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>
34 #include <acpi/acpi_bus.h>
40 #include <asm/proto.h>
41 #include <asm/mach_apic.h>
45 #include <asm/msidef.h>
47 #define __apicdebuginit __init
49 int sis_apic_bug
; /* not actually supported, dummy for compile */
51 static int no_timer_check
;
53 static int disable_timer_pin_1 __initdata
;
55 int timer_over_8254 __initdata
= 0;
57 /* Where if anywhere is the i8259 connect in external int mode */
58 static struct { int pin
, apic
; } ioapic_i8259
= { -1, -1 };
60 static DEFINE_SPINLOCK(ioapic_lock
);
61 static DEFINE_SPINLOCK(vector_lock
);
64 * # of IRQ routing registers
66 int nr_ioapic_registers
[MAX_IO_APICS
];
69 * Rough estimation of how many shared IRQs there are, can
72 #define MAX_PLUS_SHARED_IRQS NR_IRQ_VECTORS
73 #define PIN_MAP_SIZE (MAX_PLUS_SHARED_IRQS + NR_IRQS)
76 * This is performance-critical, we want to do it O(1)
78 * the indexing order of this array favors 1:1 mappings
79 * between pins and IRQs.
82 static struct irq_pin_list
{
83 short apic
, pin
, next
;
84 } irq_2_pin
[PIN_MAP_SIZE
];
86 int vector_irq
[NR_VECTORS
] __read_mostly
= { [0 ... NR_VECTORS
- 1] = -1};
88 #define vector_to_irq(vector) \
89 (platform_legacy_irq(vector) ? vector : vector_irq[vector])
91 #define vector_to_irq(vector) (vector)
94 #define __DO_ACTION(R, ACTION, FINAL) \
98 struct irq_pin_list *entry = irq_2_pin + irq; \
100 BUG_ON(irq >= NR_IRQS); \
106 reg = io_apic_read(entry->apic, 0x10 + R + pin*2); \
108 io_apic_modify(entry->apic, reg); \
111 entry = irq_2_pin + entry->next; \
117 struct { u32 w1
, w2
; };
118 struct IO_APIC_route_entry entry
;
121 static struct IO_APIC_route_entry
ioapic_read_entry(int apic
, int pin
)
123 union entry_union eu
;
125 spin_lock_irqsave(&ioapic_lock
, flags
);
126 eu
.w1
= io_apic_read(apic
, 0x10 + 2 * pin
);
127 eu
.w2
= io_apic_read(apic
, 0x11 + 2 * pin
);
128 spin_unlock_irqrestore(&ioapic_lock
, flags
);
132 static void ioapic_write_entry(int apic
, int pin
, struct IO_APIC_route_entry e
)
135 union entry_union eu
;
137 spin_lock_irqsave(&ioapic_lock
, flags
);
138 io_apic_write(apic
, 0x10 + 2*pin
, eu
.w1
);
139 io_apic_write(apic
, 0x11 + 2*pin
, eu
.w2
);
140 spin_unlock_irqrestore(&ioapic_lock
, flags
);
144 static void set_ioapic_affinity_irq(unsigned int irq
, cpumask_t mask
)
150 cpus_and(tmp
, mask
, cpu_online_map
);
154 cpus_and(mask
, tmp
, CPU_MASK_ALL
);
156 dest
= cpu_mask_to_apicid(mask
);
159 * Only the high 8 bits are valid.
161 dest
= SET_APIC_LOGICAL_ID(dest
);
163 spin_lock_irqsave(&ioapic_lock
, flags
);
164 __DO_ACTION(1, = dest
, )
165 set_irq_info(irq
, mask
);
166 spin_unlock_irqrestore(&ioapic_lock
, flags
);
170 static u8 gsi_2_irq
[NR_IRQ_VECTORS
] = { [0 ... NR_IRQ_VECTORS
-1] = 0xFF };
173 * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
174 * shared ISA-space IRQs, so we have to support them. We are super
175 * fast in the common case, and fast for shared ISA-space IRQs.
177 static void add_pin_to_irq(unsigned int irq
, int apic
, int pin
)
179 static int first_free_entry
= NR_IRQS
;
180 struct irq_pin_list
*entry
= irq_2_pin
+ irq
;
182 BUG_ON(irq
>= NR_IRQS
);
184 entry
= irq_2_pin
+ entry
->next
;
186 if (entry
->pin
!= -1) {
187 entry
->next
= first_free_entry
;
188 entry
= irq_2_pin
+ entry
->next
;
189 if (++first_free_entry
>= PIN_MAP_SIZE
)
190 panic("io_apic.c: ran out of irq_2_pin entries!");
197 #define DO_ACTION(name,R,ACTION, FINAL) \
199 static void name##_IO_APIC_irq (unsigned int irq) \
200 __DO_ACTION(R, ACTION, FINAL)
202 DO_ACTION( __mask
, 0, |= 0x00010000, io_apic_sync(entry
->apic
) )
204 DO_ACTION( __unmask
, 0, &= 0xfffeffff, )
207 static void mask_IO_APIC_irq (unsigned int irq
)
211 spin_lock_irqsave(&ioapic_lock
, flags
);
212 __mask_IO_APIC_irq(irq
);
213 spin_unlock_irqrestore(&ioapic_lock
, flags
);
216 static void unmask_IO_APIC_irq (unsigned int irq
)
220 spin_lock_irqsave(&ioapic_lock
, flags
);
221 __unmask_IO_APIC_irq(irq
);
222 spin_unlock_irqrestore(&ioapic_lock
, flags
);
225 static void clear_IO_APIC_pin(unsigned int apic
, unsigned int pin
)
227 struct IO_APIC_route_entry entry
;
229 /* Check delivery_mode to be sure we're not clearing an SMI pin */
230 entry
= ioapic_read_entry(apic
, pin
);
231 if (entry
.delivery_mode
== dest_SMI
)
234 * Disable it in the IO-APIC irq-routing table:
236 memset(&entry
, 0, sizeof(entry
));
238 ioapic_write_entry(apic
, pin
, entry
);
241 static void clear_IO_APIC (void)
245 for (apic
= 0; apic
< nr_ioapics
; apic
++)
246 for (pin
= 0; pin
< nr_ioapic_registers
[apic
]; pin
++)
247 clear_IO_APIC_pin(apic
, pin
);
250 int skip_ioapic_setup
;
253 /* dummy parsing: see setup.c */
255 static int __init
disable_ioapic_setup(char *str
)
257 skip_ioapic_setup
= 1;
260 early_param("noapic", disable_ioapic_setup
);
262 /* Actually the next is obsolete, but keep it for paranoid reasons -AK */
263 static int __init
disable_timer_pin_setup(char *arg
)
265 disable_timer_pin_1
= 1;
268 __setup("disable_timer_pin_1", disable_timer_pin_setup
);
270 static int __init
setup_disable_8254_timer(char *s
)
272 timer_over_8254
= -1;
275 static int __init
setup_enable_8254_timer(char *s
)
281 __setup("disable_8254_timer", setup_disable_8254_timer
);
282 __setup("enable_8254_timer", setup_enable_8254_timer
);
286 * Find the IRQ entry number of a certain pin.
288 static int find_irq_entry(int apic
, int pin
, int type
)
292 for (i
= 0; i
< mp_irq_entries
; i
++)
293 if (mp_irqs
[i
].mpc_irqtype
== type
&&
294 (mp_irqs
[i
].mpc_dstapic
== mp_ioapics
[apic
].mpc_apicid
||
295 mp_irqs
[i
].mpc_dstapic
== MP_APIC_ALL
) &&
296 mp_irqs
[i
].mpc_dstirq
== pin
)
303 * Find the pin to which IRQ[irq] (ISA) is connected
305 static int __init
find_isa_irq_pin(int irq
, int type
)
309 for (i
= 0; i
< mp_irq_entries
; i
++) {
310 int lbus
= mp_irqs
[i
].mpc_srcbus
;
312 if (test_bit(lbus
, mp_bus_not_pci
) &&
313 (mp_irqs
[i
].mpc_irqtype
== type
) &&
314 (mp_irqs
[i
].mpc_srcbusirq
== irq
))
316 return mp_irqs
[i
].mpc_dstirq
;
321 static int __init
find_isa_irq_apic(int irq
, int type
)
325 for (i
= 0; i
< mp_irq_entries
; i
++) {
326 int lbus
= mp_irqs
[i
].mpc_srcbus
;
328 if (test_bit(lbus
, mp_bus_not_pci
) &&
329 (mp_irqs
[i
].mpc_irqtype
== type
) &&
330 (mp_irqs
[i
].mpc_srcbusirq
== irq
))
333 if (i
< mp_irq_entries
) {
335 for(apic
= 0; apic
< nr_ioapics
; apic
++) {
336 if (mp_ioapics
[apic
].mpc_apicid
== mp_irqs
[i
].mpc_dstapic
)
345 * Find a specific PCI IRQ entry.
346 * Not an __init, possibly needed by modules
348 static int pin_2_irq(int idx
, int apic
, int pin
);
350 int IO_APIC_get_PCI_irq_vector(int bus
, int slot
, int pin
)
352 int apic
, i
, best_guess
= -1;
354 apic_printk(APIC_DEBUG
, "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
356 if (mp_bus_id_to_pci_bus
[bus
] == -1) {
357 apic_printk(APIC_VERBOSE
, "PCI BIOS passed nonexistent PCI bus %d!\n", bus
);
360 for (i
= 0; i
< mp_irq_entries
; i
++) {
361 int lbus
= mp_irqs
[i
].mpc_srcbus
;
363 for (apic
= 0; apic
< nr_ioapics
; apic
++)
364 if (mp_ioapics
[apic
].mpc_apicid
== mp_irqs
[i
].mpc_dstapic
||
365 mp_irqs
[i
].mpc_dstapic
== MP_APIC_ALL
)
368 if (!test_bit(lbus
, mp_bus_not_pci
) &&
369 !mp_irqs
[i
].mpc_irqtype
&&
371 (slot
== ((mp_irqs
[i
].mpc_srcbusirq
>> 2) & 0x1f))) {
372 int irq
= pin_2_irq(i
,apic
,mp_irqs
[i
].mpc_dstirq
);
374 if (!(apic
|| IO_APIC_IRQ(irq
)))
377 if (pin
== (mp_irqs
[i
].mpc_srcbusirq
& 3))
380 * Use the first all-but-pin matching entry as a
381 * best-guess fuzzy result for broken mptables.
387 BUG_ON(best_guess
>= NR_IRQS
);
391 /* ISA interrupts are always polarity zero edge triggered,
392 * when listed as conforming in the MP table. */
394 #define default_ISA_trigger(idx) (0)
395 #define default_ISA_polarity(idx) (0)
397 /* PCI interrupts are always polarity one level triggered,
398 * when listed as conforming in the MP table. */
400 #define default_PCI_trigger(idx) (1)
401 #define default_PCI_polarity(idx) (1)
403 static int __init
MPBIOS_polarity(int idx
)
405 int bus
= mp_irqs
[idx
].mpc_srcbus
;
409 * Determine IRQ line polarity (high active or low active):
411 switch (mp_irqs
[idx
].mpc_irqflag
& 3)
413 case 0: /* conforms, ie. bus-type dependent polarity */
414 if (test_bit(bus
, mp_bus_not_pci
))
415 polarity
= default_ISA_polarity(idx
);
417 polarity
= default_PCI_polarity(idx
);
419 case 1: /* high active */
424 case 2: /* reserved */
426 printk(KERN_WARNING
"broken BIOS!!\n");
430 case 3: /* low active */
435 default: /* invalid */
437 printk(KERN_WARNING
"broken BIOS!!\n");
445 static int MPBIOS_trigger(int idx
)
447 int bus
= mp_irqs
[idx
].mpc_srcbus
;
451 * Determine IRQ trigger mode (edge or level sensitive):
453 switch ((mp_irqs
[idx
].mpc_irqflag
>>2) & 3)
455 case 0: /* conforms, ie. bus-type dependent */
456 if (test_bit(bus
, mp_bus_not_pci
))
457 trigger
= default_ISA_trigger(idx
);
459 trigger
= default_PCI_trigger(idx
);
466 case 2: /* reserved */
468 printk(KERN_WARNING
"broken BIOS!!\n");
477 default: /* invalid */
479 printk(KERN_WARNING
"broken BIOS!!\n");
487 static inline int irq_polarity(int idx
)
489 return MPBIOS_polarity(idx
);
492 static inline int irq_trigger(int idx
)
494 return MPBIOS_trigger(idx
);
497 static int next_irq
= 16;
500 * gsi_irq_sharing -- Name overload! "irq" can be either a legacy IRQ
501 * in the range 0-15, a linux IRQ in the range 0-223, or a GSI number
502 * from ACPI, which can reach 800 in large boxen.
504 * Compact the sparse GSI space into a sequential IRQ series and reuse
505 * vectors if possible.
507 int gsi_irq_sharing(int gsi
)
509 int i
, tries
, vector
;
511 BUG_ON(gsi
>= NR_IRQ_VECTORS
);
513 if (platform_legacy_irq(gsi
))
516 if (gsi_2_irq
[gsi
] != 0xFF)
517 return (int)gsi_2_irq
[gsi
];
521 vector
= assign_irq_vector(gsi
);
524 * Sharing vectors means sharing IRQs, so scan irq_vectors for previous
525 * use of vector and if found, return that IRQ. However, we never want
526 * to share legacy IRQs, which usually have a different trigger mode
529 for (i
= 0; i
< NR_IRQS
; i
++)
530 if (IO_APIC_VECTOR(i
) == vector
)
532 if (platform_legacy_irq(i
)) {
534 IO_APIC_VECTOR(i
) = 0;
537 panic("gsi_irq_sharing: didn't find an IRQ using vector 0x%02X for GSI %d", vector
, gsi
);
541 printk(KERN_INFO
"GSI %d sharing vector 0x%02X and IRQ %d\n",
547 BUG_ON(i
>= NR_IRQS
);
549 IO_APIC_VECTOR(i
) = vector
;
550 printk(KERN_INFO
"GSI %d assigned vector 0x%02X and IRQ %d\n",
555 static int pin_2_irq(int idx
, int apic
, int pin
)
558 int bus
= mp_irqs
[idx
].mpc_srcbus
;
561 * Debugging check, we are in big trouble if this message pops up!
563 if (mp_irqs
[idx
].mpc_dstirq
!= pin
)
564 printk(KERN_ERR
"broken BIOS or MPTABLE parser, ayiee!!\n");
566 if (test_bit(bus
, mp_bus_not_pci
)) {
567 irq
= mp_irqs
[idx
].mpc_srcbusirq
;
570 * PCI IRQs are mapped in order
574 irq
+= nr_ioapic_registers
[i
++];
576 irq
= gsi_irq_sharing(irq
);
578 BUG_ON(irq
>= NR_IRQS
);
582 static inline int IO_APIC_irq_trigger(int irq
)
586 for (apic
= 0; apic
< nr_ioapics
; apic
++) {
587 for (pin
= 0; pin
< nr_ioapic_registers
[apic
]; pin
++) {
588 idx
= find_irq_entry(apic
,pin
,mp_INT
);
589 if ((idx
!= -1) && (irq
== pin_2_irq(idx
,apic
,pin
)))
590 return irq_trigger(idx
);
594 * nonexistent IRQs are edge default
599 /* irq_vectors is indexed by the sum of all RTEs in all I/O APICs. */
600 u8 irq_vector
[NR_IRQ_VECTORS
] __read_mostly
= { FIRST_DEVICE_VECTOR
, 0 };
602 int assign_irq_vector(int irq
)
604 static int current_vector
= FIRST_DEVICE_VECTOR
, offset
= 0;
608 BUG_ON(irq
!= AUTO_ASSIGN
&& (unsigned)irq
>= NR_IRQ_VECTORS
);
610 spin_lock_irqsave(&vector_lock
, flags
);
612 if (irq
!= AUTO_ASSIGN
&& IO_APIC_VECTOR(irq
) > 0) {
613 spin_unlock_irqrestore(&vector_lock
, flags
);
614 return IO_APIC_VECTOR(irq
);
618 if (current_vector
== IA32_SYSCALL_VECTOR
)
621 if (current_vector
>= FIRST_SYSTEM_VECTOR
) {
622 /* If we run out of vectors on large boxen, must share them. */
623 offset
= (offset
+ 1) % 8;
624 current_vector
= FIRST_DEVICE_VECTOR
+ offset
;
627 vector
= current_vector
;
628 vector_irq
[vector
] = irq
;
629 if (irq
!= AUTO_ASSIGN
)
630 IO_APIC_VECTOR(irq
) = vector
;
632 spin_unlock_irqrestore(&vector_lock
, flags
);
637 extern void (*interrupt
[NR_IRQS
])(void);
639 static struct irq_chip ioapic_chip
;
641 #define IOAPIC_AUTO -1
642 #define IOAPIC_EDGE 0
643 #define IOAPIC_LEVEL 1
645 static void ioapic_register_intr(int irq
, int vector
, unsigned long trigger
)
649 idx
= use_pci_vector() && !platform_legacy_irq(irq
) ? vector
: irq
;
651 if ((trigger
== IOAPIC_AUTO
&& IO_APIC_irq_trigger(irq
)) ||
652 trigger
== IOAPIC_LEVEL
)
653 set_irq_chip_and_handler(idx
, &ioapic_chip
,
656 set_irq_chip_and_handler(idx
, &ioapic_chip
,
658 set_intr_gate(vector
, interrupt
[idx
]);
661 static void __init
setup_IO_APIC_irqs(void)
663 struct IO_APIC_route_entry entry
;
664 int apic
, pin
, idx
, irq
, first_notcon
= 1, vector
;
667 apic_printk(APIC_VERBOSE
, KERN_DEBUG
"init IO_APIC IRQs\n");
669 for (apic
= 0; apic
< nr_ioapics
; apic
++) {
670 for (pin
= 0; pin
< nr_ioapic_registers
[apic
]; pin
++) {
673 * add it to the IO-APIC irq-routing table:
675 memset(&entry
,0,sizeof(entry
));
677 entry
.delivery_mode
= INT_DELIVERY_MODE
;
678 entry
.dest_mode
= INT_DEST_MODE
;
679 entry
.mask
= 0; /* enable IRQ */
680 entry
.dest
.logical
.logical_dest
= cpu_mask_to_apicid(TARGET_CPUS
);
682 idx
= find_irq_entry(apic
,pin
,mp_INT
);
685 apic_printk(APIC_VERBOSE
, KERN_DEBUG
" IO-APIC (apicid-pin) %d-%d", mp_ioapics
[apic
].mpc_apicid
, pin
);
688 apic_printk(APIC_VERBOSE
, ", %d-%d", mp_ioapics
[apic
].mpc_apicid
, pin
);
692 entry
.trigger
= irq_trigger(idx
);
693 entry
.polarity
= irq_polarity(idx
);
695 if (irq_trigger(idx
)) {
698 entry
.dest
.logical
.logical_dest
= cpu_mask_to_apicid(TARGET_CPUS
);
701 irq
= pin_2_irq(idx
, apic
, pin
);
702 add_pin_to_irq(irq
, apic
, pin
);
704 if (!apic
&& !IO_APIC_IRQ(irq
))
707 if (IO_APIC_IRQ(irq
)) {
708 vector
= assign_irq_vector(irq
);
709 entry
.vector
= vector
;
711 ioapic_register_intr(irq
, vector
, IOAPIC_AUTO
);
712 if (!apic
&& (irq
< 16))
713 disable_8259A_irq(irq
);
715 ioapic_write_entry(apic
, pin
, entry
);
717 spin_lock_irqsave(&ioapic_lock
, flags
);
718 set_native_irq_info(irq
, TARGET_CPUS
);
719 spin_unlock_irqrestore(&ioapic_lock
, flags
);
724 apic_printk(APIC_VERBOSE
," not connected.\n");
728 * Set up the 8259A-master output pin as broadcast to all
731 static void __init
setup_ExtINT_IRQ0_pin(unsigned int apic
, unsigned int pin
, int vector
)
733 struct IO_APIC_route_entry entry
;
736 memset(&entry
,0,sizeof(entry
));
738 disable_8259A_irq(0);
741 apic_write(APIC_LVT0
, APIC_LVT_MASKED
| APIC_DM_EXTINT
);
744 * We use logical delivery to get the timer IRQ
747 entry
.dest_mode
= INT_DEST_MODE
;
748 entry
.mask
= 0; /* unmask IRQ now */
749 entry
.dest
.logical
.logical_dest
= cpu_mask_to_apicid(TARGET_CPUS
);
750 entry
.delivery_mode
= INT_DELIVERY_MODE
;
753 entry
.vector
= vector
;
756 * The timer IRQ doesn't have to know that behind the
757 * scene we have a 8259A-master in AEOI mode ...
759 set_irq_chip_and_handler(0, &ioapic_chip
, handle_edge_irq
);
762 * Add it to the IO-APIC irq-routing table:
764 spin_lock_irqsave(&ioapic_lock
, flags
);
765 io_apic_write(apic
, 0x11+2*pin
, *(((int *)&entry
)+1));
766 io_apic_write(apic
, 0x10+2*pin
, *(((int *)&entry
)+0));
767 spin_unlock_irqrestore(&ioapic_lock
, flags
);
772 void __init
UNEXPECTED_IO_APIC(void)
776 void __apicdebuginit
print_IO_APIC(void)
779 union IO_APIC_reg_00 reg_00
;
780 union IO_APIC_reg_01 reg_01
;
781 union IO_APIC_reg_02 reg_02
;
784 if (apic_verbosity
== APIC_QUIET
)
787 printk(KERN_DEBUG
"number of MP IRQ sources: %d.\n", mp_irq_entries
);
788 for (i
= 0; i
< nr_ioapics
; i
++)
789 printk(KERN_DEBUG
"number of IO-APIC #%d registers: %d.\n",
790 mp_ioapics
[i
].mpc_apicid
, nr_ioapic_registers
[i
]);
793 * We are a bit conservative about what we expect. We have to
794 * know about every hardware change ASAP.
796 printk(KERN_INFO
"testing the IO APIC.......................\n");
798 for (apic
= 0; apic
< nr_ioapics
; apic
++) {
800 spin_lock_irqsave(&ioapic_lock
, flags
);
801 reg_00
.raw
= io_apic_read(apic
, 0);
802 reg_01
.raw
= io_apic_read(apic
, 1);
803 if (reg_01
.bits
.version
>= 0x10)
804 reg_02
.raw
= io_apic_read(apic
, 2);
805 spin_unlock_irqrestore(&ioapic_lock
, flags
);
808 printk(KERN_DEBUG
"IO APIC #%d......\n", mp_ioapics
[apic
].mpc_apicid
);
809 printk(KERN_DEBUG
".... register #00: %08X\n", reg_00
.raw
);
810 printk(KERN_DEBUG
"....... : physical APIC id: %02X\n", reg_00
.bits
.ID
);
811 if (reg_00
.bits
.__reserved_1
|| reg_00
.bits
.__reserved_2
)
812 UNEXPECTED_IO_APIC();
814 printk(KERN_DEBUG
".... register #01: %08X\n", *(int *)®_01
);
815 printk(KERN_DEBUG
"....... : max redirection entries: %04X\n", reg_01
.bits
.entries
);
816 if ( (reg_01
.bits
.entries
!= 0x0f) && /* older (Neptune) boards */
817 (reg_01
.bits
.entries
!= 0x17) && /* typical ISA+PCI boards */
818 (reg_01
.bits
.entries
!= 0x1b) && /* Compaq Proliant boards */
819 (reg_01
.bits
.entries
!= 0x1f) && /* dual Xeon boards */
820 (reg_01
.bits
.entries
!= 0x22) && /* bigger Xeon boards */
821 (reg_01
.bits
.entries
!= 0x2E) &&
822 (reg_01
.bits
.entries
!= 0x3F) &&
823 (reg_01
.bits
.entries
!= 0x03)
825 UNEXPECTED_IO_APIC();
827 printk(KERN_DEBUG
"....... : PRQ implemented: %X\n", reg_01
.bits
.PRQ
);
828 printk(KERN_DEBUG
"....... : IO APIC version: %04X\n", reg_01
.bits
.version
);
829 if ( (reg_01
.bits
.version
!= 0x01) && /* 82489DX IO-APICs */
830 (reg_01
.bits
.version
!= 0x02) && /* 82801BA IO-APICs (ICH2) */
831 (reg_01
.bits
.version
!= 0x10) && /* oldest IO-APICs */
832 (reg_01
.bits
.version
!= 0x11) && /* Pentium/Pro IO-APICs */
833 (reg_01
.bits
.version
!= 0x13) && /* Xeon IO-APICs */
834 (reg_01
.bits
.version
!= 0x20) /* Intel P64H (82806 AA) */
836 UNEXPECTED_IO_APIC();
837 if (reg_01
.bits
.__reserved_1
|| reg_01
.bits
.__reserved_2
)
838 UNEXPECTED_IO_APIC();
840 if (reg_01
.bits
.version
>= 0x10) {
841 printk(KERN_DEBUG
".... register #02: %08X\n", reg_02
.raw
);
842 printk(KERN_DEBUG
"....... : arbitration: %02X\n", reg_02
.bits
.arbitration
);
843 if (reg_02
.bits
.__reserved_1
|| reg_02
.bits
.__reserved_2
)
844 UNEXPECTED_IO_APIC();
847 printk(KERN_DEBUG
".... IRQ redirection table:\n");
849 printk(KERN_DEBUG
" NR Log Phy Mask Trig IRR Pol"
850 " Stat Dest Deli Vect: \n");
852 for (i
= 0; i
<= reg_01
.bits
.entries
; i
++) {
853 struct IO_APIC_route_entry entry
;
855 entry
= ioapic_read_entry(apic
, i
);
857 printk(KERN_DEBUG
" %02x %03X %02X ",
859 entry
.dest
.logical
.logical_dest
,
860 entry
.dest
.physical
.physical_dest
863 printk("%1d %1d %1d %1d %1d %1d %1d %02X\n",
868 entry
.delivery_status
,
875 if (use_pci_vector())
876 printk(KERN_INFO
"Using vector-based indexing\n");
877 printk(KERN_DEBUG
"IRQ to pin mappings:\n");
878 for (i
= 0; i
< NR_IRQS
; i
++) {
879 struct irq_pin_list
*entry
= irq_2_pin
+ i
;
882 if (use_pci_vector() && !platform_legacy_irq(i
))
883 printk(KERN_DEBUG
"IRQ%d ", IO_APIC_VECTOR(i
));
885 printk(KERN_DEBUG
"IRQ%d ", i
);
887 printk("-> %d:%d", entry
->apic
, entry
->pin
);
890 entry
= irq_2_pin
+ entry
->next
;
895 printk(KERN_INFO
".................................... done.\n");
902 static __apicdebuginit
void print_APIC_bitfield (int base
)
907 if (apic_verbosity
== APIC_QUIET
)
910 printk(KERN_DEBUG
"0123456789abcdef0123456789abcdef\n" KERN_DEBUG
);
911 for (i
= 0; i
< 8; i
++) {
912 v
= apic_read(base
+ i
*0x10);
913 for (j
= 0; j
< 32; j
++) {
923 void __apicdebuginit
print_local_APIC(void * dummy
)
925 unsigned int v
, ver
, maxlvt
;
927 if (apic_verbosity
== APIC_QUIET
)
930 printk("\n" KERN_DEBUG
"printing local APIC contents on CPU#%d/%d:\n",
931 smp_processor_id(), hard_smp_processor_id());
932 v
= apic_read(APIC_ID
);
933 printk(KERN_INFO
"... APIC ID: %08x (%01x)\n", v
, GET_APIC_ID(v
));
934 v
= apic_read(APIC_LVR
);
935 printk(KERN_INFO
"... APIC VERSION: %08x\n", v
);
936 ver
= GET_APIC_VERSION(v
);
937 maxlvt
= get_maxlvt();
939 v
= apic_read(APIC_TASKPRI
);
940 printk(KERN_DEBUG
"... APIC TASKPRI: %08x (%02x)\n", v
, v
& APIC_TPRI_MASK
);
942 v
= apic_read(APIC_ARBPRI
);
943 printk(KERN_DEBUG
"... APIC ARBPRI: %08x (%02x)\n", v
,
944 v
& APIC_ARBPRI_MASK
);
945 v
= apic_read(APIC_PROCPRI
);
946 printk(KERN_DEBUG
"... APIC PROCPRI: %08x\n", v
);
948 v
= apic_read(APIC_EOI
);
949 printk(KERN_DEBUG
"... APIC EOI: %08x\n", v
);
950 v
= apic_read(APIC_RRR
);
951 printk(KERN_DEBUG
"... APIC RRR: %08x\n", v
);
952 v
= apic_read(APIC_LDR
);
953 printk(KERN_DEBUG
"... APIC LDR: %08x\n", v
);
954 v
= apic_read(APIC_DFR
);
955 printk(KERN_DEBUG
"... APIC DFR: %08x\n", v
);
956 v
= apic_read(APIC_SPIV
);
957 printk(KERN_DEBUG
"... APIC SPIV: %08x\n", v
);
959 printk(KERN_DEBUG
"... APIC ISR field:\n");
960 print_APIC_bitfield(APIC_ISR
);
961 printk(KERN_DEBUG
"... APIC TMR field:\n");
962 print_APIC_bitfield(APIC_TMR
);
963 printk(KERN_DEBUG
"... APIC IRR field:\n");
964 print_APIC_bitfield(APIC_IRR
);
966 v
= apic_read(APIC_ESR
);
967 printk(KERN_DEBUG
"... APIC ESR: %08x\n", v
);
969 v
= apic_read(APIC_ICR
);
970 printk(KERN_DEBUG
"... APIC ICR: %08x\n", v
);
971 v
= apic_read(APIC_ICR2
);
972 printk(KERN_DEBUG
"... APIC ICR2: %08x\n", v
);
974 v
= apic_read(APIC_LVTT
);
975 printk(KERN_DEBUG
"... APIC LVTT: %08x\n", v
);
977 if (maxlvt
> 3) { /* PC is LVT#4. */
978 v
= apic_read(APIC_LVTPC
);
979 printk(KERN_DEBUG
"... APIC LVTPC: %08x\n", v
);
981 v
= apic_read(APIC_LVT0
);
982 printk(KERN_DEBUG
"... APIC LVT0: %08x\n", v
);
983 v
= apic_read(APIC_LVT1
);
984 printk(KERN_DEBUG
"... APIC LVT1: %08x\n", v
);
986 if (maxlvt
> 2) { /* ERR is LVT#3. */
987 v
= apic_read(APIC_LVTERR
);
988 printk(KERN_DEBUG
"... APIC LVTERR: %08x\n", v
);
991 v
= apic_read(APIC_TMICT
);
992 printk(KERN_DEBUG
"... APIC TMICT: %08x\n", v
);
993 v
= apic_read(APIC_TMCCT
);
994 printk(KERN_DEBUG
"... APIC TMCCT: %08x\n", v
);
995 v
= apic_read(APIC_TDCR
);
996 printk(KERN_DEBUG
"... APIC TDCR: %08x\n", v
);
1000 void print_all_local_APICs (void)
1002 on_each_cpu(print_local_APIC
, NULL
, 1, 1);
1005 void __apicdebuginit
print_PIC(void)
1008 unsigned long flags
;
1010 if (apic_verbosity
== APIC_QUIET
)
1013 printk(KERN_DEBUG
"\nprinting PIC contents\n");
1015 spin_lock_irqsave(&i8259A_lock
, flags
);
1017 v
= inb(0xa1) << 8 | inb(0x21);
1018 printk(KERN_DEBUG
"... PIC IMR: %04x\n", v
);
1020 v
= inb(0xa0) << 8 | inb(0x20);
1021 printk(KERN_DEBUG
"... PIC IRR: %04x\n", v
);
1025 v
= inb(0xa0) << 8 | inb(0x20);
1029 spin_unlock_irqrestore(&i8259A_lock
, flags
);
1031 printk(KERN_DEBUG
"... PIC ISR: %04x\n", v
);
1033 v
= inb(0x4d1) << 8 | inb(0x4d0);
1034 printk(KERN_DEBUG
"... PIC ELCR: %04x\n", v
);
1039 static void __init
enable_IO_APIC(void)
1041 union IO_APIC_reg_01 reg_01
;
1042 int i8259_apic
, i8259_pin
;
1044 unsigned long flags
;
1046 for (i
= 0; i
< PIN_MAP_SIZE
; i
++) {
1047 irq_2_pin
[i
].pin
= -1;
1048 irq_2_pin
[i
].next
= 0;
1052 * The number of IO-APIC IRQ registers (== #pins):
1054 for (apic
= 0; apic
< nr_ioapics
; apic
++) {
1055 spin_lock_irqsave(&ioapic_lock
, flags
);
1056 reg_01
.raw
= io_apic_read(apic
, 1);
1057 spin_unlock_irqrestore(&ioapic_lock
, flags
);
1058 nr_ioapic_registers
[apic
] = reg_01
.bits
.entries
+1;
1060 for(apic
= 0; apic
< nr_ioapics
; apic
++) {
1062 /* See if any of the pins is in ExtINT mode */
1063 for (pin
= 0; pin
< nr_ioapic_registers
[apic
]; pin
++) {
1064 struct IO_APIC_route_entry entry
;
1065 entry
= ioapic_read_entry(apic
, pin
);
1067 /* If the interrupt line is enabled and in ExtInt mode
1068 * I have found the pin where the i8259 is connected.
1070 if ((entry
.mask
== 0) && (entry
.delivery_mode
== dest_ExtINT
)) {
1071 ioapic_i8259
.apic
= apic
;
1072 ioapic_i8259
.pin
= pin
;
1078 /* Look to see what if the MP table has reported the ExtINT */
1079 i8259_pin
= find_isa_irq_pin(0, mp_ExtINT
);
1080 i8259_apic
= find_isa_irq_apic(0, mp_ExtINT
);
1081 /* Trust the MP table if nothing is setup in the hardware */
1082 if ((ioapic_i8259
.pin
== -1) && (i8259_pin
>= 0)) {
1083 printk(KERN_WARNING
"ExtINT not setup in hardware but reported by MP table\n");
1084 ioapic_i8259
.pin
= i8259_pin
;
1085 ioapic_i8259
.apic
= i8259_apic
;
1087 /* Complain if the MP table and the hardware disagree */
1088 if (((ioapic_i8259
.apic
!= i8259_apic
) || (ioapic_i8259
.pin
!= i8259_pin
)) &&
1089 (i8259_pin
>= 0) && (ioapic_i8259
.pin
>= 0))
1091 printk(KERN_WARNING
"ExtINT in hardware and MP table differ\n");
1095 * Do not trust the IO-APIC being empty at bootup
1101 * Not an __init, needed by the reboot code
1103 void disable_IO_APIC(void)
1106 * Clear the IO-APIC before rebooting:
1111 * If the i8259 is routed through an IOAPIC
1112 * Put that IOAPIC in virtual wire mode
1113 * so legacy interrupts can be delivered.
1115 if (ioapic_i8259
.pin
!= -1) {
1116 struct IO_APIC_route_entry entry
;
1118 memset(&entry
, 0, sizeof(entry
));
1119 entry
.mask
= 0; /* Enabled */
1120 entry
.trigger
= 0; /* Edge */
1122 entry
.polarity
= 0; /* High */
1123 entry
.delivery_status
= 0;
1124 entry
.dest_mode
= 0; /* Physical */
1125 entry
.delivery_mode
= dest_ExtINT
; /* ExtInt */
1127 entry
.dest
.physical
.physical_dest
=
1128 GET_APIC_ID(apic_read(APIC_ID
));
1131 * Add it to the IO-APIC irq-routing table:
1133 ioapic_write_entry(ioapic_i8259
.apic
, ioapic_i8259
.pin
, entry
);
1136 disconnect_bsp_APIC(ioapic_i8259
.pin
!= -1);
1140 * There is a nasty bug in some older SMP boards, their mptable lies
1141 * about the timer IRQ. We do the following to work around the situation:
1143 * - timer IRQ defaults to IO-APIC IRQ
1144 * - if this function detects that timer IRQs are defunct, then we fall
1145 * back to ISA timer IRQs
1147 static int __init
timer_irq_works(void)
1149 unsigned long t1
= jiffies
;
1152 /* Let ten ticks pass... */
1153 mdelay((10 * 1000) / HZ
);
1156 * Expect a few ticks at least, to be sure some possible
1157 * glue logic does not lock up after one or two first
1158 * ticks in a non-ExtINT mode. Also the local APIC
1159 * might have cached one ExtINT interrupt. Finally, at
1160 * least one tick may be lost due to delays.
1164 if (jiffies
- t1
> 4)
1170 * In the SMP+IOAPIC case it might happen that there are an unspecified
1171 * number of pending IRQ events unhandled. These cases are very rare,
1172 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
1173 * better to do it this way as thus we do not have to be aware of
1174 * 'pending' interrupts in the IRQ path, except at this point.
1177 * Edge triggered needs to resend any interrupt
1178 * that was delayed but this is now handled in the device
1183 * Starting up a edge-triggered IO-APIC interrupt is
1184 * nasty - we need to make sure that we get the edge.
1185 * If it is already asserted for some reason, we need
1186 * return 1 to indicate that is was pending.
1188 * This is not complete - we should be able to fake
1189 * an edge even if it isn't on the 8259A...
1192 static unsigned int startup_ioapic_irq(unsigned int irq
)
1194 int was_pending
= 0;
1195 unsigned long flags
;
1197 spin_lock_irqsave(&ioapic_lock
, flags
);
1199 disable_8259A_irq(irq
);
1200 if (i8259A_irq_pending(irq
))
1203 __unmask_IO_APIC_irq(irq
);
1204 spin_unlock_irqrestore(&ioapic_lock
, flags
);
1209 static unsigned int startup_ioapic_vector(unsigned int vector
)
1211 int irq
= vector_to_irq(vector
);
1213 return startup_ioapic_irq(irq
);
1216 static void mask_ioapic_vector (unsigned int vector
)
1218 int irq
= vector_to_irq(vector
);
1220 mask_IO_APIC_irq(irq
);
1223 static void unmask_ioapic_vector (unsigned int vector
)
1225 int irq
= vector_to_irq(vector
);
1227 unmask_IO_APIC_irq(irq
);
1231 static void set_ioapic_affinity_vector (unsigned int vector
,
1234 int irq
= vector_to_irq(vector
);
1236 set_native_irq_info(vector
, cpu_mask
);
1237 set_ioapic_affinity_irq(irq
, cpu_mask
);
1239 #endif // CONFIG_SMP
1241 static int ioapic_retrigger_vector(unsigned int vector
)
1243 int irq
= vector_to_irq(vector
);
1245 send_IPI_self(IO_APIC_VECTOR(irq
));
1251 * Level and edge triggered IO-APIC interrupts need different handling,
1252 * so we use two separate IRQ descriptors. Edge triggered IRQs can be
1253 * handled with the level-triggered descriptor, but that one has slightly
1254 * more overhead. Level-triggered interrupts cannot be handled with the
1255 * edge-triggered handler, without risking IRQ storms and other ugly
1259 static void ack_apic_edge(unsigned int irq
)
1261 move_native_irq(irq
);
1265 static void ack_apic_level(unsigned int irq
)
1267 int do_unmask_irq
= 0;
1269 #if defined(CONFIG_GENERIC_PENDING_IRQ) || defined(CONFIG_IRQBALANCE)
1270 /* If we are moving the irq we need to mask it */
1271 if (unlikely(irq_desc
[irq
].status
& IRQ_MOVE_PENDING
)) {
1273 mask_IO_APIC_irq(irq
);
1278 * We must acknowledge the irq before we move it or the acknowledge will
1279 * not propogate properly.
1283 /* Now we can move and renable the irq */
1284 move_masked_irq(irq
);
1285 if (unlikely(do_unmask_irq
))
1286 unmask_IO_APIC_irq(irq
);
1289 static struct irq_chip ioapic_chip __read_mostly
= {
1291 .startup
= startup_ioapic_vector
,
1292 .mask
= mask_ioapic_vector
,
1293 .unmask
= unmask_ioapic_vector
,
1294 .ack
= ack_apic_edge
,
1295 .eoi
= ack_apic_level
,
1297 .set_affinity
= set_ioapic_affinity_vector
,
1299 .retrigger
= ioapic_retrigger_vector
,
1302 static inline void init_IO_APIC_traps(void)
1307 * NOTE! The local APIC isn't very good at handling
1308 * multiple interrupts at the same interrupt level.
1309 * As the interrupt level is determined by taking the
1310 * vector number and shifting that right by 4, we
1311 * want to spread these out a bit so that they don't
1312 * all fall in the same interrupt level.
1314 * Also, we've got to be careful not to trash gate
1315 * 0x80, because int 0x80 is hm, kind of importantish. ;)
1317 for (irq
= 0; irq
< NR_IRQS
; irq
++) {
1319 if (use_pci_vector()) {
1320 if (!platform_legacy_irq(tmp
))
1321 if ((tmp
= vector_to_irq(tmp
)) == -1)
1324 if (IO_APIC_IRQ(tmp
) && !IO_APIC_VECTOR(tmp
)) {
1326 * Hmm.. We don't have an entry for this,
1327 * so default to an old-fashioned 8259
1328 * interrupt if we can..
1331 make_8259A_irq(irq
);
1333 /* Strange. Oh, well.. */
1334 irq_desc
[irq
].chip
= &no_irq_chip
;
1339 static void enable_lapic_irq (unsigned int irq
)
1343 v
= apic_read(APIC_LVT0
);
1344 apic_write(APIC_LVT0
, v
& ~APIC_LVT_MASKED
);
1347 static void disable_lapic_irq (unsigned int irq
)
1351 v
= apic_read(APIC_LVT0
);
1352 apic_write(APIC_LVT0
, v
| APIC_LVT_MASKED
);
1355 static void ack_lapic_irq (unsigned int irq
)
1360 static void end_lapic_irq (unsigned int i
) { /* nothing */ }
1362 static struct hw_interrupt_type lapic_irq_type __read_mostly
= {
1363 .typename
= "local-APIC-edge",
1364 .startup
= NULL
, /* startup_irq() not used for IRQ0 */
1365 .shutdown
= NULL
, /* shutdown_irq() not used for IRQ0 */
1366 .enable
= enable_lapic_irq
,
1367 .disable
= disable_lapic_irq
,
1368 .ack
= ack_lapic_irq
,
1369 .end
= end_lapic_irq
,
1372 static void setup_nmi (void)
1375 * Dirty trick to enable the NMI watchdog ...
1376 * We put the 8259A master into AEOI mode and
1377 * unmask on all local APICs LVT0 as NMI.
1379 * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
1380 * is from Maciej W. Rozycki - so we do not have to EOI from
1381 * the NMI handler or the timer interrupt.
1383 printk(KERN_INFO
"activating NMI Watchdog ...");
1385 enable_NMI_through_LVT0(NULL
);
1391 * This looks a bit hackish but it's about the only one way of sending
1392 * a few INTA cycles to 8259As and any associated glue logic. ICR does
1393 * not support the ExtINT mode, unfortunately. We need to send these
1394 * cycles as some i82489DX-based boards have glue logic that keeps the
1395 * 8259A interrupt line asserted until INTA. --macro
1397 static inline void unlock_ExtINT_logic(void)
1400 struct IO_APIC_route_entry entry0
, entry1
;
1401 unsigned char save_control
, save_freq_select
;
1402 unsigned long flags
;
1404 pin
= find_isa_irq_pin(8, mp_INT
);
1405 apic
= find_isa_irq_apic(8, mp_INT
);
1409 spin_lock_irqsave(&ioapic_lock
, flags
);
1410 *(((int *)&entry0
) + 1) = io_apic_read(apic
, 0x11 + 2 * pin
);
1411 *(((int *)&entry0
) + 0) = io_apic_read(apic
, 0x10 + 2 * pin
);
1412 spin_unlock_irqrestore(&ioapic_lock
, flags
);
1413 clear_IO_APIC_pin(apic
, pin
);
1415 memset(&entry1
, 0, sizeof(entry1
));
1417 entry1
.dest_mode
= 0; /* physical delivery */
1418 entry1
.mask
= 0; /* unmask IRQ now */
1419 entry1
.dest
.physical
.physical_dest
= hard_smp_processor_id();
1420 entry1
.delivery_mode
= dest_ExtINT
;
1421 entry1
.polarity
= entry0
.polarity
;
1425 spin_lock_irqsave(&ioapic_lock
, flags
);
1426 io_apic_write(apic
, 0x11 + 2 * pin
, *(((int *)&entry1
) + 1));
1427 io_apic_write(apic
, 0x10 + 2 * pin
, *(((int *)&entry1
) + 0));
1428 spin_unlock_irqrestore(&ioapic_lock
, flags
);
1430 save_control
= CMOS_READ(RTC_CONTROL
);
1431 save_freq_select
= CMOS_READ(RTC_FREQ_SELECT
);
1432 CMOS_WRITE((save_freq_select
& ~RTC_RATE_SELECT
) | 0x6,
1434 CMOS_WRITE(save_control
| RTC_PIE
, RTC_CONTROL
);
1439 if ((CMOS_READ(RTC_INTR_FLAGS
) & RTC_PF
) == RTC_PF
)
1443 CMOS_WRITE(save_control
, RTC_CONTROL
);
1444 CMOS_WRITE(save_freq_select
, RTC_FREQ_SELECT
);
1445 clear_IO_APIC_pin(apic
, pin
);
1447 spin_lock_irqsave(&ioapic_lock
, flags
);
1448 io_apic_write(apic
, 0x11 + 2 * pin
, *(((int *)&entry0
) + 1));
1449 io_apic_write(apic
, 0x10 + 2 * pin
, *(((int *)&entry0
) + 0));
1450 spin_unlock_irqrestore(&ioapic_lock
, flags
);
1453 int timer_uses_ioapic_pin_0
;
1456 * This code may look a bit paranoid, but it's supposed to cooperate with
1457 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ
1458 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast
1459 * fanatically on his truly buggy board.
1461 * FIXME: really need to revamp this for modern platforms only.
1463 static inline void check_timer(void)
1465 int apic1
, pin1
, apic2
, pin2
;
1469 * get/set the timer IRQ vector:
1471 disable_8259A_irq(0);
1472 vector
= assign_irq_vector(0);
1473 set_intr_gate(vector
, interrupt
[0]);
1476 * Subtle, code in do_timer_interrupt() expects an AEOI
1477 * mode for the 8259A whenever interrupts are routed
1478 * through I/O APICs. Also IRQ0 has to be enabled in
1479 * the 8259A which implies the virtual wire has to be
1480 * disabled in the local APIC.
1482 apic_write(APIC_LVT0
, APIC_LVT_MASKED
| APIC_DM_EXTINT
);
1484 if (timer_over_8254
> 0)
1485 enable_8259A_irq(0);
1487 pin1
= find_isa_irq_pin(0, mp_INT
);
1488 apic1
= find_isa_irq_apic(0, mp_INT
);
1489 pin2
= ioapic_i8259
.pin
;
1490 apic2
= ioapic_i8259
.apic
;
1493 timer_uses_ioapic_pin_0
= 1;
1495 apic_printk(APIC_VERBOSE
,KERN_INFO
"..TIMER: vector=0x%02X apic1=%d pin1=%d apic2=%d pin2=%d\n",
1496 vector
, apic1
, pin1
, apic2
, pin2
);
1500 * Ok, does IRQ0 through the IOAPIC work?
1502 unmask_IO_APIC_irq(0);
1503 if (!no_timer_check
&& timer_irq_works()) {
1504 nmi_watchdog_default();
1505 if (nmi_watchdog
== NMI_IO_APIC
) {
1506 disable_8259A_irq(0);
1508 enable_8259A_irq(0);
1510 if (disable_timer_pin_1
> 0)
1511 clear_IO_APIC_pin(0, pin1
);
1514 clear_IO_APIC_pin(apic1
, pin1
);
1515 apic_printk(APIC_QUIET
,KERN_ERR
"..MP-BIOS bug: 8254 timer not "
1516 "connected to IO-APIC\n");
1519 apic_printk(APIC_VERBOSE
,KERN_INFO
"...trying to set up timer (IRQ0) "
1520 "through the 8259A ... ");
1522 apic_printk(APIC_VERBOSE
,"\n..... (found apic %d pin %d) ...",
1525 * legacy devices should be connected to IO APIC #0
1527 setup_ExtINT_IRQ0_pin(apic2
, pin2
, vector
);
1528 if (timer_irq_works()) {
1529 apic_printk(APIC_VERBOSE
," works.\n");
1530 nmi_watchdog_default();
1531 if (nmi_watchdog
== NMI_IO_APIC
) {
1537 * Cleanup, just in case ...
1539 clear_IO_APIC_pin(apic2
, pin2
);
1541 apic_printk(APIC_VERBOSE
," failed.\n");
1543 if (nmi_watchdog
== NMI_IO_APIC
) {
1544 printk(KERN_WARNING
"timer doesn't work through the IO-APIC - disabling NMI Watchdog!\n");
1548 apic_printk(APIC_VERBOSE
, KERN_INFO
"...trying to set up timer as Virtual Wire IRQ...");
1550 disable_8259A_irq(0);
1551 irq_desc
[0].chip
= &lapic_irq_type
;
1552 apic_write(APIC_LVT0
, APIC_DM_FIXED
| vector
); /* Fixed mode */
1553 enable_8259A_irq(0);
1555 if (timer_irq_works()) {
1556 apic_printk(APIC_VERBOSE
," works.\n");
1559 apic_write(APIC_LVT0
, APIC_LVT_MASKED
| APIC_DM_FIXED
| vector
);
1560 apic_printk(APIC_VERBOSE
," failed.\n");
1562 apic_printk(APIC_VERBOSE
, KERN_INFO
"...trying to set up timer as ExtINT IRQ...");
1566 apic_write(APIC_LVT0
, APIC_DM_EXTINT
);
1568 unlock_ExtINT_logic();
1570 if (timer_irq_works()) {
1571 apic_printk(APIC_VERBOSE
," works.\n");
1574 apic_printk(APIC_VERBOSE
," failed :(.\n");
1575 panic("IO-APIC + timer doesn't work! Try using the 'noapic' kernel parameter\n");
1578 static int __init
notimercheck(char *s
)
1583 __setup("no_timer_check", notimercheck
);
1587 * IRQ's that are handled by the PIC in the MPS IOAPIC case.
1588 * - IRQ2 is the cascade IRQ, and cannot be a io-apic IRQ.
1589 * Linux doesn't really care, as it's not actually used
1590 * for any interrupt handling anyway.
1592 #define PIC_IRQS (1<<2)
1594 void __init
setup_IO_APIC(void)
1599 io_apic_irqs
= ~0; /* all IRQs go through IOAPIC */
1601 io_apic_irqs
= ~PIC_IRQS
;
1603 apic_printk(APIC_VERBOSE
, "ENABLING IO-APIC IRQs\n");
1606 setup_IO_APIC_irqs();
1607 init_IO_APIC_traps();
1613 struct sysfs_ioapic_data
{
1614 struct sys_device dev
;
1615 struct IO_APIC_route_entry entry
[0];
1617 static struct sysfs_ioapic_data
* mp_ioapic_data
[MAX_IO_APICS
];
1619 static int ioapic_suspend(struct sys_device
*dev
, pm_message_t state
)
1621 struct IO_APIC_route_entry
*entry
;
1622 struct sysfs_ioapic_data
*data
;
1625 data
= container_of(dev
, struct sysfs_ioapic_data
, dev
);
1626 entry
= data
->entry
;
1627 for (i
= 0; i
< nr_ioapic_registers
[dev
->id
]; i
++, entry
++ )
1628 *entry
= ioapic_read_entry(dev
->id
, i
);
1633 static int ioapic_resume(struct sys_device
*dev
)
1635 struct IO_APIC_route_entry
*entry
;
1636 struct sysfs_ioapic_data
*data
;
1637 unsigned long flags
;
1638 union IO_APIC_reg_00 reg_00
;
1641 data
= container_of(dev
, struct sysfs_ioapic_data
, dev
);
1642 entry
= data
->entry
;
1644 spin_lock_irqsave(&ioapic_lock
, flags
);
1645 reg_00
.raw
= io_apic_read(dev
->id
, 0);
1646 if (reg_00
.bits
.ID
!= mp_ioapics
[dev
->id
].mpc_apicid
) {
1647 reg_00
.bits
.ID
= mp_ioapics
[dev
->id
].mpc_apicid
;
1648 io_apic_write(dev
->id
, 0, reg_00
.raw
);
1650 spin_unlock_irqrestore(&ioapic_lock
, flags
);
1651 for (i
= 0; i
< nr_ioapic_registers
[dev
->id
]; i
++)
1652 ioapic_write_entry(dev
->id
, i
, entry
[i
]);
1657 static struct sysdev_class ioapic_sysdev_class
= {
1658 set_kset_name("ioapic"),
1659 .suspend
= ioapic_suspend
,
1660 .resume
= ioapic_resume
,
1663 static int __init
ioapic_init_sysfs(void)
1665 struct sys_device
* dev
;
1666 int i
, size
, error
= 0;
1668 error
= sysdev_class_register(&ioapic_sysdev_class
);
1672 for (i
= 0; i
< nr_ioapics
; i
++ ) {
1673 size
= sizeof(struct sys_device
) + nr_ioapic_registers
[i
]
1674 * sizeof(struct IO_APIC_route_entry
);
1675 mp_ioapic_data
[i
] = kmalloc(size
, GFP_KERNEL
);
1676 if (!mp_ioapic_data
[i
]) {
1677 printk(KERN_ERR
"Can't suspend/resume IOAPIC %d\n", i
);
1680 memset(mp_ioapic_data
[i
], 0, size
);
1681 dev
= &mp_ioapic_data
[i
]->dev
;
1683 dev
->cls
= &ioapic_sysdev_class
;
1684 error
= sysdev_register(dev
);
1686 kfree(mp_ioapic_data
[i
]);
1687 mp_ioapic_data
[i
] = NULL
;
1688 printk(KERN_ERR
"Can't suspend/resume IOAPIC %d\n", i
);
1696 device_initcall(ioapic_init_sysfs
);
1698 #ifdef CONFIG_PCI_MSI
1700 * Dynamic irq allocate and deallocation for MSI
1702 int create_irq(void)
1704 /* Hack of the day: irq == vector.
1706 * Ultimately this will be be more general,
1707 * and not depend on the irq to vector identity mapping.
1708 * But this version is needed until msi.c can cope with
1709 * the more general form.
1712 unsigned long flags
;
1713 vector
= assign_irq_vector(AUTO_ASSIGN
);
1717 spin_lock_irqsave(&vector_lock
, flags
);
1718 vector_irq
[vector
] = irq
;
1719 irq_vector
[irq
] = vector
;
1720 spin_unlock_irqrestore(&vector_lock
, flags
);
1722 set_intr_gate(vector
, interrupt
[irq
]);
1724 dynamic_irq_init(irq
);
1729 void destroy_irq(unsigned int irq
)
1731 unsigned long flags
;
1732 unsigned int vector
;
1734 dynamic_irq_cleanup(irq
);
1736 spin_lock_irqsave(&vector_lock
, flags
);
1737 vector
= irq_vector
[irq
];
1738 vector_irq
[vector
] = -1;
1739 irq_vector
[irq
] = 0;
1740 spin_unlock_irqrestore(&vector_lock
, flags
);
1745 * MSI mesage composition
1747 #ifdef CONFIG_PCI_MSI
1748 static int msi_msg_setup(struct pci_dev
*pdev
, unsigned int irq
, struct msi_msg
*msg
)
1750 /* For now always this code always uses physical delivery
1756 vector
= assign_irq_vector(irq
);
1761 cpu_set(first_cpu(cpu_online_map
), tmp
);
1762 dest
= cpu_mask_to_apicid(tmp
);
1764 msg
->address_hi
= MSI_ADDR_BASE_HI
;
1767 ((INT_DEST_MODE
== 0) ?
1768 MSI_ADDR_DEST_MODE_PHYSICAL
:
1769 MSI_ADDR_DEST_MODE_LOGICAL
) |
1770 ((INT_DELIVERY_MODE
!= dest_LowestPrio
) ?
1771 MSI_ADDR_REDIRECTION_CPU
:
1772 MSI_ADDR_REDIRECTION_LOWPRI
) |
1773 MSI_ADDR_DEST_ID(dest
);
1776 MSI_DATA_TRIGGER_EDGE
|
1777 MSI_DATA_LEVEL_ASSERT
|
1778 ((INT_DELIVERY_MODE
!= dest_LowestPrio
) ?
1779 MSI_DATA_DELIVERY_FIXED
:
1780 MSI_DATA_DELIVERY_LOWPRI
) |
1781 MSI_DATA_VECTOR(vector
);
1786 static void msi_msg_teardown(unsigned int irq
)
1791 static void msi_msg_set_affinity(unsigned int irq
, cpumask_t mask
, struct msi_msg
*msg
)
1796 vector
= assign_irq_vector(irq
);
1798 dest
= cpu_mask_to_apicid(mask
);
1800 msg
->data
&= ~MSI_DATA_VECTOR_MASK
;
1801 msg
->data
|= MSI_DATA_VECTOR(vector
);
1802 msg
->address_lo
&= ~MSI_ADDR_DEST_ID_MASK
;
1803 msg
->address_lo
|= MSI_ADDR_DEST_ID(dest
);
1807 struct msi_ops arch_msi_ops
= {
1808 .needs_64bit_address
= 0,
1809 .setup
= msi_msg_setup
,
1810 .teardown
= msi_msg_teardown
,
1811 .target
= msi_msg_set_affinity
,
1816 /* --------------------------------------------------------------------------
1817 ACPI-based IOAPIC Configuration
1818 -------------------------------------------------------------------------- */
1822 #define IO_APIC_MAX_ID 0xFE
1824 int __init
io_apic_get_redir_entries (int ioapic
)
1826 union IO_APIC_reg_01 reg_01
;
1827 unsigned long flags
;
1829 spin_lock_irqsave(&ioapic_lock
, flags
);
1830 reg_01
.raw
= io_apic_read(ioapic
, 1);
1831 spin_unlock_irqrestore(&ioapic_lock
, flags
);
1833 return reg_01
.bits
.entries
;
1837 int io_apic_set_pci_routing (int ioapic
, int pin
, int irq
, int triggering
, int polarity
)
1839 struct IO_APIC_route_entry entry
;
1840 unsigned long flags
;
1842 if (!IO_APIC_IRQ(irq
)) {
1843 apic_printk(APIC_QUIET
,KERN_ERR
"IOAPIC[%d]: Invalid reference to IRQ 0\n",
1849 * Generate a PCI IRQ routing entry and program the IOAPIC accordingly.
1850 * Note that we mask (disable) IRQs now -- these get enabled when the
1851 * corresponding device driver registers for this IRQ.
1854 memset(&entry
,0,sizeof(entry
));
1856 entry
.delivery_mode
= INT_DELIVERY_MODE
;
1857 entry
.dest_mode
= INT_DEST_MODE
;
1858 entry
.dest
.logical
.logical_dest
= cpu_mask_to_apicid(TARGET_CPUS
);
1859 entry
.trigger
= triggering
;
1860 entry
.polarity
= polarity
;
1861 entry
.mask
= 1; /* Disabled (masked) */
1863 irq
= gsi_irq_sharing(irq
);
1865 * IRQs < 16 are already in the irq_2_pin[] map
1868 add_pin_to_irq(irq
, ioapic
, pin
);
1870 entry
.vector
= assign_irq_vector(irq
);
1872 apic_printk(APIC_VERBOSE
,KERN_DEBUG
"IOAPIC[%d]: Set PCI routing entry (%d-%d -> 0x%x -> "
1873 "IRQ %d Mode:%i Active:%i)\n", ioapic
,
1874 mp_ioapics
[ioapic
].mpc_apicid
, pin
, entry
.vector
, irq
,
1875 triggering
, polarity
);
1877 ioapic_register_intr(irq
, entry
.vector
, triggering
);
1879 if (!ioapic
&& (irq
< 16))
1880 disable_8259A_irq(irq
);
1882 ioapic_write_entry(ioapic
, pin
, entry
);
1884 spin_lock_irqsave(&ioapic_lock
, flags
);
1885 set_native_irq_info(use_pci_vector() ? entry
.vector
: irq
, TARGET_CPUS
);
1886 spin_unlock_irqrestore(&ioapic_lock
, flags
);
1891 #endif /* CONFIG_ACPI */
1895 * This function currently is only a helper for the i386 smp boot process where
1896 * we need to reprogram the ioredtbls to cater for the cpus which have come online
1897 * so mask in all cases should simply be TARGET_CPUS
1900 void __init
setup_ioapic_dest(void)
1902 int pin
, ioapic
, irq
, irq_entry
;
1904 if (skip_ioapic_setup
== 1)
1907 for (ioapic
= 0; ioapic
< nr_ioapics
; ioapic
++) {
1908 for (pin
= 0; pin
< nr_ioapic_registers
[ioapic
]; pin
++) {
1909 irq_entry
= find_irq_entry(ioapic
, pin
, mp_INT
);
1910 if (irq_entry
== -1)
1912 irq
= pin_2_irq(irq_entry
, ioapic
, pin
);
1913 set_ioapic_affinity_irq(irq
, TARGET_CPUS
);