry
;
2143 if (skip_ioapic_setup
== 1)
2146 for (ioapic
= 0; ioapic
< nr_ioapics
; ioapic
++) {
2147 for (pin
= 0; pin
< nr_ioapic_registers
[ioapic
]; pin
++) {
2148 irq_entry
= find_irq_entry(ioapic
, pin
, mp_INT
);
2149 if (irq_entry
== -1)
2151 irq
= pin_2_irq(irq_entry
, ioapic
, pin
);
2153 /* setup_IO_APIC_irqs could fail to get vector for some device
2154 * when you have too many devices, because at that time only boot
2157 if(!irq_vector
[irq
])
2158 setup_IO_APIC_irq(ioapic
, pin
, irq_entry
, irq
);
2160 set_ioapic_affinity_irq(irq
, TARGET_CPUS
);