2 * Intel IO-APIC support for multi-Pentium hosts.
4 * Copyright (C) 1997, 1998, 1999, 2000, 2009 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/pci.h>
29 #include <linux/mc146818rtc.h>
30 #include <linux/compiler.h>
31 #include <linux/acpi.h>
32 #include <linux/module.h>
33 #include <linux/sysdev.h>
34 #include <linux/msi.h>
35 #include <linux/htirq.h>
36 #include <linux/freezer.h>
37 #include <linux/kthread.h>
38 #include <linux/jiffies.h> /* time_after() */
39 #include <linux/slab.h>
41 #include <acpi/acpi_bus.h>
43 #include <linux/bootmem.h>
44 #include <linux/dmar.h>
45 #include <linux/hpet.h>
52 #include <asm/proto.h>
55 #include <asm/timer.h>
56 #include <asm/i8259.h>
58 #include <asm/msidef.h>
59 #include <asm/hypertransport.h>
60 #include <asm/setup.h>
61 #include <asm/irq_remapping.h>
63 #include <asm/hw_irq.h>
67 #define __apicdebuginit(type) static type __init
68 #define for_each_irq_pin(entry, head) \
69 for (entry = head; entry; entry = entry->next)
72 * Is the SiS APIC rmw bug present ?
73 * -1 = don't know, 0 = no, 1 = yes
75 int sis_apic_bug
= -1;
77 static DEFINE_RAW_SPINLOCK(ioapic_lock
);
78 static DEFINE_RAW_SPINLOCK(vector_lock
);
81 * # of IRQ routing registers
83 int nr_ioapic_registers
[MAX_IO_APICS
];
85 /* I/O APIC entries */
86 struct mpc_ioapic mp_ioapics
[MAX_IO_APICS
];
89 /* IO APIC gsi routing info */
90 struct mp_ioapic_gsi mp_gsi_routing
[MAX_IO_APICS
];
92 /* The one past the highest gsi number used */
95 /* MP IRQ source entries */
96 struct mpc_intsrc mp_irqs
[MAX_IRQ_SOURCES
];
98 /* # of MP IRQ source entries */
102 static int nr_irqs_gsi
= NR_IRQS_LEGACY
;
104 #if defined (CONFIG_MCA) || defined (CONFIG_EISA)
105 int mp_bus_id_to_type
[MAX_MP_BUSSES
];
108 DECLARE_BITMAP(mp_bus_not_pci
, MAX_MP_BUSSES
);
110 int skip_ioapic_setup
;
112 void arch_disable_smp_support(void)
116 noioapicreroute
= -1;
118 skip_ioapic_setup
= 1;
121 static int __init
parse_noapic(char *str
)
123 /* disable IO-APIC */
124 arch_disable_smp_support();
127 early_param("noapic", parse_noapic
);
129 struct irq_pin_list
{
131 struct irq_pin_list
*next
;
134 static struct irq_pin_list
*get_one_free_irq_2_pin(int node
)
136 struct irq_pin_list
*pin
;
138 pin
= kzalloc_node(sizeof(*pin
), GFP_ATOMIC
, node
);
143 /* irq_cfg is indexed by the sum of all RTEs in all I/O APICs. */
144 #ifdef CONFIG_SPARSE_IRQ
145 static struct irq_cfg irq_cfgx
[NR_IRQS_LEGACY
];
147 static struct irq_cfg irq_cfgx
[NR_IRQS
];
150 int __init
arch_early_irq_init(void)
153 struct irq_desc
*desc
;
158 if (!legacy_pic
->nr_legacy_irqs
) {
164 count
= ARRAY_SIZE(irq_cfgx
);
165 node
= cpu_to_node(boot_cpu_id
);
167 for (i
= 0; i
< count
; i
++) {
168 desc
= irq_to_desc(i
);
169 desc
->chip_data
= &cfg
[i
];
170 zalloc_cpumask_var_node(&cfg
[i
].domain
, GFP_NOWAIT
, node
);
171 zalloc_cpumask_var_node(&cfg
[i
].old_domain
, GFP_NOWAIT
, node
);
173 * For legacy IRQ's, start with assigning irq0 to irq15 to
174 * IRQ0_VECTOR to IRQ15_VECTOR on cpu 0.
176 if (i
< legacy_pic
->nr_legacy_irqs
) {
177 cfg
[i
].vector
= IRQ0_VECTOR
+ i
;
178 cpumask_set_cpu(0, cfg
[i
].domain
);
185 #ifdef CONFIG_SPARSE_IRQ
186 struct irq_cfg
*irq_cfg(unsigned int irq
)
188 struct irq_cfg
*cfg
= NULL
;
189 struct irq_desc
*desc
;
191 desc
= irq_to_desc(irq
);
193 cfg
= desc
->chip_data
;
198 static struct irq_cfg
*get_one_free_irq_cfg(int node
)
202 cfg
= kzalloc_node(sizeof(*cfg
), GFP_ATOMIC
, node
);
204 if (!zalloc_cpumask_var_node(&cfg
->domain
, GFP_ATOMIC
, node
)) {
207 } else if (!zalloc_cpumask_var_node(&cfg
->old_domain
,
209 free_cpumask_var(cfg
->domain
);
218 int arch_init_chip_data(struct irq_desc
*desc
, int node
)
222 cfg
= desc
->chip_data
;
224 desc
->chip_data
= get_one_free_irq_cfg(node
);
225 if (!desc
->chip_data
) {
226 printk(KERN_ERR
"can not alloc irq_cfg\n");
234 /* for move_irq_desc */
236 init_copy_irq_2_pin(struct irq_cfg
*old_cfg
, struct irq_cfg
*cfg
, int node
)
238 struct irq_pin_list
*old_entry
, *head
, *tail
, *entry
;
240 cfg
->irq_2_pin
= NULL
;
241 old_entry
= old_cfg
->irq_2_pin
;
245 entry
= get_one_free_irq_2_pin(node
);
249 entry
->apic
= old_entry
->apic
;
250 entry
->pin
= old_entry
->pin
;
253 old_entry
= old_entry
->next
;
255 entry
= get_one_free_irq_2_pin(node
);
263 /* still use the old one */
266 entry
->apic
= old_entry
->apic
;
267 entry
->pin
= old_entry
->pin
;
270 old_entry
= old_entry
->next
;
274 cfg
->irq_2_pin
= head
;
277 static void free_irq_2_pin(struct irq_cfg
*old_cfg
, struct irq_cfg
*cfg
)
279 struct irq_pin_list
*entry
, *next
;
281 if (old_cfg
->irq_2_pin
== cfg
->irq_2_pin
)
284 entry
= old_cfg
->irq_2_pin
;
291 old_cfg
->irq_2_pin
= NULL
;
294 void arch_init_copy_chip_data(struct irq_desc
*old_desc
,
295 struct irq_desc
*desc
, int node
)
298 struct irq_cfg
*old_cfg
;
300 cfg
= get_one_free_irq_cfg(node
);
305 desc
->chip_data
= cfg
;
307 old_cfg
= old_desc
->chip_data
;
309 cfg
->vector
= old_cfg
->vector
;
310 cfg
->move_in_progress
= old_cfg
->move_in_progress
;
311 cpumask_copy(cfg
->domain
, old_cfg
->domain
);
312 cpumask_copy(cfg
->old_domain
, old_cfg
->old_domain
);
314 init_copy_irq_2_pin(old_cfg
, cfg
, node
);
317 static void free_irq_cfg(struct irq_cfg
*cfg
)
319 free_cpumask_var(cfg
->domain
);
320 free_cpumask_var(cfg
->old_domain
);
324 void arch_free_chip_data(struct irq_desc
*old_desc
, struct irq_desc
*desc
)
326 struct irq_cfg
*old_cfg
, *cfg
;
328 old_cfg
= old_desc
->chip_data
;
329 cfg
= desc
->chip_data
;
335 free_irq_2_pin(old_cfg
, cfg
);
336 free_irq_cfg(old_cfg
);
337 old_desc
->chip_data
= NULL
;
340 /* end for move_irq_desc */
343 struct irq_cfg
*irq_cfg(unsigned int irq
)
345 return irq
< nr_irqs
? irq_cfgx
+ irq
: NULL
;
352 unsigned int unused
[3];
354 unsigned int unused2
[11];
358 static __attribute_const__
struct io_apic __iomem
*io_apic_base(int idx
)
360 return (void __iomem
*) __fix_to_virt(FIX_IO_APIC_BASE_0
+ idx
)
361 + (mp_ioapics
[idx
].apicaddr
& ~PAGE_MASK
);
364 static inline void io_apic_eoi(unsigned int apic
, unsigned int vector
)
366 struct io_apic __iomem
*io_apic
= io_apic_base(apic
);
367 writel(vector
, &io_apic
->eoi
);
370 static inline unsigned int io_apic_read(unsigned int apic
, unsigned int reg
)
372 struct io_apic __iomem
*io_apic
= io_apic_base(apic
);
373 writel(reg
, &io_apic
->index
);
374 return readl(&io_apic
->data
);
377 static inline void io_apic_write(unsigned int apic
, unsigned int reg
, unsigned int value
)
379 struct io_apic __iomem
*io_apic
= io_apic_base(apic
);
380 writel(reg
, &io_apic
->index
);
381 writel(value
, &io_apic
->data
);
385 * Re-write a value: to be used for read-modify-write
386 * cycles where the read already set up the index register.
388 * Older SiS APIC requires we rewrite the index register
390 static inline void io_apic_modify(unsigned int apic
, unsigned int reg
, unsigned int value
)
392 struct io_apic __iomem
*io_apic
= io_apic_base(apic
);
395 writel(reg
, &io_apic
->index
);
396 writel(value
, &io_apic
->data
);
399 static bool io_apic_level_ack_pending(struct irq_cfg
*cfg
)
401 struct irq_pin_list
*entry
;
404 raw_spin_lock_irqsave(&ioapic_lock
, flags
);
405 for_each_irq_pin(entry
, cfg
->irq_2_pin
) {
410 reg
= io_apic_read(entry
->apic
, 0x10 + pin
*2);
411 /* Is the remote IRR bit set? */
412 if (reg
& IO_APIC_REDIR_REMOTE_IRR
) {
413 raw_spin_unlock_irqrestore(&ioapic_lock
, flags
);
417 raw_spin_unlock_irqrestore(&ioapic_lock
, flags
);
423 struct { u32 w1
, w2
; };
424 struct IO_APIC_route_entry entry
;
427 static struct IO_APIC_route_entry
ioapic_read_entry(int apic
, int pin
)
429 union entry_union eu
;
431 raw_spin_lock_irqsave(&ioapic_lock
, flags
);
432 eu
.w1
= io_apic_read(apic
, 0x10 + 2 * pin
);
433 eu
.w2
= io_apic_read(apic
, 0x11 + 2 * pin
);
434 raw_spin_unlock_irqrestore(&ioapic_lock
, flags
);
439 * When we write a new IO APIC routing entry, we need to write the high
440 * word first! If the mask bit in the low word is clear, we will enable
441 * the interrupt, and we need to make sure the entry is fully populated
442 * before that happens.
445 __ioapic_write_entry(int apic
, int pin
, struct IO_APIC_route_entry e
)
447 union entry_union eu
= {{0, 0}};
450 io_apic_write(apic
, 0x11 + 2*pin
, eu
.w2
);
451 io_apic_write(apic
, 0x10 + 2*pin
, eu
.w1
);
454 void ioapic_write_entry(int apic
, int pin
, struct IO_APIC_route_entry e
)
457 raw_spin_lock_irqsave(&ioapic_lock
, flags
);
458 __ioapic_write_entry(apic
, pin
, e
);
459 raw_spin_unlock_irqrestore(&ioapic_lock
, flags
);
463 * When we mask an IO APIC routing entry, we need to write the low
464 * word first, in order to set the mask bit before we change the
467 static void ioapic_mask_entry(int apic
, int pin
)
470 union entry_union eu
= { .entry
.mask
= 1 };
472 raw_spin_lock_irqsave(&ioapic_lock
, flags
);
473 io_apic_write(apic
, 0x10 + 2*pin
, eu
.w1
);
474 io_apic_write(apic
, 0x11 + 2*pin
, eu
.w2
);
475 raw_spin_unlock_irqrestore(&ioapic_lock
, flags
);
479 * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
480 * shared ISA-space IRQs, so we have to support them. We are super
481 * fast in the common case, and fast for shared ISA-space IRQs.
484 add_pin_to_irq_node_nopanic(struct irq_cfg
*cfg
, int node
, int apic
, int pin
)
486 struct irq_pin_list
**last
, *entry
;
488 /* don't allow duplicates */
489 last
= &cfg
->irq_2_pin
;
490 for_each_irq_pin(entry
, cfg
->irq_2_pin
) {
491 if (entry
->apic
== apic
&& entry
->pin
== pin
)
496 entry
= get_one_free_irq_2_pin(node
);
498 printk(KERN_ERR
"can not alloc irq_pin_list (%d,%d,%d)\n",
509 static void add_pin_to_irq_node(struct irq_cfg
*cfg
, int node
, int apic
, int pin
)
511 if (add_pin_to_irq_node_nopanic(cfg
, node
, apic
, pin
))
512 panic("IO-APIC: failed to add irq-pin. Can not proceed\n");
516 * Reroute an IRQ to a different pin.
518 static void __init
replace_pin_at_irq_node(struct irq_cfg
*cfg
, int node
,
519 int oldapic
, int oldpin
,
520 int newapic
, int newpin
)
522 struct irq_pin_list
*entry
;
524 for_each_irq_pin(entry
, cfg
->irq_2_pin
) {
525 if (entry
->apic
== oldapic
&& entry
->pin
== oldpin
) {
526 entry
->apic
= newapic
;
528 /* every one is different, right? */
533 /* old apic/pin didn't exist, so just add new ones */
534 add_pin_to_irq_node(cfg
, node
, newapic
, newpin
);
537 static void __io_apic_modify_irq(struct irq_pin_list
*entry
,
538 int mask_and
, int mask_or
,
539 void (*final
)(struct irq_pin_list
*entry
))
541 unsigned int reg
, pin
;
544 reg
= io_apic_read(entry
->apic
, 0x10 + pin
* 2);
547 io_apic_modify(entry
->apic
, 0x10 + pin
* 2, reg
);
552 static void io_apic_modify_irq(struct irq_cfg
*cfg
,
553 int mask_and
, int mask_or
,
554 void (*final
)(struct irq_pin_list
*entry
))
556 struct irq_pin_list
*entry
;
558 for_each_irq_pin(entry
, cfg
->irq_2_pin
)
559 __io_apic_modify_irq(entry
, mask_and
, mask_or
, final
);
562 static void __mask_and_edge_IO_APIC_irq(struct irq_pin_list
*entry
)
564 __io_apic_modify_irq(entry
, ~IO_APIC_REDIR_LEVEL_TRIGGER
,
565 IO_APIC_REDIR_MASKED
, NULL
);
568 static void __unmask_and_level_IO_APIC_irq(struct irq_pin_list
*entry
)
570 __io_apic_modify_irq(entry
, ~IO_APIC_REDIR_MASKED
,
571 IO_APIC_REDIR_LEVEL_TRIGGER
, NULL
);
574 static void __unmask_IO_APIC_irq(struct irq_cfg
*cfg
)
576 io_apic_modify_irq(cfg
, ~IO_APIC_REDIR_MASKED
, 0, NULL
);
579 static void io_apic_sync(struct irq_pin_list
*entry
)
582 * Synchronize the IO-APIC and the CPU by doing
583 * a dummy read from the IO-APIC
585 struct io_apic __iomem
*io_apic
;
586 io_apic
= io_apic_base(entry
->apic
);
587 readl(&io_apic
->data
);
590 static void __mask_IO_APIC_irq(struct irq_cfg
*cfg
)
592 io_apic_modify_irq(cfg
, ~0, IO_APIC_REDIR_MASKED
, &io_apic_sync
);
595 static void mask_IO_APIC_irq_desc(struct irq_desc
*desc
)
597 struct irq_cfg
*cfg
= desc
->chip_data
;
602 raw_spin_lock_irqsave(&ioapic_lock
, flags
);
603 __mask_IO_APIC_irq(cfg
);
604 raw_spin_unlock_irqrestore(&ioapic_lock
, flags
);
607 static void unmask_IO_APIC_irq_desc(struct irq_desc
*desc
)
609 struct irq_cfg
*cfg
= desc
->chip_data
;
612 raw_spin_lock_irqsave(&ioapic_lock
, flags
);
613 __unmask_IO_APIC_irq(cfg
);
614 raw_spin_unlock_irqrestore(&ioapic_lock
, flags
);
617 static void mask_IO_APIC_irq(unsigned int irq
)
619 struct irq_desc
*desc
= irq_to_desc(irq
);
621 mask_IO_APIC_irq_desc(desc
);
623 static void unmask_IO_APIC_irq(unsigned int irq
)
625 struct irq_desc
*desc
= irq_to_desc(irq
);
627 unmask_IO_APIC_irq_desc(desc
);
630 static void clear_IO_APIC_pin(unsigned int apic
, unsigned int pin
)
632 struct IO_APIC_route_entry entry
;
634 /* Check delivery_mode to be sure we're not clearing an SMI pin */
635 entry
= ioapic_read_entry(apic
, pin
);
636 if (entry
.delivery_mode
== dest_SMI
)
639 * Disable it in the IO-APIC irq-routing table:
641 ioapic_mask_entry(apic
, pin
);
644 static void clear_IO_APIC (void)
648 for (apic
= 0; apic
< nr_ioapics
; apic
++)
649 for (pin
= 0; pin
< nr_ioapic_registers
[apic
]; pin
++)
650 clear_IO_APIC_pin(apic
, pin
);
655 * support for broken MP BIOSs, enables hand-redirection of PIRQ0-7 to
656 * specific CPU-side IRQs.
660 static int pirq_entries
[MAX_PIRQS
] = {
661 [0 ... MAX_PIRQS
- 1] = -1
664 static int __init
ioapic_pirq_setup(char *str
)
667 int ints
[MAX_PIRQS
+1];
669 get_options(str
, ARRAY_SIZE(ints
), ints
);
671 apic_printk(APIC_VERBOSE
, KERN_INFO
672 "PIRQ redirection, working around broken MP-BIOS.\n");
674 if (ints
[0] < MAX_PIRQS
)
677 for (i
= 0; i
< max
; i
++) {
678 apic_printk(APIC_VERBOSE
, KERN_DEBUG
679 "... PIRQ%d -> IRQ %d\n", i
, ints
[i
+1]);
681 * PIRQs are mapped upside down, usually.
683 pirq_entries
[MAX_PIRQS
-i
-1] = ints
[i
+1];
688 __setup("pirq=", ioapic_pirq_setup
);
689 #endif /* CONFIG_X86_32 */
691 struct IO_APIC_route_entry
**alloc_ioapic_entries(void)
694 struct IO_APIC_route_entry
**ioapic_entries
;
696 ioapic_entries
= kzalloc(sizeof(*ioapic_entries
) * nr_ioapics
,
701 for (apic
= 0; apic
< nr_ioapics
; apic
++) {
702 ioapic_entries
[apic
] =
703 kzalloc(sizeof(struct IO_APIC_route_entry
) *
704 nr_ioapic_registers
[apic
], GFP_ATOMIC
);
705 if (!ioapic_entries
[apic
])
709 return ioapic_entries
;
713 kfree(ioapic_entries
[apic
]);
714 kfree(ioapic_entries
);
720 * Saves all the IO-APIC RTE's
722 int save_IO_APIC_setup(struct IO_APIC_route_entry
**ioapic_entries
)
729 for (apic
= 0; apic
< nr_ioapics
; apic
++) {
730 if (!ioapic_entries
[apic
])
733 for (pin
= 0; pin
< nr_ioapic_registers
[apic
]; pin
++)
734 ioapic_entries
[apic
][pin
] =
735 ioapic_read_entry(apic
, pin
);
742 * Mask all IO APIC entries.
744 void mask_IO_APIC_setup(struct IO_APIC_route_entry
**ioapic_entries
)
751 for (apic
= 0; apic
< nr_ioapics
; apic
++) {
752 if (!ioapic_entries
[apic
])
755 for (pin
= 0; pin
< nr_ioapic_registers
[apic
]; pin
++) {
756 struct IO_APIC_route_entry entry
;
758 entry
= ioapic_entries
[apic
][pin
];
761 ioapic_write_entry(apic
, pin
, entry
);
768 * Restore IO APIC entries which was saved in ioapic_entries.
770 int restore_IO_APIC_setup(struct IO_APIC_route_entry
**ioapic_entries
)
777 for (apic
= 0; apic
< nr_ioapics
; apic
++) {
778 if (!ioapic_entries
[apic
])
781 for (pin
= 0; pin
< nr_ioapic_registers
[apic
]; pin
++)
782 ioapic_write_entry(apic
, pin
,
783 ioapic_entries
[apic
][pin
]);
788 void free_ioapic_entries(struct IO_APIC_route_entry
**ioapic_entries
)
792 for (apic
= 0; apic
< nr_ioapics
; apic
++)
793 kfree(ioapic_entries
[apic
]);
795 kfree(ioapic_entries
);
799 * Find the IRQ entry number of a certain pin.
801 static int find_irq_entry(int apic
, int pin
, int type
)
805 for (i
= 0; i
< mp_irq_entries
; i
++)
806 if (mp_irqs
[i
].irqtype
== type
&&
807 (mp_irqs
[i
].dstapic
== mp_ioapics
[apic
].apicid
||
808 mp_irqs
[i
].dstapic
== MP_APIC_ALL
) &&
809 mp_irqs
[i
].dstirq
== pin
)
816 * Find the pin to which IRQ[irq] (ISA) is connected
818 static int __init
find_isa_irq_pin(int irq
, int type
)
822 for (i
= 0; i
< mp_irq_entries
; i
++) {
823 int lbus
= mp_irqs
[i
].srcbus
;
825 if (test_bit(lbus
, mp_bus_not_pci
) &&
826 (mp_irqs
[i
].irqtype
== type
) &&
827 (mp_irqs
[i
].srcbusirq
== irq
))
829 return mp_irqs
[i
].dstirq
;
834 static int __init
find_isa_irq_apic(int irq
, int type
)
838 for (i
= 0; i
< mp_irq_entries
; i
++) {
839 int lbus
= mp_irqs
[i
].srcbus
;
841 if (test_bit(lbus
, mp_bus_not_pci
) &&
842 (mp_irqs
[i
].irqtype
== type
) &&
843 (mp_irqs
[i
].srcbusirq
== irq
))
846 if (i
< mp_irq_entries
) {
848 for(apic
= 0; apic
< nr_ioapics
; apic
++) {
849 if (mp_ioapics
[apic
].apicid
== mp_irqs
[i
].dstapic
)
857 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
859 * EISA Edge/Level control register, ELCR
861 static int EISA_ELCR(unsigned int irq
)
863 if (irq
< legacy_pic
->nr_legacy_irqs
) {
864 unsigned int port
= 0x4d0 + (irq
>> 3);
865 return (inb(port
) >> (irq
& 7)) & 1;
867 apic_printk(APIC_VERBOSE
, KERN_INFO
868 "Broken MPtable reports ISA irq %d\n", irq
);
874 /* ISA interrupts are always polarity zero edge triggered,
875 * when listed as conforming in the MP table. */
877 #define default_ISA_trigger(idx) (0)
878 #define default_ISA_polarity(idx) (0)
880 /* EISA interrupts are always polarity zero and can be edge or level
881 * trigger depending on the ELCR value. If an interrupt is listed as
882 * EISA conforming in the MP table, that means its trigger type must
883 * be read in from the ELCR */
885 #define default_EISA_trigger(idx) (EISA_ELCR(mp_irqs[idx].srcbusirq))
886 #define default_EISA_polarity(idx) default_ISA_polarity(idx)
888 /* PCI interrupts are always polarity one level triggered,
889 * when listed as conforming in the MP table. */
891 #define default_PCI_trigger(idx) (1)
892 #define default_PCI_polarity(idx) (1)
894 /* MCA interrupts are always polarity zero level triggered,
895 * when listed as conforming in the MP table. */
897 #define default_MCA_trigger(idx) (1)
898 #define default_MCA_polarity(idx) default_ISA_polarity(idx)
900 static int MPBIOS_polarity(int idx
)
902 int bus
= mp_irqs
[idx
].srcbus
;
906 * Determine IRQ line polarity (high active or low active):
908 switch (mp_irqs
[idx
].irqflag
& 3)
910 case 0: /* conforms, ie. bus-type dependent polarity */
911 if (test_bit(bus
, mp_bus_not_pci
))
912 polarity
= default_ISA_polarity(idx
);
914 polarity
= default_PCI_polarity(idx
);
916 case 1: /* high active */
921 case 2: /* reserved */
923 printk(KERN_WARNING
"broken BIOS!!\n");
927 case 3: /* low active */
932 default: /* invalid */
934 printk(KERN_WARNING
"broken BIOS!!\n");
942 static int MPBIOS_trigger(int idx
)
944 int bus
= mp_irqs
[idx
].srcbus
;
948 * Determine IRQ trigger mode (edge or level sensitive):
950 switch ((mp_irqs
[idx
].irqflag
>>2) & 3)
952 case 0: /* conforms, ie. bus-type dependent */
953 if (test_bit(bus
, mp_bus_not_pci
))
954 trigger
= default_ISA_trigger(idx
);
956 trigger
= default_PCI_trigger(idx
);
957 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
958 switch (mp_bus_id_to_type
[bus
]) {
959 case MP_BUS_ISA
: /* ISA pin */
961 /* set before the switch */
964 case MP_BUS_EISA
: /* EISA pin */
966 trigger
= default_EISA_trigger(idx
);
969 case MP_BUS_PCI
: /* PCI pin */
971 /* set before the switch */
974 case MP_BUS_MCA
: /* MCA pin */
976 trigger
= default_MCA_trigger(idx
);
981 printk(KERN_WARNING
"broken BIOS!!\n");
993 case 2: /* reserved */
995 printk(KERN_WARNING
"broken BIOS!!\n");
1004 default: /* invalid */
1006 printk(KERN_WARNING
"broken BIOS!!\n");
1014 static inline int irq_polarity(int idx
)
1016 return MPBIOS_polarity(idx
);
1019 static inline int irq_trigger(int idx
)
1021 return MPBIOS_trigger(idx
);
1024 static int pin_2_irq(int idx
, int apic
, int pin
)
1027 int bus
= mp_irqs
[idx
].srcbus
;
1030 * Debugging check, we are in big trouble if this message pops up!
1032 if (mp_irqs
[idx
].dstirq
!= pin
)
1033 printk(KERN_ERR
"broken BIOS or MPTABLE parser, ayiee!!\n");
1035 if (test_bit(bus
, mp_bus_not_pci
)) {
1036 irq
= mp_irqs
[idx
].srcbusirq
;
1038 u32 gsi
= mp_gsi_routing
[apic
].gsi_base
+ pin
;
1040 if (gsi
>= NR_IRQS_LEGACY
)
1043 irq
= gsi_top
+ gsi
;
1046 #ifdef CONFIG_X86_32
1048 * PCI IRQ command line redirection. Yes, limits are hardcoded.
1050 if ((pin
>= 16) && (pin
<= 23)) {
1051 if (pirq_entries
[pin
-16] != -1) {
1052 if (!pirq_entries
[pin
-16]) {
1053 apic_printk(APIC_VERBOSE
, KERN_DEBUG
1054 "disabling PIRQ%d\n", pin
-16);
1056 irq
= pirq_entries
[pin
-16];
1057 apic_printk(APIC_VERBOSE
, KERN_DEBUG
1058 "using PIRQ%d -> IRQ %d\n",
1069 * Find a specific PCI IRQ entry.
1070 * Not an __init, possibly needed by modules
1072 int IO_APIC_get_PCI_irq_vector(int bus
, int slot
, int pin
,
1073 struct io_apic_irq_attr
*irq_attr
)
1075 int apic
, i
, best_guess
= -1;
1077 apic_printk(APIC_DEBUG
,
1078 "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
1080 if (test_bit(bus
, mp_bus_not_pci
)) {
1081 apic_printk(APIC_VERBOSE
,
1082 "PCI BIOS passed nonexistent PCI bus %d!\n", bus
);
1085 for (i
= 0; i
< mp_irq_entries
; i
++) {
1086 int lbus
= mp_irqs
[i
].srcbus
;
1088 for (apic
= 0; apic
< nr_ioapics
; apic
++)
1089 if (mp_ioapics
[apic
].apicid
== mp_irqs
[i
].dstapic
||
1090 mp_irqs
[i
].dstapic
== MP_APIC_ALL
)
1093 if (!test_bit(lbus
, mp_bus_not_pci
) &&
1094 !mp_irqs
[i
].irqtype
&&
1096 (slot
== ((mp_irqs
[i
].srcbusirq
>> 2) & 0x1f))) {
1097 int irq
= pin_2_irq(i
, apic
, mp_irqs
[i
].dstirq
);
1099 if (!(apic
|| IO_APIC_IRQ(irq
)))
1102 if (pin
== (mp_irqs
[i
].srcbusirq
& 3)) {
1103 set_io_apic_irq_attr(irq_attr
, apic
,
1110 * Use the first all-but-pin matching entry as a
1111 * best-guess fuzzy result for broken mptables.
1113 if (best_guess
< 0) {
1114 set_io_apic_irq_attr(irq_attr
, apic
,
1124 EXPORT_SYMBOL(IO_APIC_get_PCI_irq_vector
);
1126 void lock_vector_lock(void)
1128 /* Used to the online set of cpus does not change
1129 * during assign_irq_vector.
1131 raw_spin_lock(&vector_lock
);
1134 void unlock_vector_lock(void)
1136 raw_spin_unlock(&vector_lock
);
1140 __assign_irq_vector(int irq
, struct irq_cfg
*cfg
, const struct cpumask
*mask
)
1143 * NOTE! The local APIC isn't very good at handling
1144 * multiple interrupts at the same interrupt level.
1145 * As the interrupt level is determined by taking the
1146 * vector number and shifting that right by 4, we
1147 * want to spread these out a bit so that they don't
1148 * all fall in the same interrupt level.
1150 * Also, we've got to be careful not to trash gate
1151 * 0x80, because int 0x80 is hm, kind of importantish. ;)
1153 static int current_vector
= FIRST_EXTERNAL_VECTOR
+ VECTOR_OFFSET_START
;
1154 static int current_offset
= VECTOR_OFFSET_START
% 8;
1155 unsigned int old_vector
;
1157 cpumask_var_t tmp_mask
;
1159 if (cfg
->move_in_progress
)
1162 if (!alloc_cpumask_var(&tmp_mask
, GFP_ATOMIC
))
1165 old_vector
= cfg
->vector
;
1167 cpumask_and(tmp_mask
, mask
, cpu_online_mask
);
1168 cpumask_and(tmp_mask
, cfg
->domain
, tmp_mask
);
1169 if (!cpumask_empty(tmp_mask
)) {
1170 free_cpumask_var(tmp_mask
);
1175 /* Only try and allocate irqs on cpus that are present */
1177 for_each_cpu_and(cpu
, mask
, cpu_online_mask
) {
1181 apic
->vector_allocation_domain(cpu
, tmp_mask
);
1183 vector
= current_vector
;
1184 offset
= current_offset
;
1187 if (vector
>= first_system_vector
) {
1188 /* If out of vectors on large boxen, must share them. */
1189 offset
= (offset
+ 1) % 8;
1190 vector
= FIRST_EXTERNAL_VECTOR
+ offset
;
1192 if (unlikely(current_vector
== vector
))
1195 if (test_bit(vector
, used_vectors
))
1198 for_each_cpu_and(new_cpu
, tmp_mask
, cpu_online_mask
)
1199 if (per_cpu(vector_irq
, new_cpu
)[vector
] != -1)
1202 current_vector
= vector
;
1203 current_offset
= offset
;
1205 cfg
->move_in_progress
= 1;
1206 cpumask_copy(cfg
->old_domain
, cfg
->domain
);
1208 for_each_cpu_and(new_cpu
, tmp_mask
, cpu_online_mask
)
1209 per_cpu(vector_irq
, new_cpu
)[vector
] = irq
;
1210 cfg
->vector
= vector
;
1211 cpumask_copy(cfg
->domain
, tmp_mask
);
1215 free_cpumask_var(tmp_mask
);
1219 int assign_irq_vector(int irq
, struct irq_cfg
*cfg
, const struct cpumask
*mask
)
1222 unsigned long flags
;
1224 raw_spin_lock_irqsave(&vector_lock
, flags
);
1225 err
= __assign_irq_vector(irq
, cfg
, mask
);
1226 raw_spin_unlock_irqrestore(&vector_lock
, flags
);
1230 static void __clear_irq_vector(int irq
, struct irq_cfg
*cfg
)
1234 BUG_ON(!cfg
->vector
);
1236 vector
= cfg
->vector
;
1237 for_each_cpu_and(cpu
, cfg
->domain
, cpu_online_mask
)
1238 per_cpu(vector_irq
, cpu
)[vector
] = -1;
1241 cpumask_clear(cfg
->domain
);
1243 if (likely(!cfg
->move_in_progress
))
1245 for_each_cpu_and(cpu
, cfg
->old_domain
, cpu_online_mask
) {
1246 for (vector
= FIRST_EXTERNAL_VECTOR
; vector
< NR_VECTORS
;
1248 if (per_cpu(vector_irq
, cpu
)[vector
] != irq
)
1250 per_cpu(vector_irq
, cpu
)[vector
] = -1;
1254 cfg
->move_in_progress
= 0;
1257 void __setup_vector_irq(int cpu
)
1259 /* Initialize vector_irq on a new cpu */
1261 struct irq_cfg
*cfg
;
1262 struct irq_desc
*desc
;
1265 * vector_lock will make sure that we don't run into irq vector
1266 * assignments that might be happening on another cpu in parallel,
1267 * while we setup our initial vector to irq mappings.
1269 raw_spin_lock(&vector_lock
);
1270 /* Mark the inuse vectors */
1271 for_each_irq_desc(irq
, desc
) {
1272 cfg
= desc
->chip_data
;
1275 * If it is a legacy IRQ handled by the legacy PIC, this cpu
1276 * will be part of the irq_cfg's domain.
1278 if (irq
< legacy_pic
->nr_legacy_irqs
&& !IO_APIC_IRQ(irq
))
1279 cpumask_set_cpu(cpu
, cfg
->domain
);
1281 if (!cpumask_test_cpu(cpu
, cfg
->domain
))
1283 vector
= cfg
->vector
;
1284 per_cpu(vector_irq
, cpu
)[vector
] = irq
;
1286 /* Mark the free vectors */
1287 for (vector
= 0; vector
< NR_VECTORS
; ++vector
) {
1288 irq
= per_cpu(vector_irq
, cpu
)[vector
];
1293 if (!cpumask_test_cpu(cpu
, cfg
->domain
))
1294 per_cpu(vector_irq
, cpu
)[vector
] = -1;
1296 raw_spin_unlock(&vector_lock
);
1299 static struct irq_chip ioapic_chip
;
1300 static struct irq_chip ir_ioapic_chip
;
1302 #define IOAPIC_AUTO -1
1303 #define IOAPIC_EDGE 0
1304 #define IOAPIC_LEVEL 1
1306 #ifdef CONFIG_X86_32
1307 static inline int IO_APIC_irq_trigger(int irq
)
1311 for (apic
= 0; apic
< nr_ioapics
; apic
++) {
1312 for (pin
= 0; pin
< nr_ioapic_registers
[apic
]; pin
++) {
1313 idx
= find_irq_entry(apic
, pin
, mp_INT
);
1314 if ((idx
!= -1) && (irq
== pin_2_irq(idx
, apic
, pin
)))
1315 return irq_trigger(idx
);
1319 * nonexistent IRQs are edge default
1324 static inline int IO_APIC_irq_trigger(int irq
)
1330 static void ioapic_register_intr(int irq
, struct irq_desc
*desc
, unsigned long trigger
)
1333 if ((trigger
== IOAPIC_AUTO
&& IO_APIC_irq_trigger(irq
)) ||
1334 trigger
== IOAPIC_LEVEL
)
1335 desc
->status
|= IRQ_LEVEL
;
1337 desc
->status
&= ~IRQ_LEVEL
;
1339 if (irq_remapped(irq
)) {
1340 desc
->status
|= IRQ_MOVE_PCNTXT
;
1342 set_irq_chip_and_handler_name(irq
, &ir_ioapic_chip
,
1346 set_irq_chip_and_handler_name(irq
, &ir_ioapic_chip
,
1347 handle_edge_irq
, "edge");
1351 if ((trigger
== IOAPIC_AUTO
&& IO_APIC_irq_trigger(irq
)) ||
1352 trigger
== IOAPIC_LEVEL
)
1353 set_irq_chip_and_handler_name(irq
, &ioapic_chip
,
1357 set_irq_chip_and_handler_name(irq
, &ioapic_chip
,
1358 handle_edge_irq
, "edge");
1361 int setup_ioapic_entry(int apic_id
, int irq
,
1362 struct IO_APIC_route_entry
*entry
,
1363 unsigned int destination
, int trigger
,
1364 int polarity
, int vector
, int pin
)
1367 * add it to the IO-APIC irq-routing table:
1369 memset(entry
,0,sizeof(*entry
));
1371 if (intr_remapping_enabled
) {
1372 struct intel_iommu
*iommu
= map_ioapic_to_ir(apic_id
);
1374 struct IR_IO_APIC_route_entry
*ir_entry
=
1375 (struct IR_IO_APIC_route_entry
*) entry
;
1379 panic("No mapping iommu for ioapic %d\n", apic_id
);
1381 index
= alloc_irte(iommu
, irq
, 1);
1383 panic("Failed to allocate IRTE for ioapic %d\n", apic_id
);
1385 memset(&irte
, 0, sizeof(irte
));
1388 irte
.dst_mode
= apic
->irq_dest_mode
;
1390 * Trigger mode in the IRTE will always be edge, and the
1391 * actual level or edge trigger will be setup in the IO-APIC
1392 * RTE. This will help simplify level triggered irq migration.
1393 * For more details, see the comments above explainig IO-APIC
1394 * irq migration in the presence of interrupt-remapping.
1396 irte
.trigger_mode
= 0;
1397 irte
.dlvry_mode
= apic
->irq_delivery_mode
;
1398 irte
.vector
= vector
;
1399 irte
.dest_id
= IRTE_DEST(destination
);
1401 /* Set source-id of interrupt request */
1402 set_ioapic_sid(&irte
, apic_id
);
1404 modify_irte(irq
, &irte
);
1406 ir_entry
->index2
= (index
>> 15) & 0x1;
1408 ir_entry
->format
= 1;
1409 ir_entry
->index
= (index
& 0x7fff);
1411 * IO-APIC RTE will be configured with virtual vector.
1412 * irq handler will do the explicit EOI to the io-apic.
1414 ir_entry
->vector
= pin
;
1416 entry
->delivery_mode
= apic
->irq_delivery_mode
;
1417 entry
->dest_mode
= apic
->irq_dest_mode
;
1418 entry
->dest
= destination
;
1419 entry
->vector
= vector
;
1422 entry
->mask
= 0; /* enable IRQ */
1423 entry
->trigger
= trigger
;
1424 entry
->polarity
= polarity
;
1426 /* Mask level triggered irqs.
1427 * Use IRQ_DELAYED_DISABLE for edge triggered irqs.
1434 static void setup_IO_APIC_irq(int apic_id
, int pin
, unsigned int irq
, struct irq_desc
*desc
,
1435 int trigger
, int polarity
)
1437 struct irq_cfg
*cfg
;
1438 struct IO_APIC_route_entry entry
;
1441 if (!IO_APIC_IRQ(irq
))
1444 cfg
= desc
->chip_data
;
1447 * For legacy irqs, cfg->domain starts with cpu 0 for legacy
1448 * controllers like 8259. Now that IO-APIC can handle this irq, update
1451 if (irq
< legacy_pic
->nr_legacy_irqs
&& cpumask_test_cpu(0, cfg
->domain
))
1452 apic
->vector_allocation_domain(0, cfg
->domain
);
1454 if (assign_irq_vector(irq
, cfg
, apic
->target_cpus()))
1457 dest
= apic
->cpu_mask_to_apicid_and(cfg
->domain
, apic
->target_cpus());
1459 apic_printk(APIC_VERBOSE
,KERN_DEBUG
1460 "IOAPIC[%d]: Set routing entry (%d-%d -> 0x%x -> "
1461 "IRQ %d Mode:%i Active:%i)\n",
1462 apic_id
, mp_ioapics
[apic_id
].apicid
, pin
, cfg
->vector
,
1463 irq
, trigger
, polarity
);
1466 if (setup_ioapic_entry(mp_ioapics
[apic_id
].apicid
, irq
, &entry
,
1467 dest
, trigger
, polarity
, cfg
->vector
, pin
)) {
1468 printk("Failed to setup ioapic entry for ioapic %d, pin %d\n",
1469 mp_ioapics
[apic_id
].apicid
, pin
);
1470 __clear_irq_vector(irq
, cfg
);
1474 ioapic_register_intr(irq
, desc
, trigger
);
1475 if (irq
< legacy_pic
->nr_legacy_irqs
)
1476 legacy_pic
->chip
->mask(irq
);
1478 ioapic_write_entry(apic_id
, pin
, entry
);
1482 DECLARE_BITMAP(pin_programmed
, MP_MAX_IOAPIC_PIN
+ 1);
1483 } mp_ioapic_routing
[MAX_IO_APICS
];
1485 static void __init
setup_IO_APIC_irqs(void)
1487 int apic_id
, pin
, idx
, irq
;
1489 struct irq_desc
*desc
;
1490 struct irq_cfg
*cfg
;
1491 int node
= cpu_to_node(boot_cpu_id
);
1493 apic_printk(APIC_VERBOSE
, KERN_DEBUG
"init IO_APIC IRQs\n");
1495 for (apic_id
= 0; apic_id
< nr_ioapics
; apic_id
++)
1496 for (pin
= 0; pin
< nr_ioapic_registers
[apic_id
]; pin
++) {
1497 idx
= find_irq_entry(apic_id
, pin
, mp_INT
);
1501 apic_printk(APIC_VERBOSE
,
1502 KERN_DEBUG
" %d-%d",
1503 mp_ioapics
[apic_id
].apicid
, pin
);
1505 apic_printk(APIC_VERBOSE
, " %d-%d",
1506 mp_ioapics
[apic_id
].apicid
, pin
);
1510 apic_printk(APIC_VERBOSE
,
1511 " (apicid-pin) not connected\n");
1515 irq
= pin_2_irq(idx
, apic_id
, pin
);
1517 if ((apic_id
> 0) && (irq
> 16))
1521 * Skip the timer IRQ if there's a quirk handler
1522 * installed and if it returns 1:
1524 if (apic
->multi_timer_check
&&
1525 apic
->multi_timer_check(apic_id
, irq
))
1528 desc
= irq_to_desc_alloc_node(irq
, node
);
1530 printk(KERN_INFO
"can not get irq_desc for %d\n", irq
);
1533 cfg
= desc
->chip_data
;
1534 add_pin_to_irq_node(cfg
, node
, apic_id
, pin
);
1536 * don't mark it in pin_programmed, so later acpi could
1537 * set it correctly when irq < 16
1539 setup_IO_APIC_irq(apic_id
, pin
, irq
, desc
,
1540 irq_trigger(idx
), irq_polarity(idx
));
1544 apic_printk(APIC_VERBOSE
,
1545 " (apicid-pin) not connected\n");
1549 * for the gsit that is not in first ioapic
1550 * but could not use acpi_register_gsi()
1551 * like some special sci in IBM x3330
1553 void setup_IO_APIC_irq_extra(u32 gsi
)
1555 int apic_id
= 0, pin
, idx
, irq
;
1556 int node
= cpu_to_node(boot_cpu_id
);
1557 struct irq_desc
*desc
;
1558 struct irq_cfg
*cfg
;
1561 * Convert 'gsi' to 'ioapic.pin'.
1563 apic_id
= mp_find_ioapic(gsi
);
1567 pin
= mp_find_ioapic_pin(apic_id
, gsi
);
1568 idx
= find_irq_entry(apic_id
, pin
, mp_INT
);
1572 irq
= pin_2_irq(idx
, apic_id
, pin
);
1573 #ifdef CONFIG_SPARSE_IRQ
1574 desc
= irq_to_desc(irq
);
1578 desc
= irq_to_desc_alloc_node(irq
, node
);
1580 printk(KERN_INFO
"can not get irq_desc for %d\n", irq
);
1584 cfg
= desc
->chip_data
;
1585 add_pin_to_irq_node(cfg
, node
, apic_id
, pin
);
1587 if (test_bit(pin
, mp_ioapic_routing
[apic_id
].pin_programmed
)) {
1588 pr_debug("Pin %d-%d already programmed\n",
1589 mp_ioapics
[apic_id
].apicid
, pin
);
1592 set_bit(pin
, mp_ioapic_routing
[apic_id
].pin_programmed
);
1594 setup_IO_APIC_irq(apic_id
, pin
, irq
, desc
,
1595 irq_trigger(idx
), irq_polarity(idx
));
1599 * Set up the timer pin, possibly with the 8259A-master behind.
1601 static void __init
setup_timer_IRQ0_pin(unsigned int apic_id
, unsigned int pin
,
1604 struct IO_APIC_route_entry entry
;
1606 if (intr_remapping_enabled
)
1609 memset(&entry
, 0, sizeof(entry
));
1612 * We use logical delivery to get the timer IRQ
1615 entry
.dest_mode
= apic
->irq_dest_mode
;
1616 entry
.mask
= 0; /* don't mask IRQ for edge */
1617 entry
.dest
= apic
->cpu_mask_to_apicid(apic
->target_cpus());
1618 entry
.delivery_mode
= apic
->irq_delivery_mode
;
1621 entry
.vector
= vector
;
1624 * The timer IRQ doesn't have to know that behind the
1625 * scene we may have a 8259A-master in AEOI mode ...
1627 set_irq_chip_and_handler_name(0, &ioapic_chip
, handle_edge_irq
, "edge");
1630 * Add it to the IO-APIC irq-routing table:
1632 ioapic_write_entry(apic_id
, pin
, entry
);
1636 __apicdebuginit(void) print_IO_APIC(void)
1639 union IO_APIC_reg_00 reg_00
;
1640 union IO_APIC_reg_01 reg_01
;
1641 union IO_APIC_reg_02 reg_02
;
1642 union IO_APIC_reg_03 reg_03
;
1643 unsigned long flags
;
1644 struct irq_cfg
*cfg
;
1645 struct irq_desc
*desc
;
1648 printk(KERN_DEBUG
"number of MP IRQ sources: %d.\n", mp_irq_entries
);
1649 for (i
= 0; i
< nr_ioapics
; i
++)
1650 printk(KERN_DEBUG
"number of IO-APIC #%d registers: %d.\n",
1651 mp_ioapics
[i
].apicid
, nr_ioapic_registers
[i
]);
1654 * We are a bit conservative about what we expect. We have to
1655 * know about every hardware change ASAP.
1657 printk(KERN_INFO
"testing the IO APIC.......................\n");
1659 for (apic
= 0; apic
< nr_ioapics
; apic
++) {
1661 raw_spin_lock_irqsave(&ioapic_lock
, flags
);
1662 reg_00
.raw
= io_apic_read(apic
, 0);
1663 reg_01
.raw
= io_apic_read(apic
, 1);
1664 if (reg_01
.bits
.version
>= 0x10)
1665 reg_02
.raw
= io_apic_read(apic
, 2);
1666 if (reg_01
.bits
.version
>= 0x20)
1667 reg_03
.raw
= io_apic_read(apic
, 3);
1668 raw_spin_unlock_irqrestore(&ioapic_lock
, flags
);
1671 printk(KERN_DEBUG
"IO APIC #%d......\n", mp_ioapics
[apic
].apicid
);
1672 printk(KERN_DEBUG
".... register #00: %08X\n", reg_00
.raw
);
1673 printk(KERN_DEBUG
"....... : physical APIC id: %02X\n", reg_00
.bits
.ID
);
1674 printk(KERN_DEBUG
"....... : Delivery Type: %X\n", reg_00
.bits
.delivery_type
);
1675 printk(KERN_DEBUG
"....... : LTS : %X\n", reg_00
.bits
.LTS
);
1677 printk(KERN_DEBUG
".... register #01: %08X\n", *(int *)®_01
);
1678 printk(KERN_DEBUG
"....... : max redirection entries: %04X\n", reg_01
.bits
.entries
);
1680 printk(KERN_DEBUG
"....... : PRQ implemented: %X\n", reg_01
.bits
.PRQ
);
1681 printk(KERN_DEBUG
"....... : IO APIC version: %04X\n", reg_01
.bits
.version
);
1684 * Some Intel chipsets with IO APIC VERSION of 0x1? don't have reg_02,
1685 * but the value of reg_02 is read as the previous read register
1686 * value, so ignore it if reg_02 == reg_01.
1688 if (reg_01
.bits
.version
>= 0x10 && reg_02
.raw
!= reg_01
.raw
) {
1689 printk(KERN_DEBUG
".... register #02: %08X\n", reg_02
.raw
);
1690 printk(KERN_DEBUG
"....... : arbitration: %02X\n", reg_02
.bits
.arbitration
);
1694 * Some Intel chipsets with IO APIC VERSION of 0x2? don't have reg_02
1695 * or reg_03, but the value of reg_0[23] is read as the previous read
1696 * register value, so ignore it if reg_03 == reg_0[12].
1698 if (reg_01
.bits
.version
>= 0x20 && reg_03
.raw
!= reg_02
.raw
&&
1699 reg_03
.raw
!= reg_01
.raw
) {
1700 printk(KERN_DEBUG
".... register #03: %08X\n", reg_03
.raw
);
1701 printk(KERN_DEBUG
"....... : Boot DT : %X\n", reg_03
.bits
.boot_DT
);
1704 printk(KERN_DEBUG
".... IRQ redirection table:\n");
1706 printk(KERN_DEBUG
" NR Dst Mask Trig IRR Pol"
1707 " Stat Dmod Deli Vect:\n");
1709 for (i
= 0; i
<= reg_01
.bits
.entries
; i
++) {
1710 struct IO_APIC_route_entry entry
;
1712 entry
= ioapic_read_entry(apic
, i
);
1714 printk(KERN_DEBUG
" %02x %03X ",
1719 printk("%1d %1d %1d %1d %1d %1d %1d %02X\n",
1724 entry
.delivery_status
,
1726 entry
.delivery_mode
,
1731 printk(KERN_DEBUG
"IRQ to pin mappings:\n");
1732 for_each_irq_desc(irq
, desc
) {
1733 struct irq_pin_list
*entry
;
1735 cfg
= desc
->chip_data
;
1738 entry
= cfg
->irq_2_pin
;
1741 printk(KERN_DEBUG
"IRQ%d ", irq
);
1742 for_each_irq_pin(entry
, cfg
->irq_2_pin
)
1743 printk("-> %d:%d", entry
->apic
, entry
->pin
);
1747 printk(KERN_INFO
".................................... done.\n");
1752 __apicdebuginit(void) print_APIC_field(int base
)
1758 for (i
= 0; i
< 8; i
++)
1759 printk(KERN_CONT
"%08x", apic_read(base
+ i
*0x10));
1761 printk(KERN_CONT
"\n");
1764 __apicdebuginit(void) print_local_APIC(void *dummy
)
1766 unsigned int i
, v
, ver
, maxlvt
;
1769 printk(KERN_DEBUG
"printing local APIC contents on CPU#%d/%d:\n",
1770 smp_processor_id(), hard_smp_processor_id());
1771 v
= apic_read(APIC_ID
);
1772 printk(KERN_INFO
"... APIC ID: %08x (%01x)\n", v
, read_apic_id());
1773 v
= apic_read(APIC_LVR
);
1774 printk(KERN_INFO
"... APIC VERSION: %08x\n", v
);
1775 ver
= GET_APIC_VERSION(v
);
1776 maxlvt
= lapic_get_maxlvt();
1778 v
= apic_read(APIC_TASKPRI
);
1779 printk(KERN_DEBUG
"... APIC TASKPRI: %08x (%02x)\n", v
, v
& APIC_TPRI_MASK
);
1781 if (APIC_INTEGRATED(ver
)) { /* !82489DX */
1782 if (!APIC_XAPIC(ver
)) {
1783 v
= apic_read(APIC_ARBPRI
);
1784 printk(KERN_DEBUG
"... APIC ARBPRI: %08x (%02x)\n", v
,
1785 v
& APIC_ARBPRI_MASK
);
1787 v
= apic_read(APIC_PROCPRI
);
1788 printk(KERN_DEBUG
"... APIC PROCPRI: %08x\n", v
);
1792 * Remote read supported only in the 82489DX and local APIC for
1793 * Pentium processors.
1795 if (!APIC_INTEGRATED(ver
) || maxlvt
== 3) {
1796 v
= apic_read(APIC_RRR
);
1797 printk(KERN_DEBUG
"... APIC RRR: %08x\n", v
);
1800 v
= apic_read(APIC_LDR
);
1801 printk(KERN_DEBUG
"... APIC LDR: %08x\n", v
);
1802 if (!x2apic_enabled()) {
1803 v
= apic_read(APIC_DFR
);
1804 printk(KERN_DEBUG
"... APIC DFR: %08x\n", v
);
1806 v
= apic_read(APIC_SPIV
);
1807 printk(KERN_DEBUG
"... APIC SPIV: %08x\n", v
);
1809 printk(KERN_DEBUG
"... APIC ISR field:\n");
1810 print_APIC_field(APIC_ISR
);
1811 printk(KERN_DEBUG
"... APIC TMR field:\n");
1812 print_APIC_field(APIC_TMR
);
1813 printk(KERN_DEBUG
"... APIC IRR field:\n");
1814 print_APIC_field(APIC_IRR
);
1816 if (APIC_INTEGRATED(ver
)) { /* !82489DX */
1817 if (maxlvt
> 3) /* Due to the Pentium erratum 3AP. */
1818 apic_write(APIC_ESR
, 0);
1820 v
= apic_read(APIC_ESR
);
1821 printk(KERN_DEBUG
"... APIC ESR: %08x\n", v
);
1824 icr
= apic_icr_read();
1825 printk(KERN_DEBUG
"... APIC ICR: %08x\n", (u32
)icr
);
1826 printk(KERN_DEBUG
"... APIC ICR2: %08x\n", (u32
)(icr
>> 32));
1828 v
= apic_read(APIC_LVTT
);
1829 printk(KERN_DEBUG
"... APIC LVTT: %08x\n", v
);
1831 if (maxlvt
> 3) { /* PC is LVT#4. */
1832 v
= apic_read(APIC_LVTPC
);
1833 printk(KERN_DEBUG
"... APIC LVTPC: %08x\n", v
);
1835 v
= apic_read(APIC_LVT0
);
1836 printk(KERN_DEBUG
"... APIC LVT0: %08x\n", v
);
1837 v
= apic_read(APIC_LVT1
);
1838 printk(KERN_DEBUG
"... APIC LVT1: %08x\n", v
);
1840 if (maxlvt
> 2) { /* ERR is LVT#3. */
1841 v
= apic_read(APIC_LVTERR
);
1842 printk(KERN_DEBUG
"... APIC LVTERR: %08x\n", v
);
1845 v
= apic_read(APIC_TMICT
);
1846 printk(KERN_DEBUG
"... APIC TMICT: %08x\n", v
);
1847 v
= apic_read(APIC_TMCCT
);
1848 printk(KERN_DEBUG
"... APIC TMCCT: %08x\n", v
);
1849 v
= apic_read(APIC_TDCR
);
1850 printk(KERN_DEBUG
"... APIC TDCR: %08x\n", v
);
1852 if (boot_cpu_has(X86_FEATURE_EXTAPIC
)) {
1853 v
= apic_read(APIC_EFEAT
);
1854 maxlvt
= (v
>> 16) & 0xff;
1855 printk(KERN_DEBUG
"... APIC EFEAT: %08x\n", v
);
1856 v
= apic_read(APIC_ECTRL
);
1857 printk(KERN_DEBUG
"... APIC ECTRL: %08x\n", v
);
1858 for (i
= 0; i
< maxlvt
; i
++) {
1859 v
= apic_read(APIC_EILVTn(i
));
1860 printk(KERN_DEBUG
"... APIC EILVT%d: %08x\n", i
, v
);
1866 __apicdebuginit(void) print_local_APICs(int maxcpu
)
1874 for_each_online_cpu(cpu
) {
1877 smp_call_function_single(cpu
, print_local_APIC
, NULL
, 1);
1882 __apicdebuginit(void) print_PIC(void)
1885 unsigned long flags
;
1887 if (!legacy_pic
->nr_legacy_irqs
)
1890 printk(KERN_DEBUG
"\nprinting PIC contents\n");
1892 raw_spin_lock_irqsave(&i8259A_lock
, flags
);
1894 v
= inb(0xa1) << 8 | inb(0x21);
1895 printk(KERN_DEBUG
"... PIC IMR: %04x\n", v
);
1897 v
= inb(0xa0) << 8 | inb(0x20);
1898 printk(KERN_DEBUG
"... PIC IRR: %04x\n", v
);
1902 v
= inb(0xa0) << 8 | inb(0x20);
1906 raw_spin_unlock_irqrestore(&i8259A_lock
, flags
);
1908 printk(KERN_DEBUG
"... PIC ISR: %04x\n", v
);
1910 v
= inb(0x4d1) << 8 | inb(0x4d0);
1911 printk(KERN_DEBUG
"... PIC ELCR: %04x\n", v
);
1914 static int __initdata show_lapic
= 1;
1915 static __init
int setup_show_lapic(char *arg
)
1919 if (strcmp(arg
, "all") == 0) {
1920 show_lapic
= CONFIG_NR_CPUS
;
1922 get_option(&arg
, &num
);
1929 __setup("show_lapic=", setup_show_lapic
);
1931 __apicdebuginit(int) print_ICs(void)
1933 if (apic_verbosity
== APIC_QUIET
)
1938 /* don't print out if apic is not there */
1939 if (!cpu_has_apic
&& !apic_from_smp_config())
1942 print_local_APICs(show_lapic
);
1948 fs_initcall(print_ICs
);
1951 /* Where if anywhere is the i8259 connect in external int mode */
1952 static struct { int pin
, apic
; } ioapic_i8259
= { -1, -1 };
1954 void __init
enable_IO_APIC(void)
1956 int i8259_apic
, i8259_pin
;
1959 if (!legacy_pic
->nr_legacy_irqs
)
1962 for(apic
= 0; apic
< nr_ioapics
; apic
++) {
1964 /* See if any of the pins is in ExtINT mode */
1965 for (pin
= 0; pin
< nr_ioapic_registers
[apic
]; pin
++) {
1966 struct IO_APIC_route_entry entry
;
1967 entry
= ioapic_read_entry(apic
, pin
);
1969 /* If the interrupt line is enabled and in ExtInt mode
1970 * I have found the pin where the i8259 is connected.
1972 if ((entry
.mask
== 0) && (entry
.delivery_mode
== dest_ExtINT
)) {
1973 ioapic_i8259
.apic
= apic
;
1974 ioapic_i8259
.pin
= pin
;
1980 /* Look to see what if the MP table has reported the ExtINT */
1981 /* If we could not find the appropriate pin by looking at the ioapic
1982 * the i8259 probably is not connected the ioapic but give the
1983 * mptable a chance anyway.
1985 i8259_pin
= find_isa_irq_pin(0, mp_ExtINT
);
1986 i8259_apic
= find_isa_irq_apic(0, mp_ExtINT
);
1987 /* Trust the MP table if nothing is setup in the hardware */
1988 if ((ioapic_i8259
.pin
== -1) && (i8259_pin
>= 0)) {
1989 printk(KERN_WARNING
"ExtINT not setup in hardware but reported by MP table\n");
1990 ioapic_i8259
.pin
= i8259_pin
;
1991 ioapic_i8259
.apic
= i8259_apic
;
1993 /* Complain if the MP table and the hardware disagree */
1994 if (((ioapic_i8259
.apic
!= i8259_apic
) || (ioapic_i8259
.pin
!= i8259_pin
)) &&
1995 (i8259_pin
>= 0) && (ioapic_i8259
.pin
>= 0))
1997 printk(KERN_WARNING
"ExtINT in hardware and MP table differ\n");
2001 * Do not trust the IO-APIC being empty at bootup
2007 * Not an __init, needed by the reboot code
2009 void disable_IO_APIC(void)
2012 * Clear the IO-APIC before rebooting:
2016 if (!legacy_pic
->nr_legacy_irqs
)
2020 * If the i8259 is routed through an IOAPIC
2021 * Put that IOAPIC in virtual wire mode
2022 * so legacy interrupts can be delivered.
2024 * With interrupt-remapping, for now we will use virtual wire A mode,
2025 * as virtual wire B is little complex (need to configure both
2026 * IOAPIC RTE aswell as interrupt-remapping table entry).
2027 * As this gets called during crash dump, keep this simple for now.
2029 if (ioapic_i8259
.pin
!= -1 && !intr_remapping_enabled
) {
2030 struct IO_APIC_route_entry entry
;
2032 memset(&entry
, 0, sizeof(entry
));
2033 entry
.mask
= 0; /* Enabled */
2034 entry
.trigger
= 0; /* Edge */
2036 entry
.polarity
= 0; /* High */
2037 entry
.delivery_status
= 0;
2038 entry
.dest_mode
= 0; /* Physical */
2039 entry
.delivery_mode
= dest_ExtINT
; /* ExtInt */
2041 entry
.dest
= read_apic_id();
2044 * Add it to the IO-APIC irq-routing table:
2046 ioapic_write_entry(ioapic_i8259
.apic
, ioapic_i8259
.pin
, entry
);
2050 * Use virtual wire A mode when interrupt remapping is enabled.
2052 if (cpu_has_apic
|| apic_from_smp_config())
2053 disconnect_bsp_APIC(!intr_remapping_enabled
&&
2054 ioapic_i8259
.pin
!= -1);
2057 #ifdef CONFIG_X86_32
2059 * function to set the IO-APIC physical IDs based on the
2060 * values stored in the MPC table.
2062 * by Matt Domsch <Matt_Domsch@dell.com> Tue Dec 21 12:25:05 CST 1999
2065 void __init
setup_ioapic_ids_from_mpc(void)
2067 union IO_APIC_reg_00 reg_00
;
2068 physid_mask_t phys_id_present_map
;
2071 unsigned char old_id
;
2072 unsigned long flags
;
2077 * Don't check I/O APIC IDs for xAPIC systems. They have
2078 * no meaning without the serial APIC bus.
2080 if (!(boot_cpu_data
.x86_vendor
== X86_VENDOR_INTEL
)
2081 || APIC_XAPIC(apic_version
[boot_cpu_physical_apicid
]))
2084 * This is broken; anything with a real cpu count has to
2085 * circumvent this idiocy regardless.
2087 apic
->ioapic_phys_id_map(&phys_cpu_present_map
, &phys_id_present_map
);
2090 * Set the IOAPIC ID to the value stored in the MPC table.
2092 for (apic_id
= 0; apic_id
< nr_ioapics
; apic_id
++) {
2094 /* Read the register 0 value */
2095 raw_spin_lock_irqsave(&ioapic_lock
, flags
);
2096 reg_00
.raw
= io_apic_read(apic_id
, 0);
2097 raw_spin_unlock_irqrestore(&ioapic_lock
, flags
);
2099 old_id
= mp_ioapics
[apic_id
].apicid
;
2101 if (mp_ioapics
[apic_id
].apicid
>= get_physical_broadcast()) {
2102 printk(KERN_ERR
"BIOS bug, IO-APIC#%d ID is %d in the MPC table!...\n",
2103 apic_id
, mp_ioapics
[apic_id
].apicid
);
2104 printk(KERN_ERR
"... fixing up to %d. (tell your hw vendor)\n",
2106 mp_ioapics
[apic_id
].apicid
= reg_00
.bits
.ID
;
2110 * Sanity check, is the ID really free? Every APIC in a
2111 * system must have a unique ID or we get lots of nice
2112 * 'stuck on smp_invalidate_needed IPI wait' messages.
2114 if (apic
->check_apicid_used(&phys_id_present_map
,
2115 mp_ioapics
[apic_id
].apicid
)) {
2116 printk(KERN_ERR
"BIOS bug, IO-APIC#%d ID %d is already used!...\n",
2117 apic_id
, mp_ioapics
[apic_id
].apicid
);
2118 for (i
= 0; i
< get_physical_broadcast(); i
++)
2119 if (!physid_isset(i
, phys_id_present_map
))
2121 if (i
>= get_physical_broadcast())
2122 panic("Max APIC ID exceeded!\n");
2123 printk(KERN_ERR
"... fixing up to %d. (tell your hw vendor)\n",
2125 physid_set(i
, phys_id_present_map
);
2126 mp_ioapics
[apic_id
].apicid
= i
;
2129 apic
->apicid_to_cpu_present(mp_ioapics
[apic_id
].apicid
, &tmp
);
2130 apic_printk(APIC_VERBOSE
, "Setting %d in the "
2131 "phys_id_present_map\n",
2132 mp_ioapics
[apic_id
].apicid
);
2133 physids_or(phys_id_present_map
, phys_id_present_map
, tmp
);
2138 * We need to adjust the IRQ routing table
2139 * if the ID changed.
2141 if (old_id
!= mp_ioapics
[apic_id
].apicid
)
2142 for (i
= 0; i
< mp_irq_entries
; i
++)
2143 if (mp_irqs
[i
].dstapic
== old_id
)
2145 = mp_ioapics
[apic_id
].apicid
;
2148 * Read the right value from the MPC table and
2149 * write it into the ID register.
2151 apic_printk(APIC_VERBOSE
, KERN_INFO
2152 "...changing IO-APIC physical APIC ID to %d ...",
2153 mp_ioapics
[apic_id
].apicid
);
2155 reg_00
.bits
.ID
= mp_ioapics
[apic_id
].apicid
;
2156 raw_spin_lock_irqsave(&ioapic_lock
, flags
);
2157 io_apic_write(apic_id
, 0, reg_00
.raw
);
2158 raw_spin_unlock_irqrestore(&ioapic_lock
, flags
);
2163 raw_spin_lock_irqsave(&ioapic_lock
, flags
);
2164 reg_00
.raw
= io_apic_read(apic_id
, 0);
2165 raw_spin_unlock_irqrestore(&ioapic_lock
, flags
);
2166 if (reg_00
.bits
.ID
!= mp_ioapics
[apic_id
].apicid
)
2167 printk("could not set ID!\n");
2169 apic_printk(APIC_VERBOSE
, " ok.\n");
2174 int no_timer_check __initdata
;
2176 static int __init
notimercheck(char *s
)
2181 __setup("no_timer_check", notimercheck
);
2184 * There is a nasty bug in some older SMP boards, their mptable lies
2185 * about the timer IRQ. We do the following to work around the situation:
2187 * - timer IRQ defaults to IO-APIC IRQ
2188 * - if this function detects that timer IRQs are defunct, then we fall
2189 * back to ISA timer IRQs
2191 static int __init
timer_irq_works(void)
2193 unsigned long t1
= jiffies
;
2194 unsigned long flags
;
2199 local_save_flags(flags
);
2201 /* Let ten ticks pass... */
2202 mdelay((10 * 1000) / HZ
);
2203 local_irq_restore(flags
);
2206 * Expect a few ticks at least, to be sure some possible
2207 * glue logic does not lock up after one or two first
2208 * ticks in a non-ExtINT mode. Also the local APIC
2209 * might have cached one ExtINT interrupt. Finally, at
2210 * least one tick may be lost due to delays.
2214 if (time_after(jiffies
, t1
+ 4))
2220 * In the SMP+IOAPIC case it might happen that there are an unspecified
2221 * number of pending IRQ events unhandled. These cases are very rare,
2222 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
2223 * better to do it this way as thus we do not have to be aware of
2224 * 'pending' interrupts in the IRQ path, except at this point.
2227 * Edge triggered needs to resend any interrupt
2228 * that was delayed but this is now handled in the device
2233 * Starting up a edge-triggered IO-APIC interrupt is
2234 * nasty - we need to make sure that we get the edge.
2235 * If it is already asserted for some reason, we need
2236 * return 1 to indicate that is was pending.
2238 * This is not complete - we should be able to fake
2239 * an edge even if it isn't on the 8259A...
2242 static unsigned int startup_ioapic_irq(unsigned int irq
)
2244 int was_pending
= 0;
2245 unsigned long flags
;
2246 struct irq_cfg
*cfg
;
2248 raw_spin_lock_irqsave(&ioapic_lock
, flags
);
2249 if (irq
< legacy_pic
->nr_legacy_irqs
) {
2250 legacy_pic
->chip
->mask(irq
);
2251 if (legacy_pic
->irq_pending(irq
))
2255 __unmask_IO_APIC_irq(cfg
);
2256 raw_spin_unlock_irqrestore(&ioapic_lock
, flags
);
2261 static int ioapic_retrigger_irq(unsigned int irq
)
2264 struct irq_cfg
*cfg
= irq_cfg(irq
);
2265 unsigned long flags
;
2267 raw_spin_lock_irqsave(&vector_lock
, flags
);
2268 apic
->send_IPI_mask(cpumask_of(cpumask_first(cfg
->domain
)), cfg
->vector
);
2269 raw_spin_unlock_irqrestore(&vector_lock
, flags
);
2275 * Level and edge triggered IO-APIC interrupts need different handling,
2276 * so we use two separate IRQ descriptors. Edge triggered IRQs can be
2277 * handled with the level-triggered descriptor, but that one has slightly
2278 * more overhead. Level-triggered interrupts cannot be handled with the
2279 * edge-triggered handler, without risking IRQ storms and other ugly
2284 void send_cleanup_vector(struct irq_cfg
*cfg
)
2286 cpumask_var_t cleanup_mask
;
2288 if (unlikely(!alloc_cpumask_var(&cleanup_mask
, GFP_ATOMIC
))) {
2290 for_each_cpu_and(i
, cfg
->old_domain
, cpu_online_mask
)
2291 apic
->send_IPI_mask(cpumask_of(i
), IRQ_MOVE_CLEANUP_VECTOR
);
2293 cpumask_and(cleanup_mask
, cfg
->old_domain
, cpu_online_mask
);
2294 apic
->send_IPI_mask(cleanup_mask
, IRQ_MOVE_CLEANUP_VECTOR
);
2295 free_cpumask_var(cleanup_mask
);
2297 cfg
->move_in_progress
= 0;
2300 static void __target_IO_APIC_irq(unsigned int irq
, unsigned int dest
, struct irq_cfg
*cfg
)
2303 struct irq_pin_list
*entry
;
2304 u8 vector
= cfg
->vector
;
2306 for_each_irq_pin(entry
, cfg
->irq_2_pin
) {
2312 * With interrupt-remapping, destination information comes
2313 * from interrupt-remapping table entry.
2315 if (!irq_remapped(irq
))
2316 io_apic_write(apic
, 0x11 + pin
*2, dest
);
2317 reg
= io_apic_read(apic
, 0x10 + pin
*2);
2318 reg
&= ~IO_APIC_REDIR_VECTOR_MASK
;
2320 io_apic_modify(apic
, 0x10 + pin
*2, reg
);
2325 * Either sets desc->affinity to a valid value, and returns
2326 * ->cpu_mask_to_apicid of that in dest_id, or returns -1 and
2327 * leaves desc->affinity untouched.
2330 set_desc_affinity(struct irq_desc
*desc
, const struct cpumask
*mask
,
2331 unsigned int *dest_id
)
2333 struct irq_cfg
*cfg
;
2336 if (!cpumask_intersects(mask
, cpu_online_mask
))
2340 cfg
= desc
->chip_data
;
2341 if (assign_irq_vector(irq
, cfg
, mask
))
2344 cpumask_copy(desc
->affinity
, mask
);
2346 *dest_id
= apic
->cpu_mask_to_apicid_and(desc
->affinity
, cfg
->domain
);
2351 set_ioapic_affinity_irq_desc(struct irq_desc
*desc
, const struct cpumask
*mask
)
2353 struct irq_cfg
*cfg
;
2354 unsigned long flags
;
2360 cfg
= desc
->chip_data
;
2362 raw_spin_lock_irqsave(&ioapic_lock
, flags
);
2363 ret
= set_desc_affinity(desc
, mask
, &dest
);
2365 /* Only the high 8 bits are valid. */
2366 dest
= SET_APIC_LOGICAL_ID(dest
);
2367 __target_IO_APIC_irq(irq
, dest
, cfg
);
2369 raw_spin_unlock_irqrestore(&ioapic_lock
, flags
);
2375 set_ioapic_affinity_irq(unsigned int irq
, const struct cpumask
*mask
)
2377 struct irq_desc
*desc
;
2379 desc
= irq_to_desc(irq
);
2381 return set_ioapic_affinity_irq_desc(desc
, mask
);
2384 #ifdef CONFIG_INTR_REMAP
2387 * Migrate the IO-APIC irq in the presence of intr-remapping.
2389 * For both level and edge triggered, irq migration is a simple atomic
2390 * update(of vector and cpu destination) of IRTE and flush the hardware cache.
2392 * For level triggered, we eliminate the io-apic RTE modification (with the
2393 * updated vector information), by using a virtual vector (io-apic pin number).
2394 * Real vector that is used for interrupting cpu will be coming from
2395 * the interrupt-remapping table entry.
2398 migrate_ioapic_irq_desc(struct irq_desc
*desc
, const struct cpumask
*mask
)
2400 struct irq_cfg
*cfg
;
2406 if (!cpumask_intersects(mask
, cpu_online_mask
))
2410 if (get_irte(irq
, &irte
))
2413 cfg
= desc
->chip_data
;
2414 if (assign_irq_vector(irq
, cfg
, mask
))
2417 dest
= apic
->cpu_mask_to_apicid_and(cfg
->domain
, mask
);
2419 irte
.vector
= cfg
->vector
;
2420 irte
.dest_id
= IRTE_DEST(dest
);
2423 * Modified the IRTE and flushes the Interrupt entry cache.
2425 modify_irte(irq
, &irte
);
2427 if (cfg
->move_in_progress
)
2428 send_cleanup_vector(cfg
);
2430 cpumask_copy(desc
->affinity
, mask
);
2436 * Migrates the IRQ destination in the process context.
2438 static int set_ir_ioapic_affinity_irq_desc(struct irq_desc
*desc
,
2439 const struct cpumask
*mask
)
2441 return migrate_ioapic_irq_desc(desc
, mask
);
2443 static int set_ir_ioapic_affinity_irq(unsigned int irq
,
2444 const struct cpumask
*mask
)
2446 struct irq_desc
*desc
= irq_to_desc(irq
);
2448 return set_ir_ioapic_affinity_irq_desc(desc
, mask
);
2451 static inline int set_ir_ioapic_affinity_irq_desc(struct irq_desc
*desc
,
2452 const struct cpumask
*mask
)
2458 asmlinkage
void smp_irq_move_cleanup_interrupt(void)
2460 unsigned vector
, me
;
2466 me
= smp_processor_id();
2467 for (vector
= FIRST_EXTERNAL_VECTOR
; vector
< NR_VECTORS
; vector
++) {
2470 struct irq_desc
*desc
;
2471 struct irq_cfg
*cfg
;
2472 irq
= __get_cpu_var(vector_irq
)[vector
];
2477 desc
= irq_to_desc(irq
);
2482 raw_spin_lock(&desc
->lock
);
2485 * Check if the irq migration is in progress. If so, we
2486 * haven't received the cleanup request yet for this irq.
2488 if (cfg
->move_in_progress
)
2491 if (vector
== cfg
->vector
&& cpumask_test_cpu(me
, cfg
->domain
))
2494 irr
= apic_read(APIC_IRR
+ (vector
/ 32 * 0x10));
2496 * Check if the vector that needs to be cleanedup is
2497 * registered at the cpu's IRR. If so, then this is not
2498 * the best time to clean it up. Lets clean it up in the
2499 * next attempt by sending another IRQ_MOVE_CLEANUP_VECTOR
2502 if (irr
& (1 << (vector
% 32))) {
2503 apic
->send_IPI_self(IRQ_MOVE_CLEANUP_VECTOR
);
2506 __get_cpu_var(vector_irq
)[vector
] = -1;
2508 raw_spin_unlock(&desc
->lock
);
2514 static void __irq_complete_move(struct irq_desc
**descp
, unsigned vector
)
2516 struct irq_desc
*desc
= *descp
;
2517 struct irq_cfg
*cfg
= desc
->chip_data
;
2520 if (likely(!cfg
->move_in_progress
))
2523 me
= smp_processor_id();
2525 if (vector
== cfg
->vector
&& cpumask_test_cpu(me
, cfg
->domain
))
2526 send_cleanup_vector(cfg
);
2529 static void irq_complete_move(struct irq_desc
**descp
)
2531 __irq_complete_move(descp
, ~get_irq_regs()->orig_ax
);
2534 void irq_force_complete_move(int irq
)
2536 struct irq_desc
*desc
= irq_to_desc(irq
);
2537 struct irq_cfg
*cfg
= desc
->chip_data
;
2542 __irq_complete_move(&desc
, cfg
->vector
);
2545 static inline void irq_complete_move(struct irq_desc
**descp
) {}
2548 static void ack_apic_edge(unsigned int irq
)
2550 struct irq_desc
*desc
= irq_to_desc(irq
);
2552 irq_complete_move(&desc
);
2553 move_native_irq(irq
);
2557 atomic_t irq_mis_count
;
2560 * IO-APIC versions below 0x20 don't support EOI register.
2561 * For the record, here is the information about various versions:
2563 * 1Xh I/OAPIC or I/O(x)APIC which are not PCI 2.2 Compliant
2564 * 2Xh I/O(x)APIC which is PCI 2.2 Compliant
2567 * Some of the Intel ICH Specs (ICH2 to ICH5) documents the io-apic
2568 * version as 0x2. This is an error with documentation and these ICH chips
2569 * use io-apic's of version 0x20.
2571 * For IO-APIC's with EOI register, we use that to do an explicit EOI.
2572 * Otherwise, we simulate the EOI message manually by changing the trigger
2573 * mode to edge and then back to level, with RTE being masked during this.
2575 static void __eoi_ioapic_irq(unsigned int irq
, struct irq_cfg
*cfg
)
2577 struct irq_pin_list
*entry
;
2579 for_each_irq_pin(entry
, cfg
->irq_2_pin
) {
2580 if (mp_ioapics
[entry
->apic
].apicver
>= 0x20) {
2582 * Intr-remapping uses pin number as the virtual vector
2583 * in the RTE. Actual vector is programmed in
2584 * intr-remapping table entry. Hence for the io-apic
2585 * EOI we use the pin number.
2587 if (irq_remapped(irq
))
2588 io_apic_eoi(entry
->apic
, entry
->pin
);
2590 io_apic_eoi(entry
->apic
, cfg
->vector
);
2592 __mask_and_edge_IO_APIC_irq(entry
);
2593 __unmask_and_level_IO_APIC_irq(entry
);
2598 static void eoi_ioapic_irq(struct irq_desc
*desc
)
2600 struct irq_cfg
*cfg
;
2601 unsigned long flags
;
2605 cfg
= desc
->chip_data
;
2607 raw_spin_lock_irqsave(&ioapic_lock
, flags
);
2608 __eoi_ioapic_irq(irq
, cfg
);
2609 raw_spin_unlock_irqrestore(&ioapic_lock
, flags
);
2612 static void ack_apic_level(unsigned int irq
)
2614 struct irq_desc
*desc
= irq_to_desc(irq
);
2617 struct irq_cfg
*cfg
;
2618 int do_unmask_irq
= 0;
2620 irq_complete_move(&desc
);
2621 #ifdef CONFIG_GENERIC_PENDING_IRQ
2622 /* If we are moving the irq we need to mask it */
2623 if (unlikely(desc
->status
& IRQ_MOVE_PENDING
)) {
2625 mask_IO_APIC_irq_desc(desc
);
2630 * It appears there is an erratum which affects at least version 0x11
2631 * of I/O APIC (that's the 82093AA and cores integrated into various
2632 * chipsets). Under certain conditions a level-triggered interrupt is
2633 * erroneously delivered as edge-triggered one but the respective IRR
2634 * bit gets set nevertheless. As a result the I/O unit expects an EOI
2635 * message but it will never arrive and further interrupts are blocked
2636 * from the source. The exact reason is so far unknown, but the
2637 * phenomenon was observed when two consecutive interrupt requests
2638 * from a given source get delivered to the same CPU and the source is
2639 * temporarily disabled in between.
2641 * A workaround is to simulate an EOI message manually. We achieve it
2642 * by setting the trigger mode to edge and then to level when the edge
2643 * trigger mode gets detected in the TMR of a local APIC for a
2644 * level-triggered interrupt. We mask the source for the time of the
2645 * operation to prevent an edge-triggered interrupt escaping meanwhile.
2646 * The idea is from Manfred Spraul. --macro
2648 * Also in the case when cpu goes offline, fixup_irqs() will forward
2649 * any unhandled interrupt on the offlined cpu to the new cpu
2650 * destination that is handling the corresponding interrupt. This
2651 * interrupt forwarding is done via IPI's. Hence, in this case also
2652 * level-triggered io-apic interrupt will be seen as an edge
2653 * interrupt in the IRR. And we can't rely on the cpu's EOI
2654 * to be broadcasted to the IO-APIC's which will clear the remoteIRR
2655 * corresponding to the level-triggered interrupt. Hence on IO-APIC's
2656 * supporting EOI register, we do an explicit EOI to clear the
2657 * remote IRR and on IO-APIC's which don't have an EOI register,
2658 * we use the above logic (mask+edge followed by unmask+level) from
2659 * Manfred Spraul to clear the remote IRR.
2661 cfg
= desc
->chip_data
;
2663 v
= apic_read(APIC_TMR
+ ((i
& ~0x1f) >> 1));
2666 * We must acknowledge the irq before we move it or the acknowledge will
2667 * not propagate properly.
2672 * Tail end of clearing remote IRR bit (either by delivering the EOI
2673 * message via io-apic EOI register write or simulating it using
2674 * mask+edge followed by unnask+level logic) manually when the
2675 * level triggered interrupt is seen as the edge triggered interrupt
2678 if (!(v
& (1 << (i
& 0x1f)))) {
2679 atomic_inc(&irq_mis_count
);
2681 eoi_ioapic_irq(desc
);
2684 /* Now we can move and renable the irq */
2685 if (unlikely(do_unmask_irq
)) {
2686 /* Only migrate the irq if the ack has been received.
2688 * On rare occasions the broadcast level triggered ack gets
2689 * delayed going to ioapics, and if we reprogram the
2690 * vector while Remote IRR is still set the irq will never
2693 * To prevent this scenario we read the Remote IRR bit
2694 * of the ioapic. This has two effects.
2695 * - On any sane system the read of the ioapic will
2696 * flush writes (and acks) going to the ioapic from
2698 * - We get to see if the ACK has actually been delivered.
2700 * Based on failed experiments of reprogramming the
2701 * ioapic entry from outside of irq context starting
2702 * with masking the ioapic entry and then polling until
2703 * Remote IRR was clear before reprogramming the
2704 * ioapic I don't trust the Remote IRR bit to be
2705 * completey accurate.
2707 * However there appears to be no other way to plug
2708 * this race, so if the Remote IRR bit is not
2709 * accurate and is causing problems then it is a hardware bug
2710 * and you can go talk to the chipset vendor about it.
2712 cfg
= desc
->chip_data
;
2713 if (!io_apic_level_ack_pending(cfg
))
2714 move_masked_irq(irq
);
2715 unmask_IO_APIC_irq_desc(desc
);
2719 #ifdef CONFIG_INTR_REMAP
2720 static void ir_ack_apic_edge(unsigned int irq
)
2725 static void ir_ack_apic_level(unsigned int irq
)
2727 struct irq_desc
*desc
= irq_to_desc(irq
);
2730 eoi_ioapic_irq(desc
);
2732 #endif /* CONFIG_INTR_REMAP */
2734 static struct irq_chip ioapic_chip __read_mostly
= {
2736 .startup
= startup_ioapic_irq
,
2737 .mask
= mask_IO_APIC_irq
,
2738 .unmask
= unmask_IO_APIC_irq
,
2739 .ack
= ack_apic_edge
,
2740 .eoi
= ack_apic_level
,
2742 .set_affinity
= set_ioapic_affinity_irq
,
2744 .retrigger
= ioapic_retrigger_irq
,
2747 static struct irq_chip ir_ioapic_chip __read_mostly
= {
2748 .name
= "IR-IO-APIC",
2749 .startup
= startup_ioapic_irq
,
2750 .mask
= mask_IO_APIC_irq
,
2751 .unmask
= unmask_IO_APIC_irq
,
2752 #ifdef CONFIG_INTR_REMAP
2753 .ack
= ir_ack_apic_edge
,
2754 .eoi
= ir_ack_apic_level
,
2756 .set_affinity
= set_ir_ioapic_affinity_irq
,
2759 .retrigger
= ioapic_retrigger_irq
,
2762 static inline void init_IO_APIC_traps(void)
2765 struct irq_desc
*desc
;
2766 struct irq_cfg
*cfg
;
2769 * NOTE! The local APIC isn't very good at handling
2770 * multiple interrupts at the same interrupt level.
2771 * As the interrupt level is determined by taking the
2772 * vector number and shifting that right by 4, we
2773 * want to spread these out a bit so that they don't
2774 * all fall in the same interrupt level.
2776 * Also, we've got to be careful not to trash gate
2777 * 0x80, because int 0x80 is hm, kind of importantish. ;)
2779 for_each_irq_desc(irq
, desc
) {
2780 cfg
= desc
->chip_data
;
2781 if (IO_APIC_IRQ(irq
) && cfg
&& !cfg
->vector
) {
2783 * Hmm.. We don't have an entry for this,
2784 * so default to an old-fashioned 8259
2785 * interrupt if we can..
2787 if (irq
< legacy_pic
->nr_legacy_irqs
)
2788 legacy_pic
->make_irq(irq
);
2790 /* Strange. Oh, well.. */
2791 desc
->chip
= &no_irq_chip
;
2797 * The local APIC irq-chip implementation:
2800 static void mask_lapic_irq(unsigned int irq
)
2804 v
= apic_read(APIC_LVT0
);
2805 apic_write(APIC_LVT0
, v
| APIC_LVT_MASKED
);
2808 static void unmask_lapic_irq(unsigned int irq
)
2812 v
= apic_read(APIC_LVT0
);
2813 apic_write(APIC_LVT0
, v
& ~APIC_LVT_MASKED
);
2816 static void ack_lapic_irq(unsigned int irq
)
2821 static struct irq_chip lapic_chip __read_mostly
= {
2822 .name
= "local-APIC",
2823 .mask
= mask_lapic_irq
,
2824 .unmask
= unmask_lapic_irq
,
2825 .ack
= ack_lapic_irq
,
2828 static void lapic_register_intr(int irq
, struct irq_desc
*desc
)
2830 desc
->status
&= ~IRQ_LEVEL
;
2831 set_irq_chip_and_handler_name(irq
, &lapic_chip
, handle_edge_irq
,
2835 static void __init
setup_nmi(void)
2838 * Dirty trick to enable the NMI watchdog ...
2839 * We put the 8259A master into AEOI mode and
2840 * unmask on all local APICs LVT0 as NMI.
2842 * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
2843 * is from Maciej W. Rozycki - so we do not have to EOI from
2844 * the NMI handler or the timer interrupt.
2846 apic_printk(APIC_VERBOSE
, KERN_INFO
"activating NMI Watchdog ...");
2848 enable_NMI_through_LVT0();
2850 apic_printk(APIC_VERBOSE
, " done.\n");
2854 * This looks a bit hackish but it's about the only one way of sending
2855 * a few INTA cycles to 8259As and any associated glue logic. ICR does
2856 * not support the ExtINT mode, unfortunately. We need to send these
2857 * cycles as some i82489DX-based boards have glue logic that keeps the
2858 * 8259A interrupt line asserted until INTA. --macro
2860 static inline void __init
unlock_ExtINT_logic(void)
2863 struct IO_APIC_route_entry entry0
, entry1
;
2864 unsigned char save_control
, save_freq_select
;
2866 pin
= find_isa_irq_pin(8, mp_INT
);
2871 apic
= find_isa_irq_apic(8, mp_INT
);
2877 entry0
= ioapic_read_entry(apic
, pin
);
2878 clear_IO_APIC_pin(apic
, pin
);
2880 memset(&entry1
, 0, sizeof(entry1
));
2882 entry1
.dest_mode
= 0; /* physical delivery */
2883 entry1
.mask
= 0; /* unmask IRQ now */
2884 entry1
.dest
= hard_smp_processor_id();
2885 entry1
.delivery_mode
= dest_ExtINT
;
2886 entry1
.polarity
= entry0
.polarity
;
2890 ioapic_write_entry(apic
, pin
, entry1
);
2892 save_control
= CMOS_READ(RTC_CONTROL
);
2893 save_freq_select
= CMOS_READ(RTC_FREQ_SELECT
);
2894 CMOS_WRITE((save_freq_select
& ~RTC_RATE_SELECT
) | 0x6,
2896 CMOS_WRITE(save_control
| RTC_PIE
, RTC_CONTROL
);
2901 if ((CMOS_READ(RTC_INTR_FLAGS
) & RTC_PF
) == RTC_PF
)
2905 CMOS_WRITE(save_control
, RTC_CONTROL
);
2906 CMOS_WRITE(save_freq_select
, RTC_FREQ_SELECT
);
2907 clear_IO_APIC_pin(apic
, pin
);
2909 ioapic_write_entry(apic
, pin
, entry0
);
2912 static int disable_timer_pin_1 __initdata
;
2913 /* Actually the next is obsolete, but keep it for paranoid reasons -AK */
2914 static int __init
disable_timer_pin_setup(char *arg
)
2916 disable_timer_pin_1
= 1;
2919 early_param("disable_timer_pin_1", disable_timer_pin_setup
);
2921 int timer_through_8259 __initdata
;
2924 * This code may look a bit paranoid, but it's supposed to cooperate with
2925 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ
2926 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast
2927 * fanatically on his truly buggy board.
2929 * FIXME: really need to revamp this for all platforms.
2931 static inline void __init
check_timer(void)
2933 struct irq_desc
*desc
= irq_to_desc(0);
2934 struct irq_cfg
*cfg
= desc
->chip_data
;
2935 int node
= cpu_to_node(boot_cpu_id
);
2936 int apic1
, pin1
, apic2
, pin2
;
2937 unsigned long flags
;
2940 local_irq_save(flags
);
2943 * get/set the timer IRQ vector:
2945 legacy_pic
->chip
->mask(0);
2946 assign_irq_vector(0, cfg
, apic
->target_cpus());
2949 * As IRQ0 is to be enabled in the 8259A, the virtual
2950 * wire has to be disabled in the local APIC. Also
2951 * timer interrupts need to be acknowledged manually in
2952 * the 8259A for the i82489DX when using the NMI
2953 * watchdog as that APIC treats NMIs as level-triggered.
2954 * The AEOI mode will finish them in the 8259A
2957 apic_write(APIC_LVT0
, APIC_LVT_MASKED
| APIC_DM_EXTINT
);
2958 legacy_pic
->init(1);
2959 #ifdef CONFIG_X86_32
2963 ver
= apic_read(APIC_LVR
);
2964 ver
= GET_APIC_VERSION(ver
);
2965 timer_ack
= (nmi_watchdog
== NMI_IO_APIC
&& !APIC_INTEGRATED(ver
));
2969 pin1
= find_isa_irq_pin(0, mp_INT
);
2970 apic1
= find_isa_irq_apic(0, mp_INT
);
2971 pin2
= ioapic_i8259
.pin
;
2972 apic2
= ioapic_i8259
.apic
;
2974 apic_printk(APIC_QUIET
, KERN_INFO
"..TIMER: vector=0x%02X "
2975 "apic1=%d pin1=%d apic2=%d pin2=%d\n",
2976 cfg
->vector
, apic1
, pin1
, apic2
, pin2
);
2979 * Some BIOS writers are clueless and report the ExtINTA
2980 * I/O APIC input from the cascaded 8259A as the timer
2981 * interrupt input. So just in case, if only one pin
2982 * was found above, try it both directly and through the
2986 if (intr_remapping_enabled
)
2987 panic("BIOS bug: timer not connected to IO-APIC");
2991 } else if (pin2
== -1) {
2998 * Ok, does IRQ0 through the IOAPIC work?
3001 add_pin_to_irq_node(cfg
, node
, apic1
, pin1
);
3002 setup_timer_IRQ0_pin(apic1
, pin1
, cfg
->vector
);
3004 /* for edge trigger, setup_IO_APIC_irq already
3005 * leave it unmasked.
3006 * so only need to unmask if it is level-trigger
3007 * do we really have level trigger timer?
3010 idx
= find_irq_entry(apic1
, pin1
, mp_INT
);
3011 if (idx
!= -1 && irq_trigger(idx
))
3012 unmask_IO_APIC_irq_desc(desc
);
3014 if (timer_irq_works()) {
3015 if (nmi_watchdog
== NMI_IO_APIC
) {
3017 legacy_pic
->chip
->unmask(0);
3019 if (disable_timer_pin_1
> 0)
3020 clear_IO_APIC_pin(0, pin1
);
3023 if (intr_remapping_enabled
)
3024 panic("timer doesn't work through Interrupt-remapped IO-APIC");
3025 local_irq_disable();
3026 clear_IO_APIC_pin(apic1
, pin1
);
3028 apic_printk(APIC_QUIET
, KERN_ERR
"..MP-BIOS bug: "
3029 "8254 timer not connected to IO-APIC\n");
3031 apic_printk(APIC_QUIET
, KERN_INFO
"...trying to set up timer "
3032 "(IRQ0) through the 8259A ...\n");
3033 apic_printk(APIC_QUIET
, KERN_INFO
3034 "..... (found apic %d pin %d) ...\n", apic2
, pin2
);
3036 * legacy devices should be connected to IO APIC #0
3038 replace_pin_at_irq_node(cfg
, node
, apic1
, pin1
, apic2
, pin2
);
3039 setup_timer_IRQ0_pin(apic2
, pin2
, cfg
->vector
);
3040 legacy_pic
->chip
->unmask(0);
3041 if (timer_irq_works()) {
3042 apic_printk(APIC_QUIET
, KERN_INFO
"....... works.\n");
3043 timer_through_8259
= 1;
3044 if (nmi_watchdog
== NMI_IO_APIC
) {
3045 legacy_pic
->chip
->mask(0);
3047 legacy_pic
->chip
->unmask(0);
3052 * Cleanup, just in case ...
3054 local_irq_disable();
3055 legacy_pic
->chip
->mask(0);
3056 clear_IO_APIC_pin(apic2
, pin2
);
3057 apic_printk(APIC_QUIET
, KERN_INFO
"....... failed.\n");
3060 if (nmi_watchdog
== NMI_IO_APIC
) {
3061 apic_printk(APIC_QUIET
, KERN_WARNING
"timer doesn't work "
3062 "through the IO-APIC - disabling NMI Watchdog!\n");
3063 nmi_watchdog
= NMI_NONE
;
3065 #ifdef CONFIG_X86_32
3069 apic_printk(APIC_QUIET
, KERN_INFO
3070 "...trying to set up timer as Virtual Wire IRQ...\n");
3072 lapic_register_intr(0, desc
);
3073 apic_write(APIC_LVT0
, APIC_DM_FIXED
| cfg
->vector
); /* Fixed mode */
3074 legacy_pic
->chip
->unmask(0);
3076 if (timer_irq_works()) {
3077 apic_printk(APIC_QUIET
, KERN_INFO
"..... works.\n");
3080 local_irq_disable();
3081 legacy_pic
->chip
->mask(0);
3082 apic_write(APIC_LVT0
, APIC_LVT_MASKED
| APIC_DM_FIXED
| cfg
->vector
);
3083 apic_printk(APIC_QUIET
, KERN_INFO
"..... failed.\n");
3085 apic_printk(APIC_QUIET
, KERN_INFO
3086 "...trying to set up timer as ExtINT IRQ...\n");
3088 legacy_pic
->init(0);
3089 legacy_pic
->make_irq(0);
3090 apic_write(APIC_LVT0
, APIC_DM_EXTINT
);
3092 unlock_ExtINT_logic();
3094 if (timer_irq_works()) {
3095 apic_printk(APIC_QUIET
, KERN_INFO
"..... works.\n");
3098 local_irq_disable();
3099 apic_printk(APIC_QUIET
, KERN_INFO
"..... failed :(.\n");
3100 panic("IO-APIC + timer doesn't work! Boot with apic=debug and send a "
3101 "report. Then try booting with the 'noapic' option.\n");
3103 local_irq_restore(flags
);
3107 * Traditionally ISA IRQ2 is the cascade IRQ, and is not available
3108 * to devices. However there may be an I/O APIC pin available for
3109 * this interrupt regardless. The pin may be left unconnected, but
3110 * typically it will be reused as an ExtINT cascade interrupt for
3111 * the master 8259A. In the MPS case such a pin will normally be
3112 * reported as an ExtINT interrupt in the MP table. With ACPI
3113 * there is no provision for ExtINT interrupts, and in the absence
3114 * of an override it would be treated as an ordinary ISA I/O APIC
3115 * interrupt, that is edge-triggered and unmasked by default. We
3116 * used to do this, but it caused problems on some systems because
3117 * of the NMI watchdog and sometimes IRQ0 of the 8254 timer using
3118 * the same ExtINT cascade interrupt to drive the local APIC of the
3119 * bootstrap processor. Therefore we refrain from routing IRQ2 to
3120 * the I/O APIC in all cases now. No actual device should request
3121 * it anyway. --macro
3123 #define PIC_IRQS (1UL << PIC_CASCADE_IR)
3125 void __init
setup_IO_APIC(void)
3129 * calling enable_IO_APIC() is moved to setup_local_APIC for BP
3131 io_apic_irqs
= legacy_pic
->nr_legacy_irqs
? ~PIC_IRQS
: ~0UL;
3133 apic_printk(APIC_VERBOSE
, "ENABLING IO-APIC IRQs\n");
3135 * Set up IO-APIC IRQ routing.
3137 x86_init
.mpparse
.setup_ioapic_ids();
3140 setup_IO_APIC_irqs();
3141 init_IO_APIC_traps();
3142 if (legacy_pic
->nr_legacy_irqs
)
3147 * Called after all the initialization is done. If we didnt find any
3148 * APIC bugs then we can allow the modify fast path
3151 static int __init
io_apic_bug_finalize(void)
3153 if (sis_apic_bug
== -1)
3158 late_initcall(io_apic_bug_finalize
);
3160 struct sysfs_ioapic_data
{
3161 struct sys_device dev
;
3162 struct IO_APIC_route_entry entry
[0];
3164 static struct sysfs_ioapic_data
* mp_ioapic_data
[MAX_IO_APICS
];
3166 static int ioapic_suspend(struct sys_device
*dev
, pm_message_t state
)
3168 struct IO_APIC_route_entry
*entry
;
3169 struct sysfs_ioapic_data
*data
;
3172 data
= container_of(dev
, struct sysfs_ioapic_data
, dev
);
3173 entry
= data
->entry
;
3174 for (i
= 0; i
< nr_ioapic_registers
[dev
->id
]; i
++, entry
++ )
3175 *entry
= ioapic_read_entry(dev
->id
, i
);
3180 static int ioapic_resume(struct sys_device
*dev
)
3182 struct IO_APIC_route_entry
*entry
;
3183 struct sysfs_ioapic_data
*data
;
3184 unsigned long flags
;
3185 union IO_APIC_reg_00 reg_00
;
3188 data
= container_of(dev
, struct sysfs_ioapic_data
, dev
);
3189 entry
= data
->entry
;
3191 raw_spin_lock_irqsave(&ioapic_lock
, flags
);
3192 reg_00
.raw
= io_apic_read(dev
->id
, 0);
3193 if (reg_00
.bits
.ID
!= mp_ioapics
[dev
->id
].apicid
) {
3194 reg_00
.bits
.ID
= mp_ioapics
[dev
->id
].apicid
;
3195 io_apic_write(dev
->id
, 0, reg_00
.raw
);
3197 raw_spin_unlock_irqrestore(&ioapic_lock
, flags
);
3198 for (i
= 0; i
< nr_ioapic_registers
[dev
->id
]; i
++)
3199 ioapic_write_entry(dev
->id
, i
, entry
[i
]);
3204 static struct sysdev_class ioapic_sysdev_class
= {
3206 .suspend
= ioapic_suspend
,
3207 .resume
= ioapic_resume
,
3210 static int __init
ioapic_init_sysfs(void)
3212 struct sys_device
* dev
;
3215 error
= sysdev_class_register(&ioapic_sysdev_class
);
3219 for (i
= 0; i
< nr_ioapics
; i
++ ) {
3220 size
= sizeof(struct sys_device
) + nr_ioapic_registers
[i
]
3221 * sizeof(struct IO_APIC_route_entry
);
3222 mp_ioapic_data
[i
] = kzalloc(size
, GFP_KERNEL
);
3223 if (!mp_ioapic_data
[i
]) {
3224 printk(KERN_ERR
"Can't suspend/resume IOAPIC %d\n", i
);
3227 dev
= &mp_ioapic_data
[i
]->dev
;
3229 dev
->cls
= &ioapic_sysdev_class
;
3230 error
= sysdev_register(dev
);
3232 kfree(mp_ioapic_data
[i
]);
3233 mp_ioapic_data
[i
] = NULL
;
3234 printk(KERN_ERR
"Can't suspend/resume IOAPIC %d\n", i
);
3242 device_initcall(ioapic_init_sysfs
);
3245 * Dynamic irq allocate and deallocation
3247 unsigned int create_irq_nr(unsigned int irq_want
, int node
)
3249 /* Allocate an unused irq */
3252 unsigned long flags
;
3253 struct irq_cfg
*cfg_new
= NULL
;
3254 struct irq_desc
*desc_new
= NULL
;
3257 if (irq_want
< nr_irqs_gsi
)
3258 irq_want
= nr_irqs_gsi
;
3260 raw_spin_lock_irqsave(&vector_lock
, flags
);
3261 for (new = irq_want
; new < nr_irqs
; new++) {
3262 desc_new
= irq_to_desc_alloc_node(new, node
);
3264 printk(KERN_INFO
"can not get irq_desc for %d\n", new);
3267 cfg_new
= desc_new
->chip_data
;
3269 if (cfg_new
->vector
!= 0)
3272 desc_new
= move_irq_desc(desc_new
, node
);
3273 cfg_new
= desc_new
->chip_data
;
3275 if (__assign_irq_vector(new, cfg_new
, apic
->target_cpus()) == 0)
3279 raw_spin_unlock_irqrestore(&vector_lock
, flags
);
3282 dynamic_irq_init_keep_chip_data(irq
);
3287 int create_irq(void)
3289 int node
= cpu_to_node(boot_cpu_id
);
3290 unsigned int irq_want
;
3293 irq_want
= nr_irqs_gsi
;
3294 irq
= create_irq_nr(irq_want
, node
);
3302 void destroy_irq(unsigned int irq
)
3304 unsigned long flags
;
3306 dynamic_irq_cleanup_keep_chip_data(irq
);
3309 raw_spin_lock_irqsave(&vector_lock
, flags
);
3310 __clear_irq_vector(irq
, get_irq_chip_data(irq
));
3311 raw_spin_unlock_irqrestore(&vector_lock
, flags
);
3315 * MSI message composition
3317 #ifdef CONFIG_PCI_MSI
3318 static int msi_compose_msg(struct pci_dev
*pdev
, unsigned int irq
,
3319 struct msi_msg
*msg
, u8 hpet_id
)
3321 struct irq_cfg
*cfg
;
3329 err
= assign_irq_vector(irq
, cfg
, apic
->target_cpus());
3333 dest
= apic
->cpu_mask_to_apicid_and(cfg
->domain
, apic
->target_cpus());
3335 if (irq_remapped(irq
)) {
3340 ir_index
= map_irq_to_irte_handle(irq
, &sub_handle
);
3341 BUG_ON(ir_index
== -1);
3343 memset (&irte
, 0, sizeof(irte
));
3346 irte
.dst_mode
= apic
->irq_dest_mode
;
3347 irte
.trigger_mode
= 0; /* edge */
3348 irte
.dlvry_mode
= apic
->irq_delivery_mode
;
3349 irte
.vector
= cfg
->vector
;
3350 irte
.dest_id
= IRTE_DEST(dest
);
3352 /* Set source-id of interrupt request */
3354 set_msi_sid(&irte
, pdev
);
3356 set_hpet_sid(&irte
, hpet_id
);
3358 modify_irte(irq
, &irte
);
3360 msg
->address_hi
= MSI_ADDR_BASE_HI
;
3361 msg
->data
= sub_handle
;
3362 msg
->address_lo
= MSI_ADDR_BASE_LO
| MSI_ADDR_IR_EXT_INT
|
3364 MSI_ADDR_IR_INDEX1(ir_index
) |
3365 MSI_ADDR_IR_INDEX2(ir_index
);
3367 if (x2apic_enabled())
3368 msg
->address_hi
= MSI_ADDR_BASE_HI
|
3369 MSI_ADDR_EXT_DEST_ID(dest
);
3371 msg
->address_hi
= MSI_ADDR_BASE_HI
;
3375 ((apic
->irq_dest_mode
== 0) ?
3376 MSI_ADDR_DEST_MODE_PHYSICAL
:
3377 MSI_ADDR_DEST_MODE_LOGICAL
) |
3378 ((apic
->irq_delivery_mode
!= dest_LowestPrio
) ?
3379 MSI_ADDR_REDIRECTION_CPU
:
3380 MSI_ADDR_REDIRECTION_LOWPRI
) |
3381 MSI_ADDR_DEST_ID(dest
);
3384 MSI_DATA_TRIGGER_EDGE
|
3385 MSI_DATA_LEVEL_ASSERT
|
3386 ((apic
->irq_delivery_mode
!= dest_LowestPrio
) ?
3387 MSI_DATA_DELIVERY_FIXED
:
3388 MSI_DATA_DELIVERY_LOWPRI
) |
3389 MSI_DATA_VECTOR(cfg
->vector
);
3395 static int set_msi_irq_affinity(unsigned int irq
, const struct cpumask
*mask
)
3397 struct irq_desc
*desc
= irq_to_desc(irq
);
3398 struct irq_cfg
*cfg
;
3402 if (set_desc_affinity(desc
, mask
, &dest
))
3405 cfg
= desc
->chip_data
;
3407 get_cached_msi_msg_desc(desc
, &msg
);
3409 msg
.data
&= ~MSI_DATA_VECTOR_MASK
;
3410 msg
.data
|= MSI_DATA_VECTOR(cfg
->vector
);
3411 msg
.address_lo
&= ~MSI_ADDR_DEST_ID_MASK
;
3412 msg
.address_lo
|= MSI_ADDR_DEST_ID(dest
);
3414 write_msi_msg_desc(desc
, &msg
);
3418 #ifdef CONFIG_INTR_REMAP
3420 * Migrate the MSI irq to another cpumask. This migration is
3421 * done in the process context using interrupt-remapping hardware.
3424 ir_set_msi_irq_affinity(unsigned int irq
, const struct cpumask
*mask
)
3426 struct irq_desc
*desc
= irq_to_desc(irq
);
3427 struct irq_cfg
*cfg
= desc
->chip_data
;
3431 if (get_irte(irq
, &irte
))
3434 if (set_desc_affinity(desc
, mask
, &dest
))
3437 irte
.vector
= cfg
->vector
;
3438 irte
.dest_id
= IRTE_DEST(dest
);
3441 * atomically update the IRTE with the new destination and vector.
3443 modify_irte(irq
, &irte
);
3446 * After this point, all the interrupts will start arriving
3447 * at the new destination. So, time to cleanup the previous
3448 * vector allocation.
3450 if (cfg
->move_in_progress
)
3451 send_cleanup_vector(cfg
);
3457 #endif /* CONFIG_SMP */
3460 * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
3461 * which implement the MSI or MSI-X Capability Structure.
3463 static struct irq_chip msi_chip
= {
3465 .unmask
= unmask_msi_irq
,
3466 .mask
= mask_msi_irq
,
3467 .ack
= ack_apic_edge
,
3469 .set_affinity
= set_msi_irq_affinity
,
3471 .retrigger
= ioapic_retrigger_irq
,
3474 static struct irq_chip msi_ir_chip
= {
3475 .name
= "IR-PCI-MSI",
3476 .unmask
= unmask_msi_irq
,
3477 .mask
= mask_msi_irq
,
3478 #ifdef CONFIG_INTR_REMAP
3479 .ack
= ir_ack_apic_edge
,
3481 .set_affinity
= ir_set_msi_irq_affinity
,
3484 .retrigger
= ioapic_retrigger_irq
,
3488 * Map the PCI dev to the corresponding remapping hardware unit
3489 * and allocate 'nvec' consecutive interrupt-remapping table entries
3492 static int msi_alloc_irte(struct pci_dev
*dev
, int irq
, int nvec
)
3494 struct intel_iommu
*iommu
;
3497 iommu
= map_dev_to_ir(dev
);
3500 "Unable to map PCI %s to iommu\n", pci_name(dev
));
3504 index
= alloc_irte(iommu
, irq
, nvec
);
3507 "Unable to allocate %d IRTE for PCI %s\n", nvec
,
3514 static int setup_msi_irq(struct pci_dev
*dev
, struct msi_desc
*msidesc
, int irq
)
3519 ret
= msi_compose_msg(dev
, irq
, &msg
, -1);
3523 set_irq_msi(irq
, msidesc
);
3524 write_msi_msg(irq
, &msg
);
3526 if (irq_remapped(irq
)) {
3527 struct irq_desc
*desc
= irq_to_desc(irq
);
3529 * irq migration in process context
3531 desc
->status
|= IRQ_MOVE_PCNTXT
;
3532 set_irq_chip_and_handler_name(irq
, &msi_ir_chip
, handle_edge_irq
, "edge");
3534 set_irq_chip_and_handler_name(irq
, &msi_chip
, handle_edge_irq
, "edge");
3536 dev_printk(KERN_DEBUG
, &dev
->dev
, "irq %d for MSI/MSI-X\n", irq
);
3541 int arch_setup_msi_irqs(struct pci_dev
*dev
, int nvec
, int type
)
3544 int ret
, sub_handle
;
3545 struct msi_desc
*msidesc
;
3546 unsigned int irq_want
;
3547 struct intel_iommu
*iommu
= NULL
;
3551 /* x86 doesn't support multiple MSI yet */
3552 if (type
== PCI_CAP_ID_MSI
&& nvec
> 1)
3555 node
= dev_to_node(&dev
->dev
);
3556 irq_want
= nr_irqs_gsi
;
3558 list_for_each_entry(msidesc
, &dev
->msi_list
, list
) {
3559 irq
= create_irq_nr(irq_want
, node
);
3563 if (!intr_remapping_enabled
)
3568 * allocate the consecutive block of IRTE's
3571 index
= msi_alloc_irte(dev
, irq
, nvec
);
3577 iommu
= map_dev_to_ir(dev
);
3583 * setup the mapping between the irq and the IRTE
3584 * base index, the sub_handle pointing to the
3585 * appropriate interrupt remap table entry.
3587 set_irte_irq(irq
, iommu
, index
, sub_handle
);
3590 ret
= setup_msi_irq(dev
, msidesc
, irq
);
3602 void arch_teardown_msi_irq(unsigned int irq
)
3607 #if defined (CONFIG_DMAR) || defined (CONFIG_INTR_REMAP)
3609 static int dmar_msi_set_affinity(unsigned int irq
, const struct cpumask
*mask
)
3611 struct irq_desc
*desc
= irq_to_desc(irq
);
3612 struct irq_cfg
*cfg
;
3616 if (set_desc_affinity(desc
, mask
, &dest
))
3619 cfg
= desc
->chip_data
;
3621 dmar_msi_read(irq
, &msg
);
3623 msg
.data
&= ~MSI_DATA_VECTOR_MASK
;
3624 msg
.data
|= MSI_DATA_VECTOR(cfg
->vector
);
3625 msg
.address_lo
&= ~MSI_ADDR_DEST_ID_MASK
;
3626 msg
.address_lo
|= MSI_ADDR_DEST_ID(dest
);
3628 dmar_msi_write(irq
, &msg
);
3633 #endif /* CONFIG_SMP */
3635 static struct irq_chip dmar_msi_type
= {
3637 .unmask
= dmar_msi_unmask
,
3638 .mask
= dmar_msi_mask
,
3639 .ack
= ack_apic_edge
,
3641 .set_affinity
= dmar_msi_set_affinity
,
3643 .retrigger
= ioapic_retrigger_irq
,
3646 int arch_setup_dmar_msi(unsigned int irq
)
3651 ret
= msi_compose_msg(NULL
, irq
, &msg
, -1);
3654 dmar_msi_write(irq
, &msg
);
3655 set_irq_chip_and_handler_name(irq
, &dmar_msi_type
, handle_edge_irq
,
3661 #ifdef CONFIG_HPET_TIMER
3664 static int hpet_msi_set_affinity(unsigned int irq
, const struct cpumask
*mask
)
3666 struct irq_desc
*desc
= irq_to_desc(irq
);
3667 struct irq_cfg
*cfg
;
3671 if (set_desc_affinity(desc
, mask
, &dest
))
3674 cfg
= desc
->chip_data
;
3676 hpet_msi_read(irq
, &msg
);
3678 msg
.data
&= ~MSI_DATA_VECTOR_MASK
;
3679 msg
.data
|= MSI_DATA_VECTOR(cfg
->vector
);
3680 msg
.address_lo
&= ~MSI_ADDR_DEST_ID_MASK
;
3681 msg
.address_lo
|= MSI_ADDR_DEST_ID(dest
);
3683 hpet_msi_write(irq
, &msg
);
3688 #endif /* CONFIG_SMP */
3690 static struct irq_chip ir_hpet_msi_type
= {
3691 .name
= "IR-HPET_MSI",
3692 .unmask
= hpet_msi_unmask
,
3693 .mask
= hpet_msi_mask
,
3694 #ifdef CONFIG_INTR_REMAP
3695 .ack
= ir_ack_apic_edge
,
3697 .set_affinity
= ir_set_msi_irq_affinity
,
3700 .retrigger
= ioapic_retrigger_irq
,
3703 static struct irq_chip hpet_msi_type
= {
3705 .unmask
= hpet_msi_unmask
,
3706 .mask
= hpet_msi_mask
,
3707 .ack
= ack_apic_edge
,
3709 .set_affinity
= hpet_msi_set_affinity
,
3711 .retrigger
= ioapic_retrigger_irq
,
3714 int arch_setup_hpet_msi(unsigned int irq
, unsigned int id
)
3718 struct irq_desc
*desc
= irq_to_desc(irq
);
3720 if (intr_remapping_enabled
) {
3721 struct intel_iommu
*iommu
= map_hpet_to_ir(id
);
3727 index
= alloc_irte(iommu
, irq
, 1);
3732 ret
= msi_compose_msg(NULL
, irq
, &msg
, id
);
3736 hpet_msi_write(irq
, &msg
);
3737 desc
->status
|= IRQ_MOVE_PCNTXT
;
3738 if (irq_remapped(irq
))
3739 set_irq_chip_and_handler_name(irq
, &ir_hpet_msi_type
,
3740 handle_edge_irq
, "edge");
3742 set_irq_chip_and_handler_name(irq
, &hpet_msi_type
,
3743 handle_edge_irq
, "edge");
3749 #endif /* CONFIG_PCI_MSI */
3751 * Hypertransport interrupt support
3753 #ifdef CONFIG_HT_IRQ
3757 static void target_ht_irq(unsigned int irq
, unsigned int dest
, u8 vector
)
3759 struct ht_irq_msg msg
;
3760 fetch_ht_irq_msg(irq
, &msg
);
3762 msg
.address_lo
&= ~(HT_IRQ_LOW_VECTOR_MASK
| HT_IRQ_LOW_DEST_ID_MASK
);
3763 msg
.address_hi
&= ~(HT_IRQ_HIGH_DEST_ID_MASK
);
3765 msg
.address_lo
|= HT_IRQ_LOW_VECTOR(vector
) | HT_IRQ_LOW_DEST_ID(dest
);
3766 msg
.address_hi
|= HT_IRQ_HIGH_DEST_ID(dest
);
3768 write_ht_irq_msg(irq
, &msg
);
3771 static int set_ht_irq_affinity(unsigned int irq
, const struct cpumask
*mask
)
3773 struct irq_desc
*desc
= irq_to_desc(irq
);
3774 struct irq_cfg
*cfg
;
3777 if (set_desc_affinity(desc
, mask
, &dest
))
3780 cfg
= desc
->chip_data
;
3782 target_ht_irq(irq
, dest
, cfg
->vector
);
3789 static struct irq_chip ht_irq_chip
= {
3791 .mask
= mask_ht_irq
,
3792 .unmask
= unmask_ht_irq
,
3793 .ack
= ack_apic_edge
,
3795 .set_affinity
= set_ht_irq_affinity
,
3797 .retrigger
= ioapic_retrigger_irq
,
3800 int arch_setup_ht_irq(unsigned int irq
, struct pci_dev
*dev
)
3802 struct irq_cfg
*cfg
;
3809 err
= assign_irq_vector(irq
, cfg
, apic
->target_cpus());
3811 struct ht_irq_msg msg
;
3814 dest
= apic
->cpu_mask_to_apicid_and(cfg
->domain
,
3815 apic
->target_cpus());
3817 msg
.address_hi
= HT_IRQ_HIGH_DEST_ID(dest
);
3821 HT_IRQ_LOW_DEST_ID(dest
) |
3822 HT_IRQ_LOW_VECTOR(cfg
->vector
) |
3823 ((apic
->irq_dest_mode
== 0) ?
3824 HT_IRQ_LOW_DM_PHYSICAL
:
3825 HT_IRQ_LOW_DM_LOGICAL
) |
3826 HT_IRQ_LOW_RQEOI_EDGE
|
3827 ((apic
->irq_delivery_mode
!= dest_LowestPrio
) ?
3828 HT_IRQ_LOW_MT_FIXED
:
3829 HT_IRQ_LOW_MT_ARBITRATED
) |
3830 HT_IRQ_LOW_IRQ_MASKED
;
3832 write_ht_irq_msg(irq
, &msg
);
3834 set_irq_chip_and_handler_name(irq
, &ht_irq_chip
,
3835 handle_edge_irq
, "edge");
3837 dev_printk(KERN_DEBUG
, &dev
->dev
, "irq %d for HT\n", irq
);
3841 #endif /* CONFIG_HT_IRQ */
3843 int __init
io_apic_get_redir_entries (int ioapic
)
3845 union IO_APIC_reg_01 reg_01
;
3846 unsigned long flags
;
3848 raw_spin_lock_irqsave(&ioapic_lock
, flags
);
3849 reg_01
.raw
= io_apic_read(ioapic
, 1);
3850 raw_spin_unlock_irqrestore(&ioapic_lock
, flags
);
3852 /* The register returns the maximum index redir index
3853 * supported, which is one less than the total number of redir
3856 return reg_01
.bits
.entries
+ 1;
3859 void __init
probe_nr_irqs_gsi(void)
3863 nr
= gsi_top
+ NR_IRQS_LEGACY
;
3864 if (nr
> nr_irqs_gsi
)
3867 printk(KERN_DEBUG
"nr_irqs_gsi: %d\n", nr_irqs_gsi
);
3870 #ifdef CONFIG_SPARSE_IRQ
3871 int __init
arch_probe_nr_irqs(void)
3875 if (nr_irqs
> (NR_VECTORS
* nr_cpu_ids
))
3876 nr_irqs
= NR_VECTORS
* nr_cpu_ids
;
3878 nr
= nr_irqs_gsi
+ 8 * nr_cpu_ids
;
3879 #if defined(CONFIG_PCI_MSI) || defined(CONFIG_HT_IRQ)
3881 * for MSI and HT dyn irq
3883 nr
+= nr_irqs_gsi
* 16;
3892 static int __io_apic_set_pci_routing(struct device
*dev
, int irq
,
3893 struct io_apic_irq_attr
*irq_attr
)
3895 struct irq_desc
*desc
;
3896 struct irq_cfg
*cfg
;
3899 int trigger
, polarity
;
3901 ioapic
= irq_attr
->ioapic
;
3902 if (!IO_APIC_IRQ(irq
)) {
3903 apic_printk(APIC_QUIET
,KERN_ERR
"IOAPIC[%d]: Invalid reference to IRQ 0\n",
3909 node
= dev_to_node(dev
);
3911 node
= cpu_to_node(boot_cpu_id
);
3913 desc
= irq_to_desc_alloc_node(irq
, node
);
3915 printk(KERN_INFO
"can not get irq_desc %d\n", irq
);
3919 pin
= irq_attr
->ioapic_pin
;
3920 trigger
= irq_attr
->trigger
;
3921 polarity
= irq_attr
->polarity
;
3924 * IRQs < 16 are already in the irq_2_pin[] map
3926 if (irq
>= legacy_pic
->nr_legacy_irqs
) {
3927 cfg
= desc
->chip_data
;
3928 if (add_pin_to_irq_node_nopanic(cfg
, node
, ioapic
, pin
)) {
3929 printk(KERN_INFO
"can not add pin %d for irq %d\n",
3935 setup_IO_APIC_irq(ioapic
, pin
, irq
, desc
, trigger
, polarity
);
3940 int io_apic_set_pci_routing(struct device
*dev
, int irq
,
3941 struct io_apic_irq_attr
*irq_attr
)
3945 * Avoid pin reprogramming. PRTs typically include entries
3946 * with redundant pin->gsi mappings (but unique PCI devices);
3947 * we only program the IOAPIC on the first.
3949 ioapic
= irq_attr
->ioapic
;
3950 pin
= irq_attr
->ioapic_pin
;
3951 if (test_bit(pin
, mp_ioapic_routing
[ioapic
].pin_programmed
)) {
3952 pr_debug("Pin %d-%d already programmed\n",
3953 mp_ioapics
[ioapic
].apicid
, pin
);
3956 set_bit(pin
, mp_ioapic_routing
[ioapic
].pin_programmed
);
3958 return __io_apic_set_pci_routing(dev
, irq
, irq_attr
);
3961 u8 __init
io_apic_unique_id(u8 id
)
3963 #ifdef CONFIG_X86_32
3964 if ((boot_cpu_data
.x86_vendor
== X86_VENDOR_INTEL
) &&
3965 !APIC_XAPIC(apic_version
[boot_cpu_physical_apicid
]))
3966 return io_apic_get_unique_id(nr_ioapics
, id
);
3971 DECLARE_BITMAP(used
, 256);
3973 bitmap_zero(used
, 256);
3974 for (i
= 0; i
< nr_ioapics
; i
++) {
3975 struct mpc_ioapic
*ia
= &mp_ioapics
[i
];
3976 __set_bit(ia
->apicid
, used
);
3978 if (!test_bit(id
, used
))
3980 return find_first_zero_bit(used
, 256);
3984 #ifdef CONFIG_X86_32
3985 int __init
io_apic_get_unique_id(int ioapic
, int apic_id
)
3987 union IO_APIC_reg_00 reg_00
;
3988 static physid_mask_t apic_id_map
= PHYSID_MASK_NONE
;
3990 unsigned long flags
;
3994 * The P4 platform supports up to 256 APIC IDs on two separate APIC
3995 * buses (one for LAPICs, one for IOAPICs), where predecessors only
3996 * supports up to 16 on one shared APIC bus.
3998 * TBD: Expand LAPIC/IOAPIC support on P4-class systems to take full
3999 * advantage of new APIC bus architecture.
4002 if (physids_empty(apic_id_map
))
4003 apic
->ioapic_phys_id_map(&phys_cpu_present_map
, &apic_id_map
);
4005 raw_spin_lock_irqsave(&ioapic_lock
, flags
);
4006 reg_00
.raw
= io_apic_read(ioapic
, 0);
4007 raw_spin_unlock_irqrestore(&ioapic_lock
, flags
);
4009 if (apic_id
>= get_physical_broadcast()) {
4010 printk(KERN_WARNING
"IOAPIC[%d]: Invalid apic_id %d, trying "
4011 "%d\n", ioapic
, apic_id
, reg_00
.bits
.ID
);
4012 apic_id
= reg_00
.bits
.ID
;
4016 * Every APIC in a system must have a unique ID or we get lots of nice
4017 * 'stuck on smp_invalidate_needed IPI wait' messages.
4019 if (apic
->check_apicid_used(&apic_id_map
, apic_id
)) {
4021 for (i
= 0; i
< get_physical_broadcast(); i
++) {
4022 if (!apic
->check_apicid_used(&apic_id_map
, i
))
4026 if (i
== get_physical_broadcast())
4027 panic("Max apic_id exceeded!\n");
4029 printk(KERN_WARNING
"IOAPIC[%d]: apic_id %d already used, "
4030 "trying %d\n", ioapic
, apic_id
, i
);
4035 apic
->apicid_to_cpu_present(apic_id
, &tmp
);
4036 physids_or(apic_id_map
, apic_id_map
, tmp
);
4038 if (reg_00
.bits
.ID
!= apic_id
) {
4039 reg_00
.bits
.ID
= apic_id
;
4041 raw_spin_lock_irqsave(&ioapic_lock
, flags
);
4042 io_apic_write(ioapic
, 0, reg_00
.raw
);
4043 reg_00
.raw
= io_apic_read(ioapic
, 0);
4044 raw_spin_unlock_irqrestore(&ioapic_lock
, flags
);
4047 if (reg_00
.bits
.ID
!= apic_id
) {
4048 printk("IOAPIC[%d]: Unable to change apic_id!\n", ioapic
);
4053 apic_printk(APIC_VERBOSE
, KERN_INFO
4054 "IOAPIC[%d]: Assigned apic_id %d\n", ioapic
, apic_id
);
4060 int __init
io_apic_get_version(int ioapic
)
4062 union IO_APIC_reg_01 reg_01
;
4063 unsigned long flags
;
4065 raw_spin_lock_irqsave(&ioapic_lock
, flags
);
4066 reg_01
.raw
= io_apic_read(ioapic
, 1);
4067 raw_spin_unlock_irqrestore(&ioapic_lock
, flags
);
4069 return reg_01
.bits
.version
;
4072 int acpi_get_override_irq(u32 gsi
, int *trigger
, int *polarity
)
4074 int ioapic
, pin
, idx
;
4076 if (skip_ioapic_setup
)
4079 ioapic
= mp_find_ioapic(gsi
);
4083 pin
= mp_find_ioapic_pin(ioapic
, gsi
);
4087 idx
= find_irq_entry(ioapic
, pin
, mp_INT
);
4091 *trigger
= irq_trigger(idx
);
4092 *polarity
= irq_polarity(idx
);
4097 * This function currently is only a helper for the i386 smp boot process where
4098 * we need to reprogram the ioredtbls to cater for the cpus which have come online
4099 * so mask in all cases should simply be apic->target_cpus()
4102 void __init
setup_ioapic_dest(void)
4104 int pin
, ioapic
, irq
, irq_entry
;
4105 struct irq_desc
*desc
;
4106 const struct cpumask
*mask
;
4108 if (skip_ioapic_setup
== 1)
4111 for (ioapic
= 0; ioapic
< nr_ioapics
; ioapic
++)
4112 for (pin
= 0; pin
< nr_ioapic_registers
[ioapic
]; pin
++) {
4113 irq_entry
= find_irq_entry(ioapic
, pin
, mp_INT
);
4114 if (irq_entry
== -1)
4116 irq
= pin_2_irq(irq_entry
, ioapic
, pin
);
4118 if ((ioapic
> 0) && (irq
> 16))
4121 desc
= irq_to_desc(irq
);
4124 * Honour affinities which have been set in early boot
4127 (IRQ_NO_BALANCING
| IRQ_AFFINITY_SET
))
4128 mask
= desc
->affinity
;
4130 mask
= apic
->target_cpus();
4132 if (intr_remapping_enabled
)
4133 set_ir_ioapic_affinity_irq_desc(desc
, mask
);
4135 set_ioapic_affinity_irq_desc(desc
, mask
);
4141 #define IOAPIC_RESOURCE_NAME_SIZE 11
4143 static struct resource
*ioapic_resources
;
4145 static struct resource
* __init
ioapic_setup_resources(int nr_ioapics
)
4148 struct resource
*res
;
4152 if (nr_ioapics
<= 0)
4155 n
= IOAPIC_RESOURCE_NAME_SIZE
+ sizeof(struct resource
);
4158 mem
= alloc_bootmem(n
);
4161 mem
+= sizeof(struct resource
) * nr_ioapics
;
4163 for (i
= 0; i
< nr_ioapics
; i
++) {
4165 res
[i
].flags
= IORESOURCE_MEM
| IORESOURCE_BUSY
;
4166 snprintf(mem
, IOAPIC_RESOURCE_NAME_SIZE
, "IOAPIC %u", i
);
4167 mem
+= IOAPIC_RESOURCE_NAME_SIZE
;
4170 ioapic_resources
= res
;
4175 void __init
ioapic_init_mappings(void)
4177 unsigned long ioapic_phys
, idx
= FIX_IO_APIC_BASE_0
;
4178 struct resource
*ioapic_res
;
4181 ioapic_res
= ioapic_setup_resources(nr_ioapics
);
4182 for (i
= 0; i
< nr_ioapics
; i
++) {
4183 if (smp_found_config
) {
4184 ioapic_phys
= mp_ioapics
[i
].apicaddr
;
4185 #ifdef CONFIG_X86_32
4188 "WARNING: bogus zero IO-APIC "
4189 "address found in MPTABLE, "
4190 "disabling IO/APIC support!\n");
4191 smp_found_config
= 0;
4192 skip_ioapic_setup
= 1;
4193 goto fake_ioapic_page
;
4197 #ifdef CONFIG_X86_32
4200 ioapic_phys
= (unsigned long)alloc_bootmem_pages(PAGE_SIZE
);
4201 ioapic_phys
= __pa(ioapic_phys
);
4203 set_fixmap_nocache(idx
, ioapic_phys
);
4204 apic_printk(APIC_VERBOSE
, "mapped IOAPIC to %08lx (%08lx)\n",
4205 __fix_to_virt(idx
) + (ioapic_phys
& ~PAGE_MASK
),
4209 ioapic_res
->start
= ioapic_phys
;
4210 ioapic_res
->end
= ioapic_phys
+ IO_APIC_SLOT_SIZE
- 1;
4215 void __init
ioapic_insert_resources(void)
4218 struct resource
*r
= ioapic_resources
;
4223 "IO APIC resources couldn't be allocated.\n");
4227 for (i
= 0; i
< nr_ioapics
; i
++) {
4228 insert_resource(&iomem_resource
, r
);
4233 int mp_find_ioapic(u32 gsi
)
4237 /* Find the IOAPIC that manages this GSI. */
4238 for (i
= 0; i
< nr_ioapics
; i
++) {
4239 if ((gsi
>= mp_gsi_routing
[i
].gsi_base
)
4240 && (gsi
<= mp_gsi_routing
[i
].gsi_end
))
4244 printk(KERN_ERR
"ERROR: Unable to locate IOAPIC for GSI %d\n", gsi
);
4248 int mp_find_ioapic_pin(int ioapic
, u32 gsi
)
4250 if (WARN_ON(ioapic
== -1))
4252 if (WARN_ON(gsi
> mp_gsi_routing
[ioapic
].gsi_end
))
4255 return gsi
- mp_gsi_routing
[ioapic
].gsi_base
;
4258 static int bad_ioapic(unsigned long address
)
4260 if (nr_ioapics
>= MAX_IO_APICS
) {
4261 printk(KERN_WARNING
"WARING: Max # of I/O APICs (%d) exceeded "
4262 "(found %d), skipping\n", MAX_IO_APICS
, nr_ioapics
);
4266 printk(KERN_WARNING
"WARNING: Bogus (zero) I/O APIC address"
4267 " found in table, skipping!\n");
4273 void __init
mp_register_ioapic(int id
, u32 address
, u32 gsi_base
)
4278 if (bad_ioapic(address
))
4283 mp_ioapics
[idx
].type
= MP_IOAPIC
;
4284 mp_ioapics
[idx
].flags
= MPC_APIC_USABLE
;
4285 mp_ioapics
[idx
].apicaddr
= address
;
4287 set_fixmap_nocache(FIX_IO_APIC_BASE_0
+ idx
, address
);
4288 mp_ioapics
[idx
].apicid
= io_apic_unique_id(id
);
4289 mp_ioapics
[idx
].apicver
= io_apic_get_version(idx
);
4292 * Build basic GSI lookup table to facilitate gsi->io_apic lookups
4293 * and to prevent reprogramming of IOAPIC pins (PCI GSIs).
4295 entries
= io_apic_get_redir_entries(idx
);
4296 mp_gsi_routing
[idx
].gsi_base
= gsi_base
;
4297 mp_gsi_routing
[idx
].gsi_end
= gsi_base
+ entries
- 1;
4300 * The number of IO-APIC IRQ registers (== #pins):
4302 nr_ioapic_registers
[idx
] = entries
;
4304 if (mp_gsi_routing
[idx
].gsi_end
>= gsi_top
)
4305 gsi_top
= mp_gsi_routing
[idx
].gsi_end
+ 1;
4307 printk(KERN_INFO
"IOAPIC[%d]: apic_id %d, version %d, address 0x%x, "
4308 "GSI %d-%d\n", idx
, mp_ioapics
[idx
].apicid
,
4309 mp_ioapics
[idx
].apicver
, mp_ioapics
[idx
].apicaddr
,
4310 mp_gsi_routing
[idx
].gsi_base
, mp_gsi_routing
[idx
].gsi_end
);
4315 /* Enable IOAPIC early just for system timer */
4316 void __init
pre_init_apic_IRQ0(void)
4318 struct irq_cfg
*cfg
;
4319 struct irq_desc
*desc
;
4321 printk(KERN_INFO
"Early APIC setup for system timer0\n");
4323 phys_cpu_present_map
= physid_mask_of_physid(boot_cpu_physical_apicid
);
4325 desc
= irq_to_desc_alloc_node(0, 0);
4330 add_pin_to_irq_node(cfg
, 0, 0, 0);
4331 set_irq_chip_and_handler_name(0, &ioapic_chip
, handle_edge_irq
, "edge");
4333 setup_IO_APIC_irq(0, 0, 0, desc
, 0, 0);